To be cited, a page has to be reachable by each platform's search crawler, indexed, and contain a clear answer with evidence a model can lift: a number, a quote, a named source. Each engine then picks a few pages per answer, so a citation is won page by page and question by question, not by domain authority alone. The method below is what we run for partners, in the order that removes the biggest blocker first.
How do ChatGPT, Perplexity and Google choose what to cite?
All three start with a retrieval step. OpenAI documents OAI-SearchBot as the crawler "used to surface websites in search results in ChatGPT's search features", with a second agent, ChatGPT-User, that visits pages when a user asks for them (OpenAI). Perplexity runs PerplexityBot to "surface and link websites in search results on Perplexity" and states it "is not used to crawl content for AI foundation models", plus Perplexity-User, which "might visit a web page to help provide an accurate answer" (Perplexity). Anthropic's Claude-SearchBot "navigates the web to improve search result quality for users" (Anthropic).
Google's AI Overviews and AI Mode draw on the normal web index. Google says "There are no additional requirements to appear in AI Overviews or AI Mode", and AI Mode works by "issuing a multitude of queries simultaneously on your behalf" and reading across the results (Google AI features, Google). None of the vendors publish a ranking formula for citations, so treat anyone who claims to know the weights with caution.
Step 1: let the search crawlers in
Check your robots.txt before you touch a single page. Many sites blocked GPTBot, ClaudeBot and CCBot in 2023 and 2024, and Cloudflare's 2025 review found those three had the most full-site disallow directives across its network (Cloudflare). Those tokens are training crawlers. A block on GPTBot does not touch OAI-SearchBot, but a blanket rule written as User-agent: * with Disallow: / does, and so does a security product that challenges every unfamiliar bot.
Allow OAI-SearchBot, PerplexityBot and Claude-SearchBot explicitly. Decide separately about GPTBot, ClaudeBot and Google-Extended, which govern training, not citations. For Google, the switch that matters is the Search generative AI control in Search Console, which defaults to Include; setting it to Exclude stops your links from appearing in AI Overviews, AI Mode and Discover's AI features (Search Console Help). The exact file is in our robots.txt guide for AI crawlers.
Step 2: write the answer where a model can lift it
Put the answer in the first two or three sentences under the heading that matches the question. Then support it with something a model can quote: a figure with its source, a short direct quotation, a date. The GEO paper measured exactly this: on its benchmark of 10,000 queries, adding quotations raised visibility by 41%, adding statistics by 33%, and citing sources by 28%, while keyword stuffing lowered it by 9% (Aggarwal et al., KDD 2024).
Google's guidance points the same way from the other side. Its generative AI guide asks for "non-commodity content" with "a unique point of view", tells you not to "just recycle what others on the internet have already said", and says "There's no requirement to break your content into tiny pieces for AI to better understand it" (Google). Original data, a real test, a documented process or a clear opinion beats a rewritten summary of the top ten results every time.
| Do | Skip |
|---|---|
| Answer in the first paragraph under a question heading | A 200 word introduction before the point |
| One sourced number or quote per section | Adjectives in place of evidence |
| Tables for comparisons, plain sentences elsewhere | Repeating the target phrase to hit a density |
| A visible author, date and last update | Anonymous, undated pages |
Step 3: make your entity unambiguous
Models cite what they can identify. Put Organization markup on your home page with name, url, logo, sameAs links to your official profiles and a contactPoint, which is what Google recommends for the page that describes your company (Google). Give every article an author with author.name and author.url pointing to a real bio page, plus datePublished and dateModified (Google).
Google is explicit that structured data "isn't required for generative AI search" (Google). The point is not a ranking boost, it is consistency: the same company name, address and people on your site, your profiles and in your markup, so a model does not confuse you with a similarly named business. Our structured data for AI search article lists the types worth keeping.
Step 4: earn coverage where the models already look
AI answers cite third party sources heavily. Pew found Wikipedia, YouTube and Reddit supplied 15% of links in Google AI summaries, and government sites 6%, three times their share in standard results (Pew Research Center). Being the cited source on your own site is one route; being described accurately on sites the engines already trust is the other.
That means real editorial coverage, answering questions in the forums your buyers use, and publishing video where a text page would not be found. Google warns against pursuing artificial "mentions" as a shortcut (Google), so the test is simple: would the coverage exist if AI engines did not? Where press is the right route, Qubba Publishings & PR lists placements in 1,679 publications with prices, so you can pick outlets that match your market rather than buy volume.
Step 5: measure citations, not rankings
Open the generative AI performance report in Search Console. It shows impressions from AI Overviews and AI Mode by page, country, device and date, and it has been available to all sites since August 31, 2026; it does not show clicks or queries (Search Console Help, Google). Rising AI impressions on a page tell you it is being selected even when sessions are flat.
For the other engines, build a fixed list of 20 to 50 questions your buyers ask, run them monthly in ChatGPT, Perplexity, Claude and Gemini, and record whether you were cited and which URL was used. Add a referral segment in your analytics for chatgpt.com, perplexity.ai, claude.ai and gemini.google.com. The full scorecard is in what to measure instead of organic traffic.
Frequently asked questions
Do I need to rank first on Google to be cited in AI Overviews?
No.
Google's only stated condition is that the page is indexed and eligible to appear with a snippet.
AI Mode issues many subqueries at once, so a page that answers one specific subquestion well can be cited even when it does not rank at the top for the main phrase.
Specific, well evidenced pages tend to be selected over broad ones.
Does blocking GPTBot stop ChatGPT from citing me?
Not by itself.
OpenAI documents GPTBot as the training crawler and OAI-SearchBot as the crawler that surfaces sites in ChatGPT search, each controlled by its own robots.txt token.
Blocking GPTBot signals that your content should not train models.
Blocking OAI-SearchBot, or blocking all bots with a wildcard rule, is what removes you from search citations.
How many pages does an AI answer cite?
Few.
Pew's analysis of Google AI summaries found 88% cited three or more sources and only 1% cited one, with a median answer of 67 words.
Each citation is therefore a scarce slot won on a specific question.
Pages that state the answer directly and attach evidence are the ones a model can quote in a sentence or two.
Does Google-Extended control AI Overviews?
No.
Google-Extended manages whether crawled content trains and grounds Gemini models, and Google states it "does not impact a site's inclusion in Google Search nor is it used as a ranking signal".
The control for AI Overviews and AI Mode is the Search generative AI setting in Search Console, along with the usual nosnippet and max-snippet directives.