{"id":200818,"date":"2026-07-09T03:56:23","date_gmt":"2026-07-09T08:56:23","guid":{"rendered":"https:\/\/ahrefs.com\/blog\/?p=200818"},"modified":"2026-07-09T05:12:51","modified_gmt":"2026-07-09T10:12:51","slug":"retrieval-augmented-generation","status":"publish","type":"post","link":"https:\/\/ahrefs.com\/blog\/retrieval-augmented-generation\/","title":{"rendered":"Retrieval Augmented Generation (RAG) Explained: How AI Decides Which Pages to Search &amp;&nbsp;Cite"},"content":{"rendered":"<div class=\"intro-txt\"> Retrieval augmented generation is a framework that determines which content AI tools retrieve and cite before generating an answer.<\/div>\n<p>You need to understand RAG because it\u2019s one of the ways ChatGPT, AI Mode and other AI search engines choose which pages get included in its answer.<\/p>\n<p>This guide explains how RAG works (in plain English), what makes content more likely to be retrieved, and how to measure your visibility in AI systems that use RAG with <a href=\"https:\/\/ahrefs.com\/brand-radar\">Ahrefs Brand Radar<\/a>.<\/p>\n<div class=\"intro-tok\" id=\"intro_tok\" style=\"display:none;\"><div class=\"intro-title\">Contents<\/div><a href=\"#\" class=\"expand-dots\"><span><\/span><span><\/span><span><\/span><\/a><\/div>\n<div class=\"post-nav-link clearfix\" id=\"section1\"><a class=\"subhead-anchor\" data-tip=\"tooltip__copielink\" rel=\"#section1\"><svg width=\"19\" height=\"19\" viewBox=\"0 0 14 14\" style><g fill=\"none\" fill-rule=\"evenodd\"><path d=\"M0 0h14v14H0z\" \/><path d=\"M7.45 9.887l-1.62 1.621c-.92.92-2.418.92-3.338 0a2.364 2.364 0 0 1 0-3.339l1.62-1.62-1.273-1.272-1.62 1.62a4.161 4.161 0 1 0 5.885 5.884l1.62-1.62L7.45 9.886zM5.527 5.135L7.17 3.492c.92-.92 2.418-.92 3.339 0 .92.92.92 2.418 0 3.339L8.866 8.473l1.272 1.273 1.644-1.643A4.161 4.161 0 1 0 5.897 2.22L4.254 3.863l1.272 1.272zm-.66 3.998a.749.749 0 0 1 0-1.06l2.208-2.206a.749.749 0 1 1 1.06 1.06L5.928 9.133a.75.75 0 0 1-1.061 0z\" style \/><\/g><\/svg><\/a><div class=\"link-text\" data-anchor=\"What is retrieval augmented generation (RAG)?\" data-section=\"what-is-rag\">\n<h2>What is retrieval augmented generation (RAG)?<\/h2>\n<\/div><\/div>\n<p>Retrieval augmented generation (RAG) is a technique where an LLM queries an index\u2014like a search engine, knowledge base, or vector database\u2014to find additional, contextually relevant information for its response\u2014rather than just defaulting to what it learned during training.<\/p>\n<p>Large language models are trained on huge datasets, but that training has a cutoff date.<\/p>\n<p>Ask an AI model what happened last week, or what\u2019s in your live production database, and you\u2019re asking it to work from memory with no reference material in front of&nbsp;it.<\/p>\n<p>When you query an AI model on information it doesn\u2019t yet have, that\u2019s when it\u2019s most likely to go rogue and start telling you that poison is good for&nbsp;you\u2026<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"700\" height=\"500\" class=\"wp-image-200824\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-1-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-1-1.png 700w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-1-1-595x425.png 595w\" sizes=\"auto, (max-width: 700px) 100vw, 700px\"><\/p>\n<p>This is otherwise known as an AI \u201challucination\u201d.<\/p>\n<p>RAG gives AI models access to the correct, up-to-date material to avoid this&nbsp;fate.<\/p>\n<p>With RAG, LLMs are either supplementing or overriding their internal knowledge\u2014known as their \u201cparametric memory\u201d\u2014in an attempt to give a more reliable answer.<\/p>\n<p>This process is also sometimes known as \u201cgrounding\u201d; anchoring the response to specific sources so the model isn\u2019t just freestyling from its training data.<\/p>\n<p>The three words map to the three stages of the process:<\/p>\n<ul>\n<li><strong>Retrieval<\/strong>: The AI model runs a search query to find (or <em>retrieve<\/em>) relevant content<\/li>\n<li><strong>Augmented<\/strong>: It adds that retrieved content to its input (<em>augmenting <\/em>its knowledge)<\/li>\n<li><strong>Generation<\/strong>: It uses the query and the retrieved content to write (or <em>generate<\/em>) a response<\/li>\n<\/ul>\n<p>Most AI tools use both RAG and trained knowledge in tandem.<\/p>\n<blockquote class=\"small\"><div class=\"quote-content\">Most AI tools have at least two things operating under the hood: the base model generates language from patterns learned during training. The retrieval layer goes looking for sources to attach.<br>\n<\/div><div class=\"quote-info clearfix\"><div class=\"quote-photo\"><img decoding=\"async\" alt=\"Dorron Shapow\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/Dorron-Shapow.jpeg\"><\/div><div class=\"extra-box\"><span class=\"quote-author\">Dorron Shapow,<\/span> <span class=\"quote-author-job\">Search User Optimization Expert<\/span><\/div><\/div><\/blockquote>\n<p>Getting into the base model\u2019s knowledge means being part of its training data, and that isn\u2019t something you can easily control.<\/p>\n<p>But getting into the retrieval results is, in many ways, an extension of&nbsp;SEO.<\/p>\n<div class=\"post-nav-link clearfix\" id=\"section1\"><a class=\"subhead-anchor\" data-tip=\"tooltip__copielink\" rel=\"#section1\"><svg width=\"19\" height=\"19\" viewBox=\"0 0 14 14\" style><g fill=\"none\" fill-rule=\"evenodd\"><path d=\"M0 0h14v14H0z\" \/><path d=\"M7.45 9.887l-1.62 1.621c-.92.92-2.418.92-3.338 0a2.364 2.364 0 0 1 0-3.339l1.62-1.62-1.273-1.272-1.62 1.62a4.161 4.161 0 1 0 5.885 5.884l1.62-1.62L7.45 9.886zM5.527 5.135L7.17 3.492c.92-.92 2.418-.92 3.339 0 .92.92.92 2.418 0 3.339L8.866 8.473l1.272 1.273 1.644-1.643A4.161 4.161 0 1 0 5.897 2.22L4.254 3.863l1.272 1.272zm-.66 3.998a.749.749 0 0 1 0-1.06l2.208-2.206a.749.749 0 1 1 1.06 1.06L5.928 9.133a.75.75 0 0 1-1.061 0z\" style \/><\/g><\/svg><\/a><div class=\"link-text\" data-anchor=\"How does RAG work?\" data-section=\"how-rag-works\">\n<h2>How does RAG&nbsp;work?<\/h2>\n<\/div><\/div>\n<p>Every RAG-powered AI answer follows three steps: search, retrieve, generate.<\/p>\n<p>To understand the details of what likely occurs at each stage, here\u2019s what we know about how ChatGPT retrieves its sources.<\/p>\n<h3>Step 1: The AI decides whether or not to run a search<\/h3>\n<p>Before anything gets retrieved, the AI will decide whether it even <em>needs<\/em> to enrich its knowledge with outside data.<\/p>\n<p>Simple fact-finding queries like \u201cWhat is a VPN?\u201d can usually be fielded by the core model based on its existing training knowledge. No retrieval needed.<\/p>\n<p>In ChatGPT\u2019s case, a smaller classifier model (part of the \u201csonicberry\u201d system according to <a href=\"https:\/\/uk.linkedin.com\/in\/david-mcsweeney-79840154\">David McSweeney<\/a>, who put in the work to find out just <a href=\"https:\/\/queryburst.com\/blog\/how-chatgpt-works\/\">how ChatGPT retrieves sources<\/a>) will run first, assigning probability scores to determine whether a query needs: <strong>no search<\/strong>, <strong>a simple search<\/strong>, or a <strong>complex multi-step search<\/strong>.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"927\" height=\"448\" class=\"wp-image-200826\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-2-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-2-1.png 927w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-2-1-680x329.png 680w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-2-1-768x371.png 768w\" sizes=\"auto, (max-width: 927px) 100vw, 927px\"><\/p>\n<p>Other AI tools will handle this step differently, but they all share the same underlying logic: not every query triggers a search.<\/p>\n<h3>Step 2: The AI runs a search<\/h3>\n<p>Whenever someone asks ChatGPT a question that requires more context, it expands that query into multiple related queries, then sends them to an external search index like Bing or Google to collect results.<\/p>\n<p>That expansion process is known as <a href=\"https:\/\/ahrefs.com\/blog\/query-fan-out\/\">query fan-out<\/a> (remember that for&nbsp;later).<\/p>\n<div class=\"further-reading\"><div class=\"reading-title\">Further reading<\/div><div class=\"reading-content\">\n<ul>\n<li><a href=\"https:\/\/ahrefs.com\/blog\/query-fan-out\/\">What is Query Fan-Out? Understanding the Hidden Queries Driving AI Search<\/a><\/li>\n<\/ul>\n<\/div><\/div>\n<p>Once a selection of pages are collected, on-page SEO factors like the title, meta description\/summary, and URL determine which page gets read in full, according to <a href=\"https:\/\/dejanmarketing.com\/gpt-search\/\">research<\/a> by AI Expert <a href=\"https:\/\/au.linkedin.com\/in\/seoguy\">Dan Petrovic<\/a>.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"1698\" height=\"976\" class=\"wp-image-200827\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-3-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-3-1.png 1698w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-3-1-680x391.png 680w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-3-1-768x441.png 768w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-3-1-1536x883.png 1536w\" sizes=\"auto, (max-width: 1698px) 100vw, 1698px\"><\/p>\n<p>From there, <a href=\"https:\/\/dejanmarketing.com\/gpt-search\/\">he discovered<\/a> that sources are shortlisted for scraping based on \u201crelevance, authority, recency, and diversity of perspective\u201d.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"1884\" height=\"1494\" class=\"wp-image-200829\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-4-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-4-1.png 1884w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-4-1-536x425.png 536w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-4-1-768x609.png 768w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-4-1-1536x1218.png 1536w\" sizes=\"auto, (max-width: 1884px) 100vw, 1884px\"><\/p>\n<div class=\"recommendation\"><div class=\"recommendation-title\">Some AI assistants have a \u201cVIP lane\u201d for certain domains<\/div><div class=\"recommendation-content\">\n<p>AI expert <a href=\"https:\/\/www.linkedin.com\/feed\/update\/urn:li:activity:7398302645997752320\/\">J\u00e9r\u00f4me Salomon<\/a>&nbsp;has found evidence that ChatGPT is building its own search index of cached content.<\/p>\n<p>In other words, it doesn\u2019t <em>always<\/em> retrieve from live search engine results pages.<\/p>\n<p>In addition to this, according to separate research carried out by <a href=\"https:\/\/www.linkedin.com\/feed\/update\/urn:li:activity:7475163074098987008?updateEntityUrn=urn%3Ali%3Afs_updateV2%3A%28urn%3Ali%3Aactivity%3A7475163074098987008%2CFEED_DETAIL%2CEMPTY%2CDEFAULT%2Cfalse%29\">Mark Williams-Cook<\/a>, <a href=\"https:\/\/queryburst.com\/blog\/how-chatgpt-works\/\">David McSweeney<\/a>, and <a href=\"https:\/\/suganthan.com\/blog\/how-chatgpt-picks-sources\/\">Suganthan Mohanadasan<\/a>, ChatGPT reportedly feeds in content from a separate, licensed \u201cVIP\u201d tier of authoritative sites and publishers\u2014many with existing content deals (e.g. Reuters, the WSJ, Wikipedia).