How to get your website cited by ChatGPT, Perplexity and Google AI Overviews

To get cited by ChatGPT, Perplexity, Copilot and Google AI Overviews, let their search crawlers in, get indexed and ranking in Google and Bing, and publish pages that answer the question in the first lines and back it with evidence nobody else has. There is no separate trick. Each engine runs a search, reads a handful of pages, and links the ones it used.

Generative engine optimisation (GEO) is the practice of making your content more likely to be retrieved, quoted and linked in answers written by AI search engines.

This guide covers how each engine finds sources, what to put in robots.txt, what the research says gets cited, and how to measure it. It ends with the checklist we used when we rebuilt our own site.

How do AI search engines choose which websites to cite?

AI search engines run one or more web searches against an index, read the top results, write an answer, and attach links to the pages they drew on. If your page is not crawlable and indexed by the engine's search system, it cannot be cited. If it is indexed but does not rank for the queries the engine generates, it will rarely be picked.

Each engine says this in its own documentation. OpenAI says ChatGPT search sends queries to third-party search providers such as Bing, and that any public website can appear in ChatGPT search. Perplexity says it searches the web when you ask a question and that each answer includes numbered footnotes linking to the original sources. Google says that to appear in AI Overviews or AI Mode, a page must be indexed and eligible to show in Google Search with a snippet, and that there are "no additional technical requirements."

The engines also search more widely than a person would. Google describes query fan-out, where the model runs several related searches at once to answer one question. Its example expands "how to fix a lawn that's full of weeds" into queries about herbicides, chemical-free removal and prevention. A page that answers one of those sub-questions well can be cited even if it would never rank for the original phrase.

That may explain a shift in vendor data. Ahrefs, an SEO tool vendor, found in July 2025 that 76% of pages cited in AI Overviews also ranked in Google's top 10 for the same query, based on 1.9 million citations. Its March 2026 update, across 863,000 keyword results pages and 4 million AI Overview URLs, put the figure at about 38%. Ahrefs notes that it changed its parsing between the two studies and that Google moved AI Overviews to Gemini 3 in January 2026, so the numbers are not directly comparable. Ahrefs offers fan-out as a possible explanation. Google has not said whether fan-out explains the change.

Which crawlers do ChatGPT, Perplexity, Google and Copilot use?

Each engine finds pages through a named crawler and an index. ChatGPT search uses OAI-SearchBot plus third-party providers such as Bing. Perplexity uses PerplexityBot. Google AI Overviews and AI Mode use Googlebot and the Google index. Copilot is a Microsoft product, controlled through Bing's own meta tags and reported in Bing Webmaster Tools.

Engine	Crawler to allow	Index it cites from	How to measure referrals and citations
ChatGPT search	OAI-SearchBot (search). GPTBot is training only. ChatGPT-User fetches pages when a user asks.	OpenAI's own crawl plus third-party search providers such as Bing	GA4 source chatgpt.com (ChatGPT adds utm_source=chatgpt.com) and the GA4 AI Assistant channel
Perplexity	PerplexityBot (search). Perplexity-User fetches pages for a user's question.	Pages crawled by PerplexityBot, plus live fetches by Perplexity-User	GA4 referral source perplexity.ai, server logs
Google AI Overviews and AI Mode	Googlebot. Google-Extended is a robots.txt token for training and grounding.	Google Search index	Search Console generative AI performance report (impressions). Clicks sit inside Google organic in GA4.
Microsoft Copilot and Bing AI summaries	Bingbot (Bing's NOCACHE and NOARCHIVE tags control AI use)	Bing	Bing Webmaster Tools AI Performance report (citations), GA4 AI Assistant channel

A few details from the documentation matter here. OpenAI says sites that opt out of OAI-SearchBot will not be shown in ChatGPT search answers. Perplexity says PerplexityBot is used to surface and link websites in its results and not to collect training data for foundation models. Both companies say user-triggered fetchers (ChatGPT-User and Perplexity-User) may not follow robots.txt, because a person asked for the page. Both say robots.txt changes take around 24 hours to apply.

