# How Do AI Answer Engines Choose What to Cite? > How ChatGPT, Perplexity, Gemini, Claude, Copilot, Google AI Overviews and AI Mode decide which brands to name and which pages to cite. Covers training data versus live search, search expansions (query fanouts), sources versus citations, the kinds of pages engines trust, crawler access with robots.txt and llms.txt, and how Revnu shows each part. Canonical: https://revnu.com/docs/how-ai-engines-choose-sources Last updated: 2026-10-18 An AI answer engine builds each answer from two places: what it learned in training, and what it finds when it searches the web right then. The brands it names and the pages it cites come from whichever pages it read. To be named, you need to be on the pages the engine reads, said clearly enough that it can repeat you. Revnu shows you every part of that chain on the **Visibility** pages: the searches the engines ran, the pages they cited, and which brands those pages named. Then it works on the pages that matter. ## Training data and live search Every engine starts from what it learned during training. That's broad but out of date, and it's how an engine can name you without linking anywhere. For most buying questions, engines also search the web before they answer. Perplexity, Google AI Overviews, Google AI Mode and Copilot search almost every time. ChatGPT, Claude and Gemini search when the question needs current or specific information, which buying questions usually do. When an engine answers from training alone, the answer shows no sources at all. Live search is where you can move fastest. A new page, listing or post can reach answers within weeks, once engines find it. ## Search expansions: one question becomes many searches An engine rarely searches the buyer's exact words. It breaks one question into several narrower searches, reads the results, and writes one answer from all of them. This is called query fanout. Revnu calls these **search expansions**. For example, "invoicing tool for freelancers in the UK" might become: - best invoicing software freelancers UK 2026 - invoicing app that handles VAT - free invoicing tool comparison - invoicing tool reviews freelancers The pages that rank for those sub-searches are the pages the answer is built from. So the sub-searches, not just the original question, are what your content has to cover. **Visibility > Search Expansions** shows them for your tracked prompts: - **Avg Queries / Execution**: how many searches an engine runs per answer, on average. - **Total Search Expansions** and **Distinct Sub-queries**: how many searches were recorded and how many were different. - **Search Expansions by Prompt**: the searches behind each tracked prompt. - **Query Variation Breakdown** and **Word Transformations**: how engines reword your buyers' questions, for example adding a year, "best", "reviews" or "free". Where an engine shows the searches it ran, Revnu records them. For the others, it records the search plan the same engine writes for that question. > **Tip:** Read **Word Transformations** for the words engines add. If they keep adding "for small teams" or "pricing", your pages should answer that too. ## Sources and citations are different - **Sources** are the pages an engine read while writing the answer. - **Citations** are the pages it shows the reader as references, usually as numbered links. An engine can read twenty pages and cite five. It can also name your business because a cited directory or review mentions you, while linking to that directory instead of your site. That's why you can be mentioned without being cited. **Visibility > AI Citations** shows which sites and pages the answers cite: - **Citation Share**: links to your own pages as a share of every citation. - **Most-cited Sources** and the **Top Citation Domains** table: the sites engines lean on for your questions. Click a domain to see the exact pages. You can search the table and export it as CSV. - **Citation Categories**: what kind of site each one is. | Category | What it covers | What usually works | |---|---|---| | **Earned Media** | News and tech publications, and software review sites such as G2 and Capterra | Get reviewed, get listed, get included in their roundups | | **Social** | Reddit, YouTube, LinkedIn, X, Quora, Product Hunt, Trustpilot and other community sites | Take part genuinely in the threads and pages that get cited | | **Institution** | Wikipedia, universities, governments and standards bodies | Make sure the facts about your category are current and correct | | **PR Wire** | Press release distribution sites | Announce real news | | **Other** | Everything else: vendors' own sites, blogs, smaller directories and "best X" lists | Open the domain to see its pages and decide case by case | Pages on your own domains count as yours. Add every domain you own, including old domains and a separate blog host, under **Your Domains** on the **AI Citations** page. ## What makes a page easy to cite Engines cite pages that answer the question directly and that they can read and trust. In practice: - **The answer comes first.** A clear one or two sentence answer near the top of the page, then the detail. - **Specific facts.** Prices, limits, integrations, who it's for, numbers with sources. Engines repeat specifics. Vague claims get skipped. - **Structure.** Clear headings phrased as questions, short paragraphs, lists and tables, and an FAQ section. - **Formats engines like.