<\/p>\n<p>These sites are tagged with the name <code>labrador<\/code> in ChatGPT\u2019s network traffic files, and are retrieved with pre-summarized, <span style=\"font-weight: 400;\">near-full-article extracts<\/span> rather than scraped and chunked like all other results.<\/p>\n<\/div><\/div>\n<h3>Step 2: Content gets broken into chunks\u2014and the closest match&nbsp;wins<\/h3>\n<p>Before it can be fully retrieved and served in the response, the scraped web content gets broken into smaller pieces called <strong>chunks<\/strong>.<\/p>\n<p>Think of <a href=\"https:\/\/ahrefs.com\/blog\/seo-chunk-optimization\/\">chunking<\/a> like tearing a book into individual chapters. The system breaks the page into pieces, then asks which piece best answers the question.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"alignnone wp-image-201010 size-full\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/unnamed.png\" alt=\"how llm guided chunking works. Document > llm extracts thought units > chunks\" width=\"1600\" height=\"800\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/unnamed.png 1600w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/unnamed-680x340.png 680w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/unnamed-768x384.png 768w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/unnamed-1536x768.png 1536w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/unnamed-400x200.png 400w\" sizes=\"auto, (max-width: 1600px) 100vw, 1600px\"><\/p>\n<p>ChatGPT converts the search query and each chunk into a numerical representation of meaning, known as an <strong>embedding<\/strong>, then measures their <strong>cosine similarity<\/strong>\u2014a score of how semantically close they&nbsp;are.<\/p>\n<p>The simplest way to picture this: imagine a giant map where similar ideas sit close together and unrelated ideas are far apart. On this map, \u201cdog\u201d and \u201cpuppy\u201d would be near each other. \u201cDog\u201d and \u201cskateboard\u201d would be on opposite ends.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"2000\" height=\"2000\" class=\"wp-image-200833\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-6-1.jpg\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-6-1.jpg 2000w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-6-1-425x425.jpg 425w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-6-1-768x768.jpg 768w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-6-1-1536x1536.jpg 1536w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-6-1-120x120.jpg 120w\" sizes=\"auto, (max-width: 2000px) 100vw, 2000px\"><\/p>\n<div class=\"sidenote\"><div class=\"sidenote-title\">Sidenote.<\/div> The connection between dog and skateboard here would obviously be in reference to <a href=\"https:\/\/www.youtube.com\/watch?v=EVnzXA9b7Ww\">Otto the Skateboarding Dog<\/a>. <\/div>\n<p>Embeddings are like GPS coordinates on that map\u2014every piece of text gets assigned coordinates based on its meaning.<\/p>\n<p>Cosine similarity is the measure of how close together two sets of coordinates are.<\/p>\n<p>The AI retrieves the chunks whose coordinates are closest to the fan-out query\u2019s coordinates, and the closest match&nbsp;wins.<\/p>\n<p>This is why specific, clear language helps retrieval\u2014it\u2019s easier to map to the correct vector \u201ccoordinate\u201d.<\/p>\n<h3><a id=\"post-200818-_39706nxn8gm2\"><\/a>Step 3: Retrieved content loads into the AI\u2019s working memory\u2014briefly<\/h3>\n<p>The matching chunks are loaded into the AI\u2019s <strong>context window<\/strong>\u2014its short-term working memory\u2014alongside the user\u2019s original question.<\/p>\n<p>It synthesizes an answer using both, then it deletes the chunks.<\/p>\n<p><a href=\"https:\/\/dejan.ai\/blog\/search-grounding-is-transient\/\">Dan Petrovic tested this directly<\/a>: he asked an AI model to retrieve information on a well-known person, then in a follow-up message asked it to recall a specific snippet from its sources. It couldn\u2019t.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"877\" height=\"551\" class=\"wp-image-200835\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-7-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-7-1.png 877w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-7-1-676x425.png 676w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-7-1-768x483.png 768w\" sizes=\"auto, (max-width: 877px) 100vw, 877px\"><\/p>\n<p>The raw content is \u201cpurged\u201d the moment a response is generated.<\/p>\n<div class=\"post-nav-link clearfix\" id=\"section1\"><a class=\"subhead-anchor\" data-tip=\"tooltip__copielink\" rel=\"#section1\"><svg width=\"19\" height=\"19\" viewBox=\"0 0 14 14\" style><g fill=\"none\" fill-rule=\"evenodd\"><path d=\"M0 0h14v14H0z\" \/><path d=\"M7.45 9.887l-1.62 1.621c-.92.92-2.418.92-3.338 0a2.364 2.364 0 0 1 0-3.339l1.62-1.62-1.273-1.272-1.62 1.62a4.161 4.161 0 1 0 5.885 5.884l1.62-1.62L7.45 9.886zM5.527 5.135L7.17 3.492c.92-.92 2.418-.92 3.339 0 .92.92.92 2.418 0 3.339L8.866 8.473l1.272 1.273 1.644-1.643A4.161 4.161 0 1 0 5.897 2.22L4.254 3.863l1.272 1.272zm-.66 3.998a.749.749 0 0 1 0-1.06l2.208-2.206a.749.749 0 1 1 1.06 1.06L5.928 9.133a.75.75 0 0 1-1.061 0z\" style \/><\/g><\/svg><\/a><div class=\"link-text\" data-anchor=\"RAG vs. training data: what\u2019s the difference?\" data-section=\"rag-vs-training\">\n<h2><a id=\"post-200818-_8g321l9tq0un\"><\/a>RAG vs.&nbsp;training data: what\u2019s the difference?<\/h2>\n<\/div><\/div>\n<p>RAG and training data often get conflated, but they do very different things.<\/p>\n<p><strong>Training data<\/strong> is what builds an AI model\u2019s parametric memory\u2014the internal knowledge I mentioned earlier.<\/p>\n<p>This happens during pretraining, when the model learns from a huge, general corpus of text scraped from the web and other sources (e.g. CommonCrawl), and can happen again during fine-tuning, when a model is further trained on a narrower dataset to shift its behavior or knowledge.<\/p>\n<p>Either way, that knowledge gets baked into the model itself. No lookup needed\u2014it\u2019s just part of what the model \u201cknows\u201d.<\/p>\n<p>But you don\u2019t get a say in it. It happens on the model developer\u2019s schedule, using whatever data they choose to train on. It\u2019s not something you can request, target, or verify happened for your content.<\/p>\n<p><strong>RAG<\/strong>, on the other hand, is a process you have some control over. The quality, structure, and indexing of your content directly affects whether it gets retrieved.<\/p>\n<p>Whenever a user\u2019s query triggers a retrieval step, the model pulls in current information from outside data without needing to be retrained.<\/p>\n<p>For most commercial AI tools, this is the mechanism behind most up-to-date answers they give&nbsp;you.<\/p>\n\n<table id=\"tablepress-546\" class=\"tablepress tablepress-id-546 tablepress-responsive tablepress-ahrefs-width-720px\">\n<thead>\n<tr class=\"row-1 odd\">\n\t<th class=\"column-1\"><\/th><th class=\"column-2\">RAG<\/th><th class=\"column-3\">Training data<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr class=\"row-2 even\">\n\t<td class=\"column-1\">How it&nbsp;works<\/td><td class=\"column-2\">Retrieves external content at query time\u2014never becomes part of the&nbsp;model<\/td><td class=\"column-3\">Content is absorbed into the model\u2019s parameters during training, becoming part of what it \u201cknows\u201d internally<\/td>\n<\/tr>\n<tr class=\"row-3 odd\">\n\t<td class=\"column-1\">Update cost<\/td><td class=\"column-2\">Low. Updates the knowledge base and the model\u2019s next answer reflects it<\/td><td class=\"column-3\">High. Only changes when the model is retrained\u2014on the developer\u2019s schedule, not&nbsp;yours<\/td>\n<\/tr>\n<tr class=\"row-4 even\">\n\t<td class=\"column-1\">Something you can influence?<\/td><td class=\"column-2\">Yes. Content quality, indexing, and structure all affect whether your content gets retrieved to surface current information, cite you as a source, or mention your specific brand<\/td><td class=\"column-3\">Not directly. You can publish content and hope it gets crawled but, unlike RAG, you can\u2019t optimize one page and hope to see it cited. What you can do is build a consistent brand narrative across enough content over time that it shapes how future models describe you.<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<!-- #tablepress-546 from cache -->\n<p>Everything we cover in this article\u2014leading with definitions, including entities, Q&amp;A structure, freshness\u2014directly influences the <strong>retrieval<\/strong> process.<\/p>\n<p>That\u2019s the layer you can actually influence through content.<\/p>\n<p>Being cited in an AI response is a win, but it\u2019s not the same as being baked into what an AI fundamentally knows about your&nbsp;brand.<\/p>\n<p>Search User Optimization expert <a href=\"https:\/\/www.linkedin.com\/in\/dshapowseo\">Dorron Shapow<\/a> puts it&nbsp;well:<\/p>\n<blockquote class=\"small\"><div class=\"quote-content\">Optimizing for retrieval isn\u2019t wrong.&nbsp;In systems that rely heavily on live search for commercial queries, it can absolutely influence what gets surfaced.&nbsp;But assuming retrieval visibility is the same as foundational model weighting is where the strategy breaks.&nbsp;One takes weeks.&nbsp;The other is the slow work of entity coherence\u2014how consistently and clearly your brand is understood across the broader web\u2014and it takes&nbsp;years.\u201d<\/div><div class=\"quote-info clearfix\"><div class=\"quote-photo\"><img decoding=\"async\" alt=\"Dorron Shapow\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/Dorron-Shapow.jpeg\"><\/div><div class=\"extra-box\"><span class=\"quote-author\">Dorron Shapow,<\/span> <span class=\"quote-author-job\">Search User Optimization Expert<\/span><\/div><\/div><\/blockquote>\n<div class=\"post-nav-link clearfix\" id=\"section1\"><a class=\"subhead-anchor\" data-tip=\"tooltip__copielink\" rel=\"#section1\"><svg width=\"19\" height=\"19\" viewBox=\"0 0 14 14\" style><g fill=\"none\" fill-rule=\"evenodd\"><path d=\"M0 0h14v14H0z\" \/><path d=\"M7.45 9.887l-1.62 1.621c-.92.92-2.418.92-3.338 0a2.364 2.364 0 0 1 0-3.339l1.62-1.62-1.273-1.272-1.62 1.62a4.161 4.161 0 1 0 5.885 5.884l1.62-1.62L7.45 9.886zM5.527 5.135L7.17 3.492c.92-.92 2.418-.92 3.339 0 .92.92.92 2.418 0 3.339L8.866 8.473l1.272 1.273 1.644-1.643A4.161 4.161 0 1 0 5.897 2.22L4.254 3.863l1.272 1.272zm-.66 3.998a.749.749 0 0 1 0-1.06l2.208-2.206a.749.749 0 1 1 1.06 1.06L5.928 9.133a.75.75 0 0 1-1.061 0z\" style \/><\/g><\/svg><\/a><div class=\"link-text\" data-anchor=\"What is query fan-out?\" data-section=\"query-fan-out\">\n<h2>What is query fan-out?<\/h2>\n<\/div><\/div>\n<p><a href=\"https:\/\/ahrefs.com\/blog\/query-fan-out\/\">Query fan-out<\/a> is the process that happens behind the scenes when you submit a query to an AI system.<\/p>\n<p>Rather than searching your exact words, it breaks your question into multiple related sub-queries, runs each one separately, and pulls sources from the combined results.<\/p>\n<p>Say someone searches <code>\u201cWhat will happen if I swap out regular flour for wholemeal flour in a lemon drizzle\u201d<\/code> in Google, the underlying AI search model wouldn\u2019t just search that phrase, it might also look&nbsp;up:<\/p>\n<ul>\n<li>Best flour for lemon drizzle<\/li>\n<li>Baking with wholemeal flour&nbsp;tips<\/li>\n<li>How does wholemeal flour affect cake density?