Google is the exception to the "separate AI crawler" pattern. AI Overviews are built from the normal Google index, so Googlebot is the control. Google-Extended is a robots.txt product token with no user agent of its own, and Google states it "does not impact a site's inclusion in Google Search."

What should my robots.txt say about AI crawlers?

For most B2B sites, allow the search crawlers (Googlebot, Bingbot, OAI-SearchBot, PerplexityBot) and decide separately whether you want your content used for model training. Blocking GPTBot or Google-Extended keeps you out of some training without removing you from ChatGPT search or Google AI Overviews.

A robots.txt that allows search and declines training looks like this:

User-agent: OAI-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: GPTBot
Disallow: /

User-agent: Google-Extended
Disallow: /

User-agent: *
Allow: /

Sitemap: https://example.com/sitemap.xml

Whether to block training is a business decision. For a services firm that wants to be known, being in the training data may help a model mention you unprompted. For a publisher that sells content, it may not. Our own robots.txt allows every crawler and only disallows our API routes.

Two cautions. Google says robots.txt is "not a mechanism for keeping a web page out of Google"; use noindex for that. OpenAI says the same for ChatGPT: to keep a page out of its results, use a noindex meta tag, which means the crawler has to be allowed to read the page.

Bing has its own controls. In 2023, when Copilot was still called Bing Chat, Microsoft said a NOCACHE meta tag limits its AI answers to the URL, title and snippet, and NOARCHIVE keeps the page out of those answers. Both leave the page in normal Bing results.

Is GEO different from SEO?

Mostly no. Google's guide to optimising for generative AI features, published in May 2026, says AEO and GEO describe work that is still SEO from Google's point of view. ChatGPT search uses providers such as Bing, Copilot is reported through Bing Webmaster Tools, and Perplexity runs its own crawler, so the foundations (crawlability, indexing, relevance, authority) carry across.

What changes is the unit of success. In classic search you win a position and hope for the click. In AI search you win a citation inside an answer, and the click is optional. Pew Research Center tracked 900 US adults in March 2025 and found they clicked a traditional result on 8% of Google visits with an AI summary, against 15% without one. Only 1% of those visits included a click on a link inside the summary. That is US data from early in the rollout, but the point holds: being the cited source matters more when fewer people scroll.

Visibility is also spread across engines. Ahrefs found in June 2025 that only 7 of the 50 most-mentioned websites were shared across ChatGPT, Perplexity and Google AI Overviews. Ranking well in one does not guarantee the others.

What does the research say gets content cited?

The main peer-reviewed study is the GEO paper by Pranjal Aggarwal and colleagues, published at KDD 2024. It tested rewriting methods on a benchmark of queries and found that adding quotations, adding statistics and citing sources raised a page's visibility in generative answers, while keyword stuffing lowered it.

In the paper's results, quotation addition did best, improving its main visibility measure by about 41% over the unmodified page. Statistics addition, fluency improvements and citing sources each improved it by roughly 27 to 31% on the same measure (our calculation from the paper's Table 1). On that measure, keyword stuffing scored below the unmodified baseline. The authors also report that results varied by domain, so no single method works everywhere.

Google's guidance points the same way from a different angle. Its AI optimisation guide calls out "non-commodity content" as the factor most likely to affect presence in generative AI search over time. Commodity content restates common knowledge, which an AI model can already produce. Non-commodity content carries first-hand experience, original data or an expert view. Google also warns that creating many pages to match fan-out variations breaches its scaled content abuse policy.

Bing's February 2026 post on AI Performance gives similar advice for Copilot: keep content fresh, use clear headings, tables and FAQ sections, and support claims with examples, data and cited sources.