** Comparisons ("X vs Y"), alternatives pages, how-to guides, use-case pages and glossaries. "Best X" lists on third-party sites get cited heavily too. - **Fresh.** A current date and current facts. Engines prefer recent pages for anything that changes. - **Readable HTML.** Engines read the HTML your server sends. Content that only appears after JavaScript runs, or sits behind a login or paywall, is invisible to them. Every post Revnu writes is researched, checked against the facts in your Memory and ships with an FAQ section. On the Revnu-hosted blog, each post also carries structured data and your blog serves an llms.txt file. See [how Revnu researches and checks each post](/docs/ai-written-posts). ## Third-party pages decide a lot Engines trust what other sites say about you, often more than what you say about yourself. For a buying question, the cited pages are typically review sites, directories, "best X" lists, comparison articles and Reddit threads. If those pages name your competitors and not you, the answer will too. That's why Revnu works off your site as well as on it: - It submits you to the free directories and "submit your tool" forms the answers cite. See [free directory listings](/docs/directory-listings). - For lists you get onto by asking a person, it gives you the link and a ready pitch. - It drafts Reddit posts for you to share, and on accounts where Reddit replies are switched on, it drafts replies for the threads AI answers cite, each waiting for your approval. See [social posts](/docs/social-posts). ## Let AI crawlers read your site AI companies run three kinds of crawler, and each has its own name in robots.txt: | Kind | What it does | Examples | |---|---|---| | **Search crawlers** | Index pages so the engine can find and cite them in answers | OAI-SearchBot, PerplexityBot, Claude-SearchBot | | **User fetchers** | Fetch a page when a user's question needs it | ChatGPT-User, Claude-User, Perplexity-User | | **Training crawlers** | Collect pages to train future models | GPTBot, ClaudeBot, Google-Extended, CCBot | Blocking a search crawler or user fetcher in robots.txt can keep you out of that engine's answers. Blocking training crawlers is a choice about future models and doesn't stop you being cited from live search. Google AI Overviews and AI Mode use Google's normal search index, so anything that keeps you out of Google search keeps you out of those too. **llms.txt** is a short markdown file at the root of your site that tells AI tools what your site is and which pages matter. It's quick to add and helps tools that read it. Ask Revnu to check what your server actually sends: > Audit my site for AI search. Check robots.txt, my sitemap, llms.txt and whether my pages have real content without JavaScript. [Try this in Revnu](/sign-up?prompt=Audit%20my%20site%20for%20AI%20search.%20Check%20robots.txt%2C%20my%20sitemap%2C%20llms.txt%20and%20whether%20my%20pages%20have%20real%20content%20without%20JavaScript.) The audit puts a fix list in **Review**, ordered by impact. See [Get a site audit](/docs/site-audit). ## Answers change from run to run Ask the same engine the same question twice and you can get a different list of brands. That's normal. It's why Revnu measures every tracked prompt repeatedly and reports shares over a window, not one answer. Judge a change over several measurements on the same prompts and engines. ## Next step [AI Visibility: What the Numbers Mean](/docs/ai-visibility) ## Frequently asked questions ### Why does ChatGPT recommend my competitor and not me? Because the pages ChatGPT read for that question name your competitor and not you, or say more about them. Open **Visibility > AI Citations** to see which pages those are, then click **Investigate in Chat** on the gap in **Improve Visibility** and Revnu proposes the next step. ### What is a query fanout? It's when an AI engine turns one question into several narrower web searches before it answers. Revnu calls them search expansions and shows them on **Visibility > Search Expansions**. ### What's the difference between a source and a citation? A source is any page the engine read. A citation is a page it shows the reader as a reference. Engines read more pages than they cite. ### Can I be mentioned in an AI answer without being cited? Yes. An engine often names you because a review site or list it cited mentions you, and links to that page instead of yours. Revnu counts mentions and citations separately. ### Should I block GPTBot in robots.txt? That's your call. GPTBot collects pages for training. Blocking it doesn't stop ChatGPT citing you from live search, which uses OAI-SearchBot and ChatGPT-User. Keep the search crawlers and user fetchers allowed if you want to be cited. ### Does llms.txt help with AI visibility? It helps the AI tools that read it understand your site quickly. It's simple to add, and the Revnu-hosted blog serves one for you. Revnu's site audit checks that yours is present and current. ### Can AI engines read pages built with JavaScript? Many can't. They read the HTML your server sends. If your content only appears after JavaScript runs, engines see an empty page. Revnu's site audit checks this for your real pages. ### Do Google AI Overviews use the same sources as Google search? They draw on Google's search index, so ranking and being indexed in Google matters for them. Connect [Search Console](/docs/search-console) so Revnu can see how Google treats your pages.