<\/li>\n<\/ul>\n<p>Before synthesizing an answer.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"2046\" height=\"1962\" class=\"wp-image-200837\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-8-1.jpg\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-8-1.jpg 2046w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-8-1-443x425.jpg 443w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-8-1-768x736.jpg 768w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-8-1-1536x1473.jpg 1536w\" sizes=\"auto, (max-width: 2046px) 100vw, 2046px\"><\/p>\n<p>AI does the same thing automatically on most complex queries.<\/p>\n<p>Some SEOs have been able to extract these internal sub-queries directly.<\/p>\n<p>For instance, <a href=\"https:\/\/metehanai.substack.com\/p\/google-ai-mode-query-fanouts-extraction\">Metehan Ye\u015filyurt has developed a technique<\/a> to prompt Google AI Mode into outputting the search queries it used for grounding.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"2048\" height=\"1091\" class=\"wp-image-200840\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-9-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-9-1.png 2048w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-9-1-680x362.png 680w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-9-1-768x409.png 768w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-9-1-1536x818.png 1536w\" sizes=\"auto, (max-width: 2048px) 100vw, 2048px\"><\/p>\n<p>But if you don\u2019t have time to go digging, you can also see the fan-out queries generated by ChatGPT, Grok, and Perplexity in the AI Responses report in <a href=\"https:\/\/ahrefs.com\/brand-radar\">Ahrefs Brand Radar<\/a>.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"1245\" height=\"505\" class=\"wp-image-200842\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-10-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-10-1.png 1245w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-10-1-680x276.png 680w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-10-1-768x312.png 768w\" sizes=\"auto, (max-width: 1245px) 100vw, 1245px\"> During query fan-out, the AI splits your question into smaller sub-queries, searches all of them at once, combines and <a href=\"https:\/\/metehan.ai\/blog\/chatgpt-is-using-reciprocal-rank-fusion-rrf\/\">re-ranks the results<\/a>, then merges the pages that do well across multiple searches into one final&nbsp;list.<\/p>\n<p>That list is what the AI actually reads to write your answer.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"1704\" height=\"2048\" class=\"wp-image-200844\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-11-1.jpg\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-11-1.jpg 1704w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-11-1-354x425.jpg 354w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-11-1-768x923.jpg 768w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-11-1-1278x1536.jpg 1278w\" sizes=\"auto, (max-width: 1704px) 100vw, 1704px\"><\/p>\n<p>We\u2019ve simplified the fan-out process here for ease of understanding, but for a deeper-dive read our guide: <a href=\"https:\/\/ahrefs.com\/blog\/query-fan-out\/\">What is Query Fan-Out? Understanding the Hidden Queries Driving AI Search<\/a>.<\/p>\n<div class=\"post-nav-link clearfix\" id=\"section1\"><a class=\"subhead-anchor\" data-tip=\"tooltip__copielink\" rel=\"#section1\"><svg width=\"19\" height=\"19\" viewBox=\"0 0 14 14\" style><g fill=\"none\" fill-rule=\"evenodd\"><path d=\"M0 0h14v14H0z\" \/><path d=\"M7.45 9.887l-1.62 1.621c-.92.92-2.418.92-3.338 0a2.364 2.364 0 0 1 0-3.339l1.62-1.62-1.273-1.272-1.62 1.62a4.161 4.161 0 1 0 5.885 5.884l1.62-1.62L7.45 9.886zM5.527 5.135L7.17 3.492c.92-.92 2.418-.92 3.339 0 .92.92.92 2.418 0 3.339L8.866 8.473l1.272 1.273 1.644-1.643A4.161 4.161 0 1 0 5.897 2.22L4.254 3.863l1.272 1.272zm-.66 3.998a.749.749 0 0 1 0-1.06l2.208-2.206a.749.749 0 1 1 1.06 1.06L5.928 9.133a.75.75 0 0 1-1.061 0z\" style \/><\/g><\/svg><\/a><div class=\"link-text\" data-anchor=\"Why RAG means SEO still matters\" data-section=\"why-rag-means-seo-still-matters\">\n<h2><a id=\"post-200818-_swj5t7kk5jpc\"><\/a>Why RAG means SEO still matters<\/h2>\n<\/div><\/div>\n<p>For ChatGPT and other AI search engines, Retrieval Augmented Generation runs on&nbsp;SEO.<\/p>\n<p>In fact, many marketers and SEOs view AI search as a \u201cwrapper\u201d on top of \u201ctraditional\u201d search engines like Google, since some AI assistants draw so heavily from&nbsp;them.<\/p>\n<p><a href=\"https:\/\/primaryposition.com\/blog\/perplexity-crawl-index\/\"><img loading=\"lazy\" decoding=\"async\" width=\"1704\" height=\"792\" class=\"wp-image-200846\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-12-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-12-1.png 1704w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-12-1-680x316.png 680w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-12-1-768x357.png 768w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-12-1-1536x714.png 1536w\" sizes=\"auto, (max-width: 1704px) 100vw, 1704px\"><\/a><\/p>\n<p>When ChatGPT, Perplexity, or Google AI Overviews need to answer a question, they run actual web searches <a href=\"https:\/\/lnkd.in\/p\/edkNJQYh\">\u00b9<\/a> <a href=\"https:\/\/www.linkedin.com\/feed\/update\/urn:li:activity:7474912135588896768\/\">\u00b2<\/a> <a href=\"https:\/\/ahrefs.com\/blog\/is-chatgpt-really-powered-by-google\/\">\u00b3<\/a><\/p>\n<p>Google Gemini and AI Overviews use Google Search. Microsoft Copilot uses Bing. ChatGPT pulls from both Google and Bing. Claude uses Brave Search.<\/p>\n\n<table id=\"tablepress-481\" class=\"tablepress tablepress-id-481 tablepress-responsive tablepress-ahrefs-width-720px\">\n<thead>\n<tr class=\"row-1 odd\">\n\t<th class=\"column-1\">AI assistant<\/th><th class=\"column-2\">Official search index\/crawler<\/th><th class=\"column-3\">Suspected search index\/crawler<\/th><th class=\"column-4\">Suspected future search index\/crawler<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr class=\"row-2 even\">\n\t<td class=\"column-1\">ChatGPT<\/td><td class=\"column-2\">Bing (<a href=\"https:\/\/help.openai.com\/en\/articles\/10093903\">OpenAI Help Center<\/a>)<\/td><td class=\"column-3\">Google Search (<a href=\"https:\/\/www.zdnet.com\/article\/chatgpt-is-reportedly-scraping-google-search-data-to-answer-your-questions-heres-how\/\">ZDNET<\/a>)<br>\n<\/td><td class=\"column-4\">Own index (<a href=\"https:\/\/drive.google.com\/file\/d\/1WfKxSJIJLnYfxEpyjfwtsyBUNI8XazY0\/view\">Leaked memo<\/a>)<\/td>\n<\/tr>\n<tr class=\"row-3 odd\">\n\t<td class=\"column-1\">Gemini<\/td><td class=\"column-2\">Google Search (<a href=\"https:\/\/ai.google.dev\/gemini-api\/docs\/google-search\">Google AI Devs<\/a>)<\/td><td class=\"column-3\">Google Knowledge Graph (<a href=\"https:\/\/developers.google.com\/knowledge-graph\/\">Google Dev Docs<\/a>)<br>\n<\/td><td class=\"column-4\">-<\/td>\n<\/tr>\n<tr class=\"row-4 even\">\n\t<td class=\"column-1\">Perplexity<\/td><td class=\"column-2\">PerplexityBot \/ own index (<a href=\"https:\/\/docs.perplexity.ai\/guides\/bots\">Perplexity docs<\/a>)<br>\n<\/td><td class=\"column-3\">Undisclosed third-party crawlers (According to <a href=\"https:\/\/www.fastcompany.com\/91144894\/perplexity-ai-ceo-aravind-srinivas-on-plagiarism-accusations#:~:text=%E2%80%9CI%20think%20there%20is%20a%20basic%20misunderstanding%20of%20the%20way%20this%20works%2C%E2%80%9D%20Srinivas%20said.%20%E2%80%9CWe%20don%E2%80%99t%20just%20rely%20on%20our%20own%20web%20crawlers%2C%20we%20rely%20on%20third%2Dparty%20web%20crawlers%20as%20well.%E2%80%9D%C2%A0%C2%A0\">Fast Company<\/a>)<br>\n and Google Search (According to <a href=\"https:\/\/searchengineland.com\/openai-chatgpt-serpapi-google-search-results-461226\">SEL<\/a>)<br>\n<\/td><td class=\"column-4\">-<\/td>\n<\/tr>\n<tr class=\"row-5 odd\">\n\t<td class=\"column-1\">Copilot<\/td><td class=\"column-2\">Bing (<a href=\"https:\/\/support.microsoft.com\/en-us\/topic\/understanding-web-search-in-microsoft-365-copilot-chat-94c45d32-1a77-4f82-8e05-58dfb9afac48\">Microsoft Support<\/a>)<br>\n<\/td><td class=\"column-3\">-<\/td><td class=\"column-4\">-<\/td>\n<\/tr>\n<tr class=\"row-6 even\">\n\t<td class=\"column-1\">Claude<\/td><td class=\"column-2\">Brave Search (<a href=\"https:\/\/app.vanta.com\/anthropic\/trust\/iz673w96495gyjer8h78n\/updates#:~:text=Anthropic%20Subprocessor%20Changes\">Anthropic Trust Center<\/a><br>\n \u2192 noted by <a href=\"https:\/\/simonwillison.net\/2025\/Mar\/21\/anthropic-use-brave\/\">Simon Willison<\/a>)<\/td><td class=\"column-3\">-<\/td><td class=\"column-4\">-<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<!-- #tablepress-481 from cache -->\n<p>That means the retrieval layer of every major AI tool is powered by a traditional search engine.<\/p>\n<ul>\n<li><strong>Indexed content is the starting pool.<\/strong> You need your content to show up in Google before it shows up in&nbsp;AI.<\/li>\n<li><strong>Search optimized content gets you cited: <\/strong>Even if search and AI <a href=\"https:\/\/ahrefs.com\/blog\/ai-overview-citations-top-10\/\">results don\u2019t always neatly overlap<\/a>, both prioritize authoritative, well-structured, well-optimized content.<\/li>\n<li><strong>Brand mentions in search correlate strongly with AI visibility:<\/strong> AI systems pick up on how often and where your brand is referenced across the web\u2014search-optimized content and digital PR directly feeds this <a href=\"https:\/\/ahrefs.com\/blog\/ai-brand-visibility-correlations\/\">\u00b9<\/a><\/li>\n<\/ul>\n<p>Despite some differences, SEO and GEO are intrinsically linked.<\/p>\n<p>If your content doesn\u2019t show up in a search index, an AI bot is going to have a hard time finding it, and if it can\u2019t find it, it can\u2019t retrieve it.<\/p>\n<div class=\"post-nav-link clearfix\" id=\"section1\"><a class=\"subhead-anchor\" data-tip=\"tooltip__copielink\" rel=\"#section1\"><svg width=\"19\" height=\"19\" viewBox=\"0 0 14 14\" style><g fill=\"none\" fill-rule=\"evenodd\"><path d=\"M0 0h14v14H0z\" \/><path d=\"M7.45 9.887l-1.62 1.621c-.92.92-2.418.92-3.338 0a2.364 2.364 0 0 1 0-3.339l1.62-1.62-1.273-1.272-1.62 1.62a4.161 4.161 0 1 0 5.885 5.884l1.62-1.62L7.45 9.886zM5.527 5.135L7.17 3.492c.92-.92 2.418-.92 3.339 0 .92.92.92 2.418 0 3.339L8.866 8.473l1.272 1.273 1.644-1.643A4.161 4.161 0 1 0 5.897 2.22L4.254 3.863l1.272 1.272zm-.66 3.998a.749.749 0 0 1 0-1.06l2.208-2.206a.749.749 0 1 1 1.06 1.06L5.928 9.133a.75.75 0 0 1-1.061 0z\" style \/><\/g><\/svg><\/a><div class=\"link-text\" data-anchor=\"How to optimize your content for RAG\" data-section=\"optimize-for-rag\">\n<h2>How to optimize your content for&nbsp;RAG<\/h2>\n<\/div><\/div>\n<p>Follow these seven best practice tips if you want to get your content cited in RAG search.<\/p>\n<h3><a id=\"post-200818-_lzfutz4w1m2f\"><\/a>Make sure content is accessible to AI crawlers<\/h3>\n<p>When they go out to fetch content, many AI crawlers are unable to read and cite certain pages.