Treat single-site case studies with care. Search Influence, a marketing agency, published 91 days of its own Copilot data in February 2026: 19,717 citations, with one comparison page taking 69% of them and daily citations falling 97% from early December levels. It is one site, one engine and a preview report, but it shows how concentrated and volatile citations can be.

How should I structure a page so AI engines can quote it?

Put a complete answer at the top of the page and under every heading, in two to four sentences that make sense on their own. Use question headings that match how people search, tables for comparisons, and a named, linked source for every figure. Mark up the author, dates and publisher so engines can see who stands behind the claims.

Google says there is no need to break content into tiny chunks or write in a special style for AI, and there is no ideal length. We agree, and still find the answer-first pattern useful because it serves people who skim as much as it serves machines. In our own articles, each ## section opens with a self-contained answer, then goes deeper.

On structured data, Google says no special schema.org markup is needed for its generative AI features, and that structured data should match the visible text. It still recommends JSON-LD for giving "explicit clues about the meaning of a page." Article markup lets you state the author, the date published and, through schema.org's citation property, the works an article references.

One change matters if you rely on FAQ markup. Google stopped showing FAQ rich results on 7 May 2026. FAQPage markup no longer earns a special result in Google, but a visible FAQ section is still the kind of structure Bing recommends for Copilot.

Should I publish an llms.txt file?

llms.txt is a proposal, published by Jeremy Howard in September 2024, for a Markdown file at your site root that gives AI agents a short summary of the site and links to its key pages. Google says it does not use llms.txt in Search, and we have not found a primary source from OpenAI, Perplexity or Microsoft confirming that their search products use it.

The llmstxt.org page describes it as "a proposal to standardise" this file. Google added a note to its AI optimisation guide in June 2026 saying these files are not needed for Google Search, will not help or hurt rankings, and are fine to keep for other systems that use them.

We publish one because it takes minutes and costs nothing. On a Next.js site it can be generated from the same content that builds the pages, so it stays in step with each deploy. Do not expect it to move citations on its own.

How do I tell search engines about new pages quickly?

Submit an XML sitemap to Google Search Console and Bing Webmaster Tools, and ping IndexNow whenever you publish or update a page. IndexNow is a protocol that notifies participating engines, including Bing, of changed URLs. Because ChatGPT search draws on providers such as Bing, faster Bing indexing can help there too.

The IndexNow documentation allows up to 10,000 URLs in one POST, verified by a key file hosted at your site root. Its FAQ lists Amazon, Bing, Naver, Seznam.cz, Yandex and Yep as participants, with a shared endpoint that passes submissions to all of them. Google is not listed. Bing's February 2026 AI Performance post recommends IndexNow to keep content current in AI answers.

Freshness matters because AI answers quote whatever version the index holds. If you change a price, a date or a claim, the old version can keep circulating until the page is recrawled.

How do I measure traffic and citations from AI search?

Use four sources together. GA4 shows visits from ChatGPT, Perplexity and Copilot. Bing Webmaster Tools shows Copilot citations. Search Console shows impressions in AI Overviews and AI Mode. Server logs show which AI crawlers are reading which pages.

Google Analytics 4. ChatGPT adds utm_source=chatgpt.com to the links it sends. Since May 2026, GA4 also has an AI Assistant default channel for referrers it recognises, naming ChatGPT, Gemini, Deepseek, Copilot and Grok as examples. Google's AI Overviews and AI Mode are excluded from that channel; those clicks count as Google organic.

Bing Webmaster Tools. The AI Performance report, in public preview since 10 February 2026, shows total citations across Copilot, Bing AI summaries and some partner integrations, the pages cited, and a sample of grounding queries (the searches the AI ran when it retrieved your content). Bing notes that citation counts do not indicate ranking or importance.

Google Search Console. The generative AI performance report shows impressions from AI Overviews and AI Mode by page, device and country. Google says it reached all websites worldwide by 31 August 2026. It is built from the Web search type in the main Performance report, so it carries the same limits.