<\/p>\n<p>JavaScript content (like tabs or accordions) or text in images is often inaccessible to AI&nbsp;bots.<\/p>\n<p>Instead, AI systems retrieve static HTML content.<\/p>\n<div id=\"attachment_200848\" style=\"width: 833px\" class=\"wp-caption alignnone\"><a href=\"https:\/\/www.linkedin.com\/feed\/update\/urn:li:activity:7350871819031089152\"><img loading=\"lazy\" decoding=\"async\" aria-describedby=\"caption-attachment-200848\" class=\"wp-image-200848\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-13-1.png\" alt width=\"823\" height=\"639\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-13-1.png 823w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-13-1-547x425.png 547w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-13-1-768x596.png 768w\" sizes=\"auto, (max-width: 823px) 100vw, 823px\"><\/a><p id=\"caption-attachment-200848\" class=\"wp-caption-text\">\u2014LinkedIn, Daniel Foley Carter.<\/p><\/div>\n<p>Here\u2019s what happens when a page contains JavaScript.<\/p>\n<p><a href=\"https:\/\/ae.linkedin.com\/in\/suganthan-mohanadasan\">Suganthan Mohanadasan<\/a> recently tapped into the <a href=\"https:\/\/suganthan.com\/blog\/how-chatgpt-picks-sources\/\">network files of dozens of ChatGPT conversations<\/a>, and studied the model\u2019s chain-of-thought process, where it describes how it sources information in layman\u2019s terms.<\/p>\n<p>For a relevant B2B SaaS query, ChatGPT located official pricing for Ahrefs but struggled to find prices for Profound and Peec, reasoning that this information was hidden within JavaScript.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"1145\" height=\"2048\" class=\"wp-image-200851\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-14-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-14-1.png 1145w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-14-1-238x425.png 238w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-14-1-768x1374.png 768w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-14-1-859x1536.png 859w\" sizes=\"auto, (max-width: 1145px) 100vw, 1145px\"><\/p>\n<p>ChatGPT deferred to third-party sources like G2 since \u201cthe official page is hard to parse and doesn\u2019t show prices\u201d.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"943\" height=\"753\" class=\"wp-image-200854\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-15-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-15-1.png 943w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-15-1-532x425.png 532w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-15-1-768x613.png 768w\" sizes=\"auto, (max-width: 943px) 100vw, 943px\"><\/p>\n<p>The moral of the story: if you want your most important information\u2014like your pricing\u2014 to be accurately portrayed in AI search, your content should ideally be served via HTML, not JavaScript.<\/p>\n<div class=\"sidenote\"><div class=\"sidenote-title\">Sidenote.<\/div> There is another possible explanation here: some companies don\u2019t disclose their pricing. This leaves AI to piece together that missing information with data from other sources. Even if <em>you<\/em> don\u2019t disclose your pricing, AI models <em>will<\/em>, and they won\u2019t always be&nbsp;right.<\/div>\n<p>JavaScript isn\u2019t the only way to lock a crawler out\u2014you also need to avoid blocking AI crawlers (like <code>OAI_SearchBot<\/code>) in your robots.txt and firewall rules if you want to be cited via retrieval <a href=\"https:\/\/www.linkedin.com\/feed\/update\/urn:li:activity:7258439559896596480\">\u00b9<\/a> <a href=\"https:\/\/searchengineland.com\/ai-optimization-how-to-optimize-your-content-for-ai-search-and-agents-451287\">\u00b2<\/a>.<\/p>\n<p>If you use Cloudflare, you can monitor how AI bots are crawling your site\u2014including which pages they visit most often and which ones they miss\u2014via <a href=\"https:\/\/ahrefs.com\/bot-analytics\">Ahrefs Bot Analytics<\/a>.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"2048\" height=\"1905\" class=\"wp-image-200856\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-16-1.jpg\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-16-1.jpg 2048w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-16-1-457x425.jpg 457w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-16-1-768x714.jpg 768w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-16-1-1536x1429.jpg 1536w\" sizes=\"auto, (max-width: 2048px) 100vw, 2048px\"><\/p>\n<div class=\"recommendation\"><div class=\"recommendation-title\">Beware of CDNs blocking AI and multipurpose crawlers<\/div><div class=\"recommendation-content\">\n<p>Check your Content Delivery Networks (CDNs) default crawl settings to make sure you\u2019re not inadvertently blocking your content from retrieval.<\/p>\n<p>For example, Cloudflare blocks all AI crawlers by default, which can limit your website\u2019s visibility on interfaces like ChatGPT, Claude, and Gemini.<\/p>\n<p>Even more crucially, it may also block multipurpose crawlers that combine AI training and search engine visibility, like <code>Googlebot<\/code> and <code>BingBot<\/code>.<\/p>\n<p><a href=\"https:\/\/www.linkedin.com\/feed\/update\/urn:li:activity:7478535383857766400\/\"><img loading=\"lazy\" decoding=\"async\" width=\"1208\" height=\"1702\" class=\"wp-image-200859\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-17-1.jpg\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-17-1.jpg 1208w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-17-1-302x425.jpg 302w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-17-1-768x1082.jpg 768w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-17-1-1090x1536.jpg 1090w\" sizes=\"auto, (max-width: 1208px) 100vw, 1208px\"><\/a><\/p>\n<p>\u2014LinkedIn, <a href=\"https:\/\/www.linkedin.com\/feed\/update\/urn:li:activity:7478535383857766400\/\">Suganthan Mohanadasan<\/a>, <a href=\"https:\/\/www.linkedin.com\/feed\/update\/urn:li:activity:7424395610159665152\/\">Dixon Jones<\/a>, and <a href=\"https:\/\/www.linkedin.com\/feed\/update\/urn:li:activity:7431751075940102145\">Mark Williams-Cook<\/a>.<\/p>\n<\/div><\/div>\n<h3>Lead with your best information<\/h3>\n<p>AI pays the most attention to the beginning of your page, but its attention drops steadily from&nbsp;there.<\/p>\n<p>According to<a href=\"https:\/\/www.growth-memo.com\/p\/the-science-of-how-ai-pays-attention\"> Kevin Indig\u2019s study of 1.2 million ChatGPT citations<\/a>, the first 30% of a page\u2019s content generates 44.2% of all citations.<\/p>\n<p>The middle third generates 31.1%, and the bottom third: just 24.7%.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"1456\" height=\"1092\" class=\"wp-image-200860\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-18-1.jpg\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-18-1.jpg 1456w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-18-1-567x425.jpg 567w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-18-1-768x576.jpg 768w\" sizes=\"auto, (max-width: 1456px) 100vw, 1456px\"><\/p>\n<p>Your most important information\u2014definitions, key claims, unique data\u2014needs to be at the very top of your content.<\/p>\n<p>This is the opposite of the traditional \u201csave the best for last\u201d approach. In content optimized for AI citations, the punchline goes&nbsp;first.<\/p>\n<p>This is known as serving the Bottom Line Up Front (BLUF).<\/p>\n<p>Answer the query immediately in the first sentence below the subheading\u2014don\u2019t bury the answer two paragraphs in.<\/p>\n<p>This directly mirrors how RAG systems match content to queries\u2014but also, how users read, so you\u2019re satisfying both beings and bots&nbsp;alike!<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"785\" height=\"392\" class=\"wp-image-200863\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-19-1.jpg\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-19-1.jpg 785w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-19-1-680x340.jpg 680w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-19-1-768x384.jpg 768w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-19-1-400x200.jpg 400w\" sizes=\"auto, (max-width: 785px) 100vw, 785px\"><\/p>\n<p>This eye-tracking data shows readers concentrate the most attention at the very top of a page and scan less and less as they move down, so if your key takeaway is buried in paragraph three, most readers never actually see it\u2014hence, \u201cbottom line up&nbsp;front\u201d.<\/p>\n<h3><a id=\"post-200818-_3qn7wy81nt3h\"><\/a>Optimize for fan-out topics<\/h3>\n<p>To show up in the fan-out results that AI systems draw on, it\u2019s helpful to create <strong>topic clusters<\/strong>\u2014the related questions, definitions, comparisons, and subtopics that AI might search for while preparing an answer.<\/p>\n<p>If you\u2019re looking for hints as to what those sub-topics might be, tap into \u201cPeople also ask\u201d boxes and \u201cPeople also search for\u201d queries at the bottom of Google.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"2048\" height=\"839\" class=\"wp-image-200864\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-20-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-20-1.png 2048w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-20-1-680x279.png 680w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-20-1-768x315.png 768w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-20-1-1536x629.png 1536w\" sizes=\"auto, (max-width: 2048px) 100vw, 2048px\"><\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"2048\" height=\"557\" class=\"wp-image-200866\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-21-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-21-1.png 2048w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-21-1-680x185.png 680w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-21-1-768x209.png 768w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-21-1-1536x418.png 1536w\" sizes=\"auto, (max-width: 2048px) 100vw, 2048px\"><\/p>\n<p>They reflect the most-asked questions and angles around your topic, which tend to be similar to the queries AI generates in a fan-out.<\/p>\n<div class=\"recommendation\"><div class=\"recommendation-title\">Tip<\/div><div class=\"recommendation-content\"><!-- notionvc: 0c9fa10b-2016-46df-9473-0d2f35b22564 -->\n<p>Check out the Questions tab in <a href=\"https:\/\/ahrefs.com\/keywords-explorer\">Ahrefs\u2019 Keywords Explorer<\/a> to find related queries being asked around your topic and map out a topic cluster.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"984\" height=\"545\" class=\"wp-image-200867\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-22-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-22-1.png 984w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-22-1-680x377.png 680w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-22-1-768x425.png 768w\" sizes=\"auto, (max-width: 984px) 100vw, 984px\"><\/p>\n<p>If you\u2019re not covering specific subtopics, you\u2019ll be invisible in a significant chunk of fan-out query search results.<\/p>\n<\/div><\/div>\n<h3><a id=\"post-200818-_yryw7fw7hd4\"><\/a>Optimize your page&nbsp;speed<\/h3>\n<p>Slow pages are bad news in any search engine, but in AI search the cost is even steeper.<\/p>\n<p>In his <a href=\"https:\/\/queryburst.com\/blog\/how-chatgpt-works\/\">breakdown of how ChatGPT works<\/a>, SEO Consultant <a href=\"https:\/\/uk.linkedin.com\/in\/david-mcsweeney-79840154\">David McSweeney<\/a> notes that ChatGPT appears to fetch grounding pages on a hard timeout of around two seconds: if your server is slow, your page gets cut, and even if it responds in time, a high <a href=\"https:\/\/ahrefs.com\/blog\/pagespeed-insights\/\">time-to-first-byte<\/a> (TTFB) means your content gets truncated.<\/p>\n<p><strong>Under 1 second TTFB:<\/strong> you\u2019re probably fine. Your full page has time to load, get chunked up, and fed to the&nbsp;model.<\/p>\n<p><strong>Over 1 second:<\/strong> you\u2019re gambling. The connection might get cut mid-download\u2014sometimes so early that only your &lt;head&gt; tag made it through, meaning the model never even saw your actual content.<\/p>\n<p>Speed decides whether you make it into the model\u2019s context window at&nbsp;all.<\/p>\n<p>Check your time-to-first-byte in <a href=\"https:\/\/ahrefs.com\/site-audit\">Site Audit<\/a>.<\/p>\n<ol>\n<li>Head to the <a href=\"https:\/\/ahrefs.com\/website-performance\">Performance<\/a> report<\/li>\n<li>Find the \u201cTime to first byte distribution\u201d chart<\/li>\n<li>Click \u201cMedium: 200\u2013300 ms\u201d for your quick-win optimization opportunities<\/li>\n<\/ol>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"906\" height=\"927\" class=\"wp-image-200869\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-23-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-23-1.png 906w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-23-1-415x425.png 415w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-23-1-768x786.png 768w\" sizes=\"auto, (max-width: 906px) 100vw, 906px\"><\/p>\n<p>Then sort by <strong>organic traffic<\/strong> to find your most important content that may need to be optimized<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"1226\" height=\"875\" class=\"wp-image-200871\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-24-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-24-1.png 1226w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-24-1-595x425.png 595w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-24-1-768x548.png 768w\" sizes=\"auto, (max-width: 1226px) 100vw, 1226px\"><\/p>\n<p>If your server is too slow, your page may never make it into an AI answer\u2014but in some cases you\u2019ll never know, because the visitor (in this case, a bot) simply gave up and&nbsp;left.<\/p>\n<p><a href=\"https:\/\/www.linkedin.com\/feed\/update\/urn:li:activity:7478056086680510464\/\">Jan-Willem Bobbink<\/a> looks for instances of this by identifying the HTTP status code 499 in his server logs.<\/p>\n<p>A 499 status code means the client closed the connection before the server finished responding.<\/p>\n<p>This is another clear signal that your site is too slow for AI retrieval.<\/p>\n<h3>Create deep, entity-led content<\/h3>\n<p>The content that gets cited most often via RAG search contains <a href=\"https:\/\/www.growth-memo.com\/p\/the-science-of-how-ai-pays-attention\">roughly 20.6% entity density<\/a>.<\/p>\n<p>Meaning, 20.6% of its words are proper nouns\u2014named tools, brands, people, companies, studies\u2014compared to 5-8% in \u201caverage\u201d content.<\/p>\n<p>An entity is any specific named thing. For example, \u201cAn SEO tool\u201d is not an entity\u2014 but \u201cAhrefs\u201d is.<\/p>\n<p>The more named entities you include, the more anchor points your content has on the meaning map\u2014making it retrievable for a broader range of related queries.<\/p>\n<p>But you\u2019re not going to win citations by randomly \u201centity stuffing\u201d. Your content, and its entities, need to be relevant to the user\u2019s query.<\/p>\n<p>Here\u2019s another reason entities matter.<\/p>\n<p>Fan-out queries often use a \u201c<a href=\"https:\/\/x.com\/top5seo\/status\/1998793512150765959\">synonym cloud<\/a>\u201d technique to steer retrieval towards specific angles and <strong>entities<\/strong>, and ultimately better match the intent of the user\u2019s original query.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"1189\" height=\"1466\" class=\"wp-image-200873\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-25-1.jpg\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-25-1.jpg 1189w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-25-1-345x425.jpg 345w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-25-1-768x947.jpg 768w\" sizes=\"auto, (max-width: 1189px) 100vw, 1189px\"><\/p>\n<p>For example, ChatGPT\u2019s frontier model may transform a query like \u201cWhat are the 10 best running shoes?\u201d into synonym-rich fan-out queries like:<\/p>\n<ul>\n<li>best running shoes&nbsp;2026<\/li>\n<li>reviews running shoes<\/li>\n<li>top picks<\/li>\n<li>awards<\/li>\n<\/ul>\n<p>To nudge the embedding toward \u201cbest of\u201d intent, as seen below via <a href=\"https:\/\/ahrefs.com\/brand-radar\">Brand Radar<\/a>.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"2048\" height=\"390\" class=\"wp-image-200875\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-26-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-26-1.png 2048w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-26-1-680x129.png 680w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-26-1-768x146.png 768w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-26-1-1536x293.png 1536w\" sizes=\"auto, (max-width: 2048px) 100vw, 2048px\"><\/p>\n<p>So what does this mean for your content?<\/p>\n<p>Well, to paraphrase David McSweeney: Generic pages that mention everything score <em>okay<\/em> across the&nbsp;board.<\/p>\n<p>But specialized pages that go deep on one angle win that angle outright.<\/p>\n<p>Getting cited is therefore about anchoring your content to <strong>specific <\/strong>entities.<\/p>\n<div class=\"recommendation\"><div class=\"recommendation-title\">Include fan-out query entities in your page&nbsp;title<\/div><div class=\"recommendation-content\">\n<p>Our study of <a href=\"https:\/\/ahrefs.com\/blog\/why-chatgpt-cites-pages\/\">1.4 million ChatGPT prompts<\/a> found <strong>cited pages<\/strong> have titles more semantically similar to ChatGPT\u2019s internal fanout queries than pages that got passed over.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"1431\" height=\"1536\" class=\"wp-image-200877\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-27-1.jpg\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-27-1.jpg 1431w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-27-1-396x425.jpg 396w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-27-1-768x824.jpg 768w\" sizes=\"auto, (max-width: 1431px) 100vw, 1431px\"><\/p>\n<p><a href=\"https:\/\/ahrefs.com\/brand-radar\">Brand Radar<\/a> shows the fan-out queries behind any prompt, so you can check whether your title entities match fan-out entities.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"1536\" height=\"921\" class=\"wp-image-200879\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-28-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-28-1.png 1536w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-28-1-680x408.png 680w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-28-1-768x461.png 768w\" sizes=\"auto, (max-width: 1536px) 100vw, 1536px\"><\/p>\n<\/div><\/div>\n<p>Here\u2019s a practical way to enrich your content with entities: go through your back catalog and replace generics with specifics.<\/p>\n<p>Change:<\/p>\n<ul>\n<li>\u201cA search engine\u201d \u2192 \u201cGoogle\u201d<\/li>\n<li>\u201cResearch suggests\u201d \u2192 \u201cA 2024 study from Waseda University found\u201d<\/li>\n<li>\u201cAn AI assistant\u201d \u2192 \u201cChatGPT\u201d or \u201cPerplexity\u201d<\/li>\n<\/ul>\n<p>You can verify your work using<a href=\"https:\/\/cloud.google.com\/natural-language\"> Google\u2019s Natural Language API<\/a>.<\/p>\n<p>The free demo version shows you every entity Google detected on your page, and the category it assigned your content to.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"1622\" height=\"1842\" class=\"wp-image-200882\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-29-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-29-1.png 1622w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-29-1-374x425.png 374w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-29-1-768x872.png 768w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-29-1-1353x1536.png 1353w\" sizes=\"auto, (max-width: 1622px) 100vw, 1622px\"><\/p>\n<p>If you pay for full access, you\u2019ll also get the salience score\u2014a value for how prominent and important Google thinks an entity is to your&nbsp;page.<\/p>\n<p>Run the API on your page, then run it on the top-ranking page for your target keyword.<\/p>\n<p>The gap between those two outputs gives you your entity optimization checklist:<\/p>\n<ul>\n<li>Entity crossover<\/li>\n<li>Entity gaps<\/li>\n<li>Salience scores (higher when the topic is named earlier and more prominently)<\/li>\n<li>Category crossover<\/li>\n<li>Category gaps<\/li>\n<\/ul>\n<p>Alternatively, run your draft through <a href=\"https:\/\/ahrefs.com\/ai-content-helper\">Ahrefs\u2019 AI Content Helper<\/a>.<\/p>\n<p>It grades your content against your top competitors for your target keyword and highlights the topics they cover that you\u2019re missing\u2014useful for catching topic gaps that might make you invisible in fan-out results.<\/p>\n<div data-mode=\"normal\" data-oembed=\"1\" data-provider=\"youtube\" id=\"arve-youtube-fanrnhgn_tk\" style=\"max-width:900px;\" class=\"arve\">\n<div class=\"arve-inner\">\n<div style=\"aspect-ratio:4\/3\" class=\"arve-embed arve-embed--has-aspect-ratio\">\n<div class=\"arve-ar\" style=\"padding-top:75.000000%\"><\/div>\n<p>\t\t\t<iframe allow=\"accelerometer 'none';autoplay 'none';bluetooth 'none';browsing-topics 'none';camera 'none';clipboard-read 'none';clipboard-write;display-capture 'none';encrypted-media 'none';gamepad 'none';geolocation 'none';gyroscope 'none';hid 'none';identity-credentials-get 'none';idle-detection 'none';keyboard-map 'none';local-fonts;magnetometer 'none';microphone 'none';midi 'none';otp-credentials 'none';payment 'none';picture-in-picture;publickey-credentials-create 'none';publickey-credentials-get 'none';screen-wake-lock 'none';serial 'none';summarizer 'none';sync-xhr;usb 'none';web-share;window-management 'none';xr-spatial-tracking 'none';\" allowfullscreen class=\"arve-iframe fitvidsignore\" credentialless data-arve=\"arve-youtube-fanrnhgn_tk\" data-lenis-prevent data-src-no-ap=\"https:\/\/www.youtube-nocookie.com\/embed\/FanRNHgN_Tk?feature=oembed&amp;iv_load_policy=3&amp;modestbranding=1&amp;rel=0&amp;autohide=1&amp;playsinline=0&amp;autoplay=0\" frameborder=\"0\" height=\"675\" loading=\"lazy\" name referrerpolicy=\"strict-origin-when-cross-origin\" sandbox=\"allow-scripts allow-same-origin allow-presentation allow-popups allow-popups-to-escape-sandbox\" scrolling=\"no\" src=\"https:\/\/www.youtube-nocookie.com\/embed\/FanRNHgN_Tk?feature=oembed&amp;iv_load_policy=3&amp;modestbranding=1&amp;rel=0&amp;autohide=1&amp;playsinline=0&amp;autoplay=0\" title width=\"900\"><\/iframe><\/p><\/div>\n<\/div>\n<p>\t<script type=\"application\/ld+json\">{\"@context\":\"http:\\\/\\\/schema.org\\\/\",\"@id\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/retrieval-augmented-generation\\\/#arve-youtube-fanrnhgn_tk\",\"@type\":\"VideoObject\",\"embedURL\":\"https:\\\/\\\/www.youtube-nocookie.com\\\/embed\\\/FanRNHgN_Tk?feature=oembed&iv_load_policy=3&modestbranding=1&rel=0&autohide=1&playsinline=0&autoplay=0\"}<\/script><\/p>\n<\/div>\n<h3><a id=\"post-200818-_b09jbftfh3uy\"><\/a>Add information gain\u2014say something the model doesn\u2019t already know<\/h3>\n<p>Entity coverage gets you retrieved, but there\u2019s something that comes before that: does your content even qualify for retrieval in the first&nbsp;place?