Server logs. Analytics only sees visits. Logs show crawls. Filtering for OAI-SearchBot, PerplexityBot, GPTBot, Bingbot and Googlebot tells you whether the engines are reading your new pages at all, which is the first thing to check when nothing is being cited.

What happened when we rebuilt our own site for AI search?

We relaunched thirdeyeopen.io on 17 October 2026 as a Next.js site, with the measures in this guide built in. In the first week, AI referrals were a small share of visits but the most engaged channel we had. It is a small sample. We treat it as a signal to keep watching.

The numbers, from our own Google Analytics and Bing Webmaster Tools in October 2026:

• 3 of 33 sessions in the first week arrived from ChatGPT or Perplexity.
• All three counted as engaged sessions. Together they averaged just under five minutes of engagement, against 39 seconds for direct visits. The two from ChatGPT averaged about seven minutes.
• Bing Webmaster Tools showed 13 Copilot citations for the domain over the previous three months.

Thirty-three sessions proves nothing statistically. What it suggests is that people who click through from an AI answer have usually read a summary already and arrive looking for detail, which fits with long, sourced pages.

Here is what we built into the site:

• robots.txt allows all crawlers, AI search and training included, and blocks only our API routes. The sitemap is declared in the same file.
• XML sitemap generated from the content files, with each article's own publish or update date.
• IndexNow on every deploy. Our build pipeline reads the live sitemap after each release and posts every URL to the IndexNow endpoint, so Bing and the other participants hear about changes within minutes. The step never fails the build.
• llms.txt generated from the same content: a short description of the company, then every service, case study and article with a one-line summary.
• Structured data. One Organization and Person for the whole site, with the Companies House number and linked profiles. Each article carries Article markup with the author, dates and a citation list built from its sources, plus BreadcrumbList and, where there are FAQs, FAQPage.
• "In short" summaries at the top of every article, written answer-first.
• A visible sources list at the foot of every article, linking each primary source.

None of this is exotic. Most of it is the same discipline we apply to moving a site to a headless CMS without losing rankings, pointed at a wider set of engines.

What is the TEO Cite-Ready checklist?

Cite-Ready is the seven-step checklist we use to get a site into AI answers: open the doors, get indexed, answer first, add evidence, say something original, structure it, and measure. Work through it in order, because each step depends on the one before.

1. Open. Allow Googlebot, Bingbot, OAI-SearchBot and PerplexityBot in robots.txt. Decide on GPTBot and Google-Extended as a separate, deliberate choice. Check your CDN or firewall is not blocking these bots by default.
2. Indexed. Verify the site in Google Search Console and Bing Webmaster Tools, submit your sitemap to both, and ping IndexNow on publish. If Bing has not indexed you, ChatGPT and Copilot are unlikely to cite you.
3. Answer first. Open the page with a two or three sentence answer to the headline question. Start each section with a self-contained answer under a question heading.
4. Evidence. Name and link a primary source for every figure, add short quotes where they help, and include a sources list. This is what the GEO study found most effective.
5. Original. Add something only you can say: your own data, a decision you made and why, a worked example from a project. Google calls this non-commodity content.
6. Structured. Use tables for comparisons and numbered steps for processes. Add Article and Organization markup that matches the visible page.
7. Measured. Watch the GA4 AI Assistant channel, Bing AI Performance and the Search Console generative AI report monthly. Check your logs for crawler visits after each publish.

If you want an outside view of where your site stands on the technical side, our Platform Health Check covers performance, architecture and delivery in one fixed-fee audit. For the content and experience side, what a UX audit should tell you explains how we review pages people actually use.

What should I not waste time on?

Skip anything that promises to game the engines. Keyword stuffing lowered visibility in the GEO study. Google says separate pages for every fan-out variation breach its scaled content abuse policy, that seeking inauthentic brand mentions is less helpful than it seems, and that no third-