<\/p>\n<p><a href=\"https:\/\/www.linkedin.com\/feed\/update\/urn:li:activity:7341098643317293056\">A leaked Claude system prompt<\/a> revealed that AI systems like Claude have a <code>never_search<\/code> command for queries about \u201ctimeless or stable\u201d information.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"800\" height=\"431\" class=\"wp-image-200883\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-30-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-30-1.png 800w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-30-1-680x366.png 680w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-30-1-768x414.png 768w\" sizes=\"auto, (max-width: 800px) 100vw, 800px\"><\/p>\n<p>Claude answers <code>never_search<\/code> questions from training data alone, and doesn\u2019t go looking for external URLs to&nbsp;cite.<\/p>\n<p>Growth Advisor <a href=\"https:\/\/www.linkedin.com\/in\/officialg\">Gaetano DiNardi<\/a> thinks other LLMs are likely following the same logic. In his&nbsp;words:<\/p>\n<blockquote><p>\u201c<em>the value of publishing pages on generalized knowledge is zero.<\/em>\u201d<\/p><\/blockquote>\n<p>This is the information gain problem.<\/p>\n<p>Think of everything a model already knows as the overculture\u2014the averaged-out, consensus version of a topic that\u2019s been indexed thousands of&nbsp;times.<\/p>\n<p>If your content only restates it, you\u2019re redundant from the RAG framework\u2014an AI model has nothing to gain from citing you.<\/p>\n<p>What it <em>does<\/em> cite is content that adds something new: proprietary data, a named theory, a specific finding from a study, a conclusion the model can\u2019t synthesize from its existing knowledge base.<\/p>\n<p>OpenAI researcher <a href=\"https:\/\/www.linkedin.com\/in\/karthik-narasimhan-65214455\/?lipi=urn%3Ali%3Apage%3Ad_flagship3_detail_base%3BjGXS4aRnRzG03EiMOpoVhw%3D%3D\">Karthik Narasimhan<\/a> published a paper on <a href=\"https:\/\/arxiv.org\/pdf\/2311.09735\">Generative Engine Optimization<\/a> that offers further proof of&nbsp;this.<\/p>\n<p>Along with peers at Princeton University, he studied which techniques are most likely to boost visibility in RAG AI systems like Perplexity.<\/p>\n<p>Their findings revealed that websites featuring unique information like quotes and statistics were most commonly referenced; seeing 30-40% visibility uplift in AI responses.<\/p>\n\n<table id=\"tablepress-359\" class=\"tablepress tablepress-id-359 tablepress-responsive tablepress-ahrefs-width-720px\">\n<thead>\n<tr class=\"row-1 odd\">\n\t<th class=\"column-1\">LLMO method tested<\/th><th class=\"column-2\">Position-adjusted word count (visibility) \ud83d\udc47<\/th><th class=\"column-3\">Subjective impression (relevance, click potential)<\/th>\n<\/tr>\n<\/thead>\n<tbody class=\"row-hover\">\n<tr class=\"row-2 even\">\n\t<td class=\"column-1\">Quotes<\/td><td class=\"column-2\">27.2<\/td><td class=\"column-3\">24.7<\/td>\n<\/tr>\n<tr class=\"row-3 odd\">\n\t<td class=\"column-1\">Statistics<\/td><td class=\"column-2\">25.2<\/td><td class=\"column-3\">23.7<\/td>\n<\/tr>\n<tr class=\"row-4 even\">\n\t<td class=\"column-1\">Fluency<\/td><td class=\"column-2\">24.7<\/td><td class=\"column-3\">21.9<\/td>\n<\/tr>\n<tr class=\"row-5 odd\">\n\t<td class=\"column-1\">Citing sources<\/td><td class=\"column-2\">24.6<\/td><td class=\"column-3\">21.9<\/td>\n<\/tr>\n<tr class=\"row-6 even\">\n\t<td class=\"column-1\">Technical terms<\/td><td class=\"column-2\">22.7<\/td><td class=\"column-3\">21.4<\/td>\n<\/tr>\n<tr class=\"row-7 odd\">\n\t<td class=\"column-1\">Easy-to-understand<\/td><td class=\"column-2\">22<\/td><td class=\"column-3\">20.5<\/td>\n<\/tr>\n<tr class=\"row-8 even\">\n\t<td class=\"column-1\">Authoritative<\/td><td class=\"column-2\">21.3<\/td><td class=\"column-3\">22.9<\/td>\n<\/tr>\n<tr class=\"row-9 odd\">\n\t<td class=\"column-1\">Unique words<\/td><td class=\"column-2\">20.5<\/td><td class=\"column-3\">20.4<\/td>\n<\/tr>\n<tr class=\"row-10 even\">\n\t<td class=\"column-1\">No optimization<\/td><td class=\"column-2\">19.3<\/td><td class=\"column-3\">19.3<\/td>\n<\/tr>\n<tr class=\"row-11 odd\">\n\t<td class=\"column-1\">Keyword stuffing<\/td><td class=\"column-2\">17.7<\/td><td class=\"column-3\">20.2<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n\n<p><a href=\"https:\/\/www.growth-memo.com\/p\/the-science-of-what-ai-actually-rewards\">Kevin Indig<\/a> also found that <strong>date<\/strong> and <strong>number<\/strong> are the entity types that predict ChatGPT citations most.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"1600\" height=\"615\" class=\"wp-image-200885\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-31-1.jpg\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-31-1.jpg 1600w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-31-1-680x261.jpg 680w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-31-1-768x295.jpg 768w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-31-1-1536x590.jpg 1536w\" sizes=\"auto, (max-width: 1600px) 100vw, 1600px\"><\/p>\n<p>And <a href=\"https:\/\/www.linkedin.com\/in\/eric-lancheres-787a772\/\">Eric Lancheres<\/a> studied 150 ranking pages and found the biggest ranking predictor was their number of <a href=\"https:\/\/api.on-page.ai\/research\/information-gain-study\">unique data points<\/a><strong>.<\/strong><\/p>\n<p><strong><img loading=\"lazy\" decoding=\"async\" width=\"1712\" height=\"1784\" class=\"wp-image-200887\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-32-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-32-1.png 1712w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-32-1-408x425.png 408w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-32-1-768x800.png 768w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-32-1-1474x1536.png 1474w\" sizes=\"auto, (max-width: 1712px) 100vw, 1712px\"><\/strong><\/p>\n<p>Having your content retrieved is a matter of surfacing fresh information and unique data, not chorusing what other pages have already covered.<\/p>\n<h3>Include a question-and-answer structure<\/h3>\n<p>Content structured as <code>question \u2192 immediate answer<\/code> is <a href=\"https:\/\/www.growth-memo.com\/p\/the-science-of-how-ai-pays-attention\">cited twice as often<\/a> as content that doesn\u2019t follow this convention (18% vs.&nbsp;8.9%), according to Kevin Indig\u2019s data.<\/p>\n<p>This is yet another example of BLUF in&nbsp;play.<\/p>\n<p>AI models try to match user queries (almost always a question) to a chunk that answers it.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"1600\" height=\"1248\" class=\"wp-image-200890\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-33-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-33-1.png 1600w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-33-1-545x425.png 545w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-33-1-768x599.png 768w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-33-1-1536x1198.png 1536w\" sizes=\"auto, (max-width: 1600px) 100vw, 1600px\"><\/p>\n<p>In the words of <a href=\"https:\/\/suganthan.com\/blog\/how-chatgpt-picks-sources\/\">Suganthan Mohanadansan:<\/a><\/p>\n<blockquote><p>\u201cCitations bind to a specific sentence, not the whole answer, so being topically relevant isn\u2019t enough, you have to be the best support for a precise claim.\u201d<\/p><\/blockquote>\n<p>Formatting your content as a Q&amp;A can help AI models like ChatGPT make a direct, unambiguous match.<\/p>\n<p><a href=\"https:\/\/suganthan.com\/blog\/how-chatgpt-picks-sources\/\">Mohanadasan<\/a> also found that ChatGPT deduplicates results by domain, so 20 thin pages on your site don\u2019t add up to 20 chances at citation.<\/p>\n<p>ChatGPT selects the one page that best matches the user\u2019s initial query and fan-out subqueries.<\/p>\n<p>Put your strongest answer on that page, not spread across all&nbsp;20.<\/p>\n<div class=\"recommendation\"><div class=\"recommendation-title\">Tip<\/div><div class=\"recommendation-content\">\n<p>In the words of <a href=\"https:\/\/www.linkedin.com\/feed\/update\/urn:li:activity:7438203479833120768\">Eli Schwartz<\/a>: \u201cThe vast majority of pages get considered and rejected before the answer is ever written.\u201d<\/p>\n<p>In <a href=\"https:\/\/ahrefs.com\/brand-radar\">Brand Radar<\/a> you can filter citations by \u201cFound but not cited\u201d to see every response where your page was pulled into ChatGPT\u2019s retrieval set and then passed over for someone else\u2019s.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"1084\" height=\"900\" class=\"wp-image-200892\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-34-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-34-1.png 1084w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-34-1-512x425.png 512w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-34-1-768x638.png 768w\" sizes=\"auto, (max-width: 1084px) 100vw, 1084px\"><\/p>\n<p>Study the pages that did get cited, and adjust your content to increase your chance of citation.<\/p>\n<\/div><\/div>\n<h3>Keep content fresh<\/h3>\n<p>RAG search systems have a preference for current content.<\/p>\n<p>We ran a study of 17 million citations, and found that <a href=\"https:\/\/ahrefs.com\/blog\/do-ai-assistants-prefer-to-cite-fresh-content\/\">AI assistants consistently prefer to cite fresher content<\/a> than search engines.<\/p>\n<p>URLs cited by AI assistants are 25.7% fresher on average than URLs in standard organic SERPs\u2014and ChatGPT and Perplexity actually order their citations from newest to oldest.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"1600\" height=\"1648\" class=\"wp-image-200893\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-35-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-35-1.png 1600w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-35-1-413x425.png 413w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-35-1-768x791.png 768w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-35-1-1491x1536.png 1491w\" sizes=\"auto, (max-width: 1600px) 100vw, 1600px\"><\/p>\n<p>But don\u2019t just take our word for it. Freshness is a confirmed, documented signal in AI retrieval.<\/p>\n<p><a href=\"https:\/\/tr.linkedin.com\/in\/metehanyesilyurt\">Metehan Ye\u015filyurt\u2019s<\/a> research confirmed this. He discovered that<a href=\"https:\/\/metehan.ai\/blog\/i-found-it-in-the-code-science-proved-it-in-the-lab-the-recency-bias-thats-reshaping-ai-search\/\"> ChatGPT has a configuration setting called<\/a> <code><a href=\"https:\/\/metehan.ai\/blog\/i-found-it-in-the-code-science-proved-it-in-the-lab-the-recency-bias-thats-reshaping-ai-search\/\"> use_freshness_scoring_profile: true<\/a><\/code>, which bakes in a systematic recency bias.<\/p>\n<p>So, your content has a much better chance of being retrieved and eventually cited if you update your key pages regularly.<\/p>\n<p>Even minor updates can reset the freshness signal. Refresh statistics and examples annually and add a visible \u201clast updated\u201d date.<\/p>\n<div class=\"sidenote\"><div class=\"sidenote-title\">Sidenote.<\/div> One thing to remember with RAG is that AI models often retrieve <strong>cached versions<\/strong> of pages rather than the live page. So if you updated your content yesterday, the AI may still be reading an older version from the search index\u2019s cache.<\/div>\n<div class=\"further-reading\"><div class=\"reading-title\">Further reading<\/div><div class=\"reading-content\">\n<ul>\n<li><a href=\"https:\/\/ahrefs.com\/blog\/republishing-content\/\">Republishing Content for SEO &amp; AI: How to Update Posts (Not Just Change Dates)<\/a><\/li>\n<li><a href=\"https:\/\/ahrefs.com\/blog\/fresh-content\/\">Fresh Content: Why Publish Dates Make or Break Rankings and AI Visibility<\/a><\/li>\n<\/ul>\n<\/div><\/div>\n<div class=\"post-nav-link clearfix\" id=\"section1\"><a class=\"subhead-anchor\" data-tip=\"tooltip__copielink\" rel=\"#section1\"><svg width=\"19\" height=\"19\" viewBox=\"0 0 14 14\" style><g fill=\"none\" fill-rule=\"evenodd\"><path d=\"M0 0h14v14H0z\" \/><path d=\"M7.45 9.887l-1.62 1.621c-.92.92-2.418.92-3.338 0a2.364 2.364 0 0 1 0-3.339l1.62-1.62-1.273-1.272-1.62 1.62a4.161 4.161 0 1 0 5.885 5.884l1.62-1.62L7.45 9.886zM5.527 5.135L7.17 3.492c.92-.92 2.418-.92 3.339 0 .92.92.92 2.418 0 3.339L8.866 8.473l1.272 1.273 1.644-1.643A4.161 4.161 0 1 0 5.897 2.22L4.254 3.863l1.272 1.272zm-.66 3.998a.749.749 0 0 1 0-1.06l2.208-2.206a.749.749 0 1 1 1.06 1.06L5.928 9.133a.75.75 0 0 1-1.061 0z\" style \/><\/g><\/svg><\/a><div class=\"link-text\" data-anchor=\"How to monitor your visibility in RAG with Ahrefs Brand Radar\" data-section=\"monitor-rag\">\n<h2>How to monitor your visibility in RAG with Ahrefs Brand&nbsp;Radar<\/h2>\n<\/div><\/div>\n<p>Optimizing your content for RAG is vital, but you need to know if it\u2019s working.<\/p>\n<p><a href=\"https:\/\/ahrefs.com\/brand-radar\">Ahrefs Brand Radar<\/a> was built to help brands monitor their visibility in retrieval augmented AI results.<\/p>\n<p>Here\u2019s how I suggest using it to improve your visibility in&nbsp;RAG.<\/p>\n<h3><a id=\"post-200818-_d9bugg9q6tcb\"><\/a>Track your baseline visibility<\/h3>\n<p>Before changing anything, find out where you actually stand.<\/p>\n<p>Search your brand in Brand Radar to see how often you\u2019re appearing in AI answers for your target topics, and which platforms are citing you.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"2048\" height=\"990\" class=\"wp-image-200894\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-36-1.jpg\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-36-1.jpg 2048w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-36-1-680x329.jpg 680w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-36-1-768x371.jpg 768w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-36-1-1536x743.jpg 1536w\" sizes=\"auto, (max-width: 2048px) 100vw, 2048px\"><\/p>\n<p>If mentions are low or absent, see who is being cited instead.<\/p>\n<h3><a id=\"post-200818-_lags5gx58wjg\"><\/a>Find out which AI platforms are citing you (and which aren\u2019t)<\/h3>\n<p>Different AI platforms have different retrieval architectures with different biases toward freshness, authority, and structure.<\/p>\n<p>Brand Radar\u2019s platform breakdown can reveal gaps like \u201cAI Mode cites us regularly, but we lack visibility in Perplexity.\u201d<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"2048\" height=\"1127\" class=\"wp-image-200895\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-37-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-37-1.png 2048w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-37-1-680x374.png 680w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-37-1-768x423.png 768w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-37-1-1536x845.png 1536w\" sizes=\"auto, (max-width: 2048px) 100vw, 2048px\"><\/p>\n<p>If your site performs badly on only one platform, the issue is likely with how that platform evaluates it\u2014not the content itself.<\/p>\n<p>For example, if a page ranked well on Google but not on Bing, we\u2019d see that as a Bing-specific signal (like links, entities, or indexing) rather than the page being low quality overall\u2014the same is true of AI visibility.<\/p>\n<h3><a id=\"post-200818-_r2jebg1ljtbw\"><\/a>Discover which queries are triggering your citations<\/h3>\n<p>Seeing the exact queries that lead to citation tells you what\u2019s working, and flags related queries where you\u2019re not appearing yet.<\/p>\n<p>Because of query fan-out, you may already be getting cited for queries you\u2019d never have thought to target.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"1327\" height=\"389\" class=\"wp-image-200896\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-38-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-38-1.png 1327w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-38-1-680x199.png 680w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-38-1-768x225.png 768w\" sizes=\"auto, (max-width: 1327px) 100vw, 1327px\"><\/p>\n<p>Brand Radar\u2019s database contains millions of existing queries, meaning you can stumble on new content opportunities you wouldn\u2019t otherwise know existed.<\/p>\n<h3><a id=\"post-200818-_fpua3uwc3b18\"><\/a>Track whether content updates change your citation rate<\/h3>\n<p>Once you\u2019ve made changes to optimize your content for retrieval\u2014applying BLUF, targeting fan-out queries, incorporating statistics\u2014monitor Brand Radar to see whether your citations grow in the following weeks.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"1127\" height=\"604\" class=\"wp-image-200897\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-39-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-39-1.png 1127w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-39-1-680x364.png 680w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-39-1-768x412.png 768w\" sizes=\"auto, (max-width: 1127px) 100vw, 1127px\"><\/p>\n<p>This lets you build a feedback loop: optimize \u2192 publish \u2192 measure \u2192 iterate.<\/p>\n<p>The same kind of methodology that works for tracking organic rankings also applies to AI citation tracking.<\/p>\n<h3><a id=\"post-200818-_ima4hf39vn2i\"><\/a>Benchmark against competitors<\/h3>\n<p>Find out which of your competitors is being consistently cited by AI for queries you care about, then analyze the structure and content of their most-cited pages.<\/p>\n<p>Just add a <code>Your brand: Not mentioned<\/code> and <code>Your brand: Found but not cited<\/code> filter to an AI Responses or Cited Pages report in Brand&nbsp;Radar.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" width=\"1083\" height=\"892\" class=\"wp-image-200898\" src=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-40-1.png\" srcset=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-40-1.png 1083w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-40-1-516x425.png 516w, https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-40-1-768x633.png 768w\" sizes=\"auto, (max-width: 1083px) 100vw, 1083px\"><\/p>\n<p>This will show you the topics and third-party discussions your brand tends to be left out&nbsp;of.<\/p>\n<p>Then it\u2019s just a case of reverse-engineering your competitors\u2019 moves to close the&nbsp;gap.<\/p>\n<div class=\"post-nav-link clearfix\" id=\"section1\"><a class=\"subhead-anchor\" data-tip=\"tooltip__copielink\" rel=\"#section1\"><svg width=\"19\" height=\"19\" viewBox=\"0 0 14 14\" style><g fill=\"none\" fill-rule=\"evenodd\"><path d=\"M0 0h14v14H0z\" \/><path d=\"M7.45 9.887l-1.62 1.621c-.92.92-2.418.92-3.338 0a2.364 2.364 0 0 1 0-3.339l1.62-1.62-1.273-1.272-1.62 1.62a4.161 4.161 0 1 0 5.885 5.884l1.62-1.62L7.45 9.886zM5.527 5.135L7.17 3.492c.92-.92 2.418-.92 3.339 0 .92.92.92 2.418 0 3.339L8.866 8.473l1.272 1.273 1.644-1.643A4.161 4.161 0 1 0 5.897 2.22L4.254 3.863l1.272 1.272zm-.66 3.998a.749.749 0 0 1 0-1.06l2.208-2.206a.749.749 0 1 1 1.06 1.06L5.928 9.133a.75.75 0 0 1-1.061 0z\" style \/><\/g><\/svg><\/a><div class=\"link-text\" data-anchor=\"Final thoughts\" data-section=\"final-thoughts\">\n<h2>Final thoughts<\/h2>\n<\/div><\/div>\n<p>RAG is the bridge between search and AI. It follows predictable rules, promoting pages it can access, fetch quickly, and topic-match directly to give the best possible answer.<\/p>\n<p>Track your AI visibility with <a href=\"https:\/\/ahrefs.com\/brand-radar\">Ahrefs Brand Radar<\/a> to see whether your content is showing up across ChatGPT, Perplexity, Google AI Overviews, and the other tools your audience actually uses.<\/p>\n<p>Got questions? Ping me on <a href=\"https:\/\/uk.linkedin.com\/in\/louise-linehan\">LinkedIn<\/a>.<\/p>\n<p>&nbsp;<\/p>\n","protected":false},"excerpt":{"rendered":"<p>You need to understand RAG because it\u2019s one of the ways ChatGPT, AI Mode and other AI search engines choose which pages get included in its answer. This guide explains how RAG works (in plain English), what makes content more<span class=\"ellipsis\">\u2026<\/span><\/p>\n<div class=\"read-more\">Read more \u203a<\/div>\n<p><!-- end of .read-more --><\/p>\n","protected":false},"author":197,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"wp_typography_post_enhancements_disabled":false,"footnotes":""},"categories":[469],"tags":[],"coauthors":[464],"class_list":["post-200818","post","type-post","status-publish","format-standard","hentry","category-ai-search","odd"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v28.0 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>Retrieval Augmented Generation (RAG) Explained: How AI Decides Which Pages to Search &amp; Cite<\/title>\n<meta name=\"description\" content=\"How retrieval augmented generation works\u2014and how to optimize your content so AI search engines like ChatGPT actually retrieve and cite it.\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/ahrefs.com\/blog\/retrieval-augmented-generation\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Retrieval Augmented Generation (RAG) Explained: How AI Decides Which Pages to Search &amp; Cite\" \/>\n<meta property=\"og:description\" content=\"How retrieval augmented generation works\u2014and how to optimize your content so AI search engines like ChatGPT actually retrieve and cite it.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/ahrefs.com\/blog\/retrieval-augmented-generation\/\" \/>\n<meta property=\"og:site_name\" content=\"SEO Blog by Ahrefs\" \/>\n<meta property=\"article:publisher\" content=\"https:\/\/www.facebook.com\/Ahrefs\/\" \/>\n<meta property=\"article:published_time\" content=\"2026-07-09T08:56:23+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2026-07-09T10:12:51+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-1-1.png\" \/>\n\t<meta property=\"og:image:width\" content=\"700\" \/>\n\t<meta property=\"og:image:height\" content=\"500\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/png\" \/>\n<meta name=\"author\" content=\"Louise Linehan\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:creator\" content=\"@ahrefs\" \/>\n<meta name=\"twitter:site\" content=\"@ahrefs\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/retrieval-augmented-generation\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/retrieval-augmented-generation\\\/\"},\"author\":{\"name\":\"Louise Linehan\",\"@id\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/#\\\/schema\\\/person\\\/444b3643c35b16b94b763446c5562388\"},\"headline\":\"Retrieval Augmented Generation (RAG) Explained: How AI Decides Which Pages to Search &amp;&nbsp;Cite\",\"datePublished\":\"2026-07-09T08:56:23+00:00\",\"dateModified\":\"2026-07-09T10:12:51+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/retrieval-augmented-generation\\\/\"},\"wordCount\":4901,\"publisher\":{\"@id\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/retrieval-augmented-generation\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/wp-content\\\/uploads\\\/2026\\\/07\\\/retrieval-augmented-generation-rag-explained-how-by-louise-linehan-ai-search.jpg\",\"articleSection\":[\"AI Search\"],\"inLanguage\":\"en-US\"},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/retrieval-augmented-generation\\\/\",\"url\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/retrieval-augmented-generation\\\/\",\"name\":\"Retrieval Augmented Generation (RAG) Explained: How AI Decides Which Pages to Search & Cite\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/retrieval-augmented-generation\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/retrieval-augmented-generation\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/wp-content\\\/uploads\\\/2026\\\/07\\\/word-image-200818-1-1.png\",\"datePublished\":\"2026-07-09T08:56:23+00:00\",\"dateModified\":\"2026-07-09T10:12:51+00:00\",\"description\":\"How retrieval augmented generation works\u2014and how to optimize your content so AI search engines like ChatGPT actually retrieve and cite it.\",\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/ahrefs.com\\\/blog\\\/retrieval-augmented-generation\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/retrieval-augmented-generation\\\/#primaryimage\",\"url\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/wp-content\\\/uploads\\\/2026\\\/07\\\/word-image-200818-1-1.png\",\"contentUrl\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/wp-content\\\/uploads\\\/2026\\\/07\\\/word-image-200818-1-1.png\",\"width\":700,\"height\":500},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/#website\",\"url\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/\",\"name\":\"SEO Blog by Ahrefs\",\"description\":\"Link Building Strategies &amp; SEO Tips\",\"publisher\":{\"@id\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/#organization\",\"name\":\"Ahrefs\",\"url\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/wp-content\\\/uploads\\\/2023\\\/06\\\/ahrefs-logo.png\",\"contentUrl\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/wp-content\\\/uploads\\\/2023\\\/06\\\/ahrefs-logo.png\",\"width\":2048,\"height\":768,\"caption\":\"Ahrefs\"},\"image\":{\"@id\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/#\\\/schema\\\/logo\\\/image\\\/\"},\"sameAs\":[\"https:\\\/\\\/www.facebook.com\\\/Ahrefs\\\/\",\"https:\\\/\\\/x.com\\\/ahrefs\",\"https:\\\/\\\/www.linkedin.com\\\/company\\\/ahrefs\\\/\",\"https:\\\/\\\/www.youtube.com\\\/c\\\/ahrefscom\"]},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/#\\\/schema\\\/person\\\/444b3643c35b16b94b763446c5562388\",\"name\":\"Louise Linehan\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/wp-content\\\/uploads\\\/2024\\\/08\\\/Louise-Linehan.jpg02b05bbed9b25ec9b04e39f0d88f15b0\",\"url\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/wp-content\\\/uploads\\\/2024\\\/08\\\/Louise-Linehan.jpg\",\"contentUrl\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/wp-content\\\/uploads\\\/2024\\\/08\\\/Louise-Linehan.jpg\",\"caption\":\"Louise Linehan\"},\"description\":\"Louise is a Content Marketer at Ahrefs. Over the past ten years, she has held senior content positions at SaaS brands: Pi Datametrics, BuzzSumo, and Cision. By day, she writes about content and SEO; by night, you'll find her playing football or screaming down the mic at karaoke.\",\"sameAs\":[\"https:\\\/\\\/www.linkedin.com\\\/in\\\/louise-linehan\\\/\"],\"url\":\"https:\\\/\\\/ahrefs.com\\\/blog\\\/author\\\/louise-linehan\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Retrieval Augmented Generation (RAG) Explained: How AI Decides Which Pages to Search & Cite","description":"How retrieval augmented generation works\u2014and how to optimize your content so AI search engines like ChatGPT actually retrieve and cite it.","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/ahrefs.com\/blog\/retrieval-augmented-generation\/","og_locale":"en_US","og_type":"article","og_title":"Retrieval Augmented Generation (RAG) Explained: How AI Decides Which Pages to Search & Cite","og_description":"How retrieval augmented generation works\u2014and how to optimize your content so AI search engines like ChatGPT actually retrieve and cite it.","og_url":"https:\/\/ahrefs.com\/blog\/retrieval-augmented-generation\/","og_site_name":"SEO Blog by Ahrefs","article_publisher":"https:\/\/www.facebook.com\/Ahrefs\/","article_published_time":"2026-07-09T08:56:23+00:00","article_modified_time":"2026-07-09T10:12:51+00:00","og_image":[{"width":700,"height":500,"url":"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-1-1.png","type":"image\/png"}],"author":"Louise Linehan","twitter_card":"summary_large_image","twitter_creator":"@ahrefs","twitter_site":"@ahrefs","schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/ahrefs.com\/blog\/retrieval-augmented-generation\/#article","isPartOf":{"@id":"https:\/\/ahrefs.com\/blog\/retrieval-augmented-generation\/"},"author":{"name":"Louise Linehan","@id":"https:\/\/ahrefs.com\/blog\/#\/schema\/person\/444b3643c35b16b94b763446c5562388"},"headline":"Retrieval Augmented Generation (RAG) Explained: How AI Decides Which Pages to Search &amp;&nbsp;Cite","datePublished":"2026-07-09T08:56:23+00:00","dateModified":"2026-07-09T10:12:51+00:00","mainEntityOfPage":{"@id":"https:\/\/ahrefs.com\/blog\/retrieval-augmented-generation\/"},"wordCount":4901,"publisher":{"@id":"https:\/\/ahrefs.com\/blog\/#organization"},"image":{"@id":"https:\/\/ahrefs.com\/blog\/retrieval-augmented-generation\/#primaryimage"},"thumbnailUrl":"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/retrieval-augmented-generation-rag-explained-how-by-louise-linehan-ai-search.jpg","articleSection":["AI Search"],"inLanguage":"en-US"},{"@type":"WebPage","@id":"https:\/\/ahrefs.com\/blog\/retrieval-augmented-generation\/","url":"https:\/\/ahrefs.com\/blog\/retrieval-augmented-generation\/","name":"Retrieval Augmented Generation (RAG) Explained: How AI Decides Which Pages to Search & Cite","isPartOf":{"@id":"https:\/\/ahrefs.com\/blog\/#website"},"primaryImageOfPage":{"@id":"https:\/\/ahrefs.com\/blog\/retrieval-augmented-generation\/#primaryimage"},"image":{"@id":"https:\/\/ahrefs.com\/blog\/retrieval-augmented-generation\/#primaryimage"},"thumbnailUrl":"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-1-1.png","datePublished":"2026-07-09T08:56:23+00:00","dateModified":"2026-07-09T10:12:51+00:00","description":"How retrieval augmented generation works\u2014and how to optimize your content so AI search engines like ChatGPT actually retrieve and cite it.","inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/ahrefs.com\/blog\/retrieval-augmented-generation\/"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/ahrefs.com\/blog\/retrieval-augmented-generation\/#primaryimage","url":"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-1-1.png","contentUrl":"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2026\/07\/word-image-200818-1-1.png","width":700,"height":500},{"@type":"WebSite","@id":"https:\/\/ahrefs.com\/blog\/#website","url":"https:\/\/ahrefs.com\/blog\/","name":"SEO Blog by Ahrefs","description":"Link Building Strategies &amp; SEO Tips","publisher":{"@id":"https:\/\/ahrefs.com\/blog\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/ahrefs.com\/blog\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/ahrefs.com\/blog\/#organization","name":"Ahrefs","url":"https:\/\/ahrefs.com\/blog\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/ahrefs.com\/blog\/#\/schema\/logo\/image\/","url":"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2023\/06\/ahrefs-logo.png","contentUrl":"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2023\/06\/ahrefs-logo.png","width":2048,"height":768,"caption":"Ahrefs"},"image":{"@id":"https:\/\/ahrefs.com\/blog\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/www.facebook.com\/Ahrefs\/","https:\/\/x.com\/ahrefs","https:\/\/www.linkedin.com\/company\/ahrefs\/","https:\/\/www.youtube.com\/c\/ahrefscom"]},{"@type":"Person","@id":"https:\/\/ahrefs.com\/blog\/#\/schema\/person\/444b3643c35b16b94b763446c5562388","name":"Louise Linehan","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2024\/08\/Louise-Linehan.jpg02b05bbed9b25ec9b04e39f0d88f15b0","url":"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2024\/08\/Louise-Linehan.jpg","contentUrl":"https:\/\/ahrefs.com\/blog\/wp-content\/uploads\/2024\/08\/Louise-Linehan.jpg","caption":"Louise Linehan"},"description":"Louise is a Content Marketer at Ahrefs. Over the past ten years, she has held senior content positions at SaaS brands: Pi Datametrics, BuzzSumo, and Cision. By day, she writes about content and SEO; by night, you'll find her playing football or screaming down the mic at karaoke.","sameAs":["https:\/\/www.linkedin.com\/in\/louise-linehan\/"],"url":"https:\/\/ahrefs.com\/blog\/author\/louise-linehan\/"}]}},"as_json":null,"as_tables":null,"as_images":null,"json_reviewers":[194],"as_coauthors":[],"as_post_info":null,"as_sticky":null,"as_hreflang":null,"_links":{"self":[{"href":"https:\/\/ahrefs.com\/blog\/wp-json\/wp\/v2\/posts\/200818","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/ahrefs.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/ahrefs.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/ahrefs.com\/blog\/wp-json\/wp\/v2\/users\/197"}],"replies":[{"embeddable":true,"href":"https:\/\/ahrefs.com\/blog\/wp-json\/wp\/v2\/comments?post=200818"}],"version-history":[{"count":55,"href":"https:\/\/ahrefs.com\/blog\/wp-json\/wp\/v2\/posts\/200818\/revisions"}],"predecessor-version":[{"id":201016,"href":"https:\/\/ahrefs.com\/blog\/wp-json\/wp\/v2\/posts\/200818\/revisions\/201016"}],"wp:attachment":[{"href":"https:\/\/ahrefs.com\/blog\/wp-json\/wp\/v2\/media?parent=200818"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/ahrefs.com\/blog\/wp-json\/wp\/v2\/categories?post=200818"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/ahrefs.com\/blog\/wp-json\/wp\/v2\/tags?post=200818"},{"taxonomy":"author","embeddable":true,"href":"https:\/\/ahrefs.com\/blog\/wp-json\/wp\/v2\/coauthors?post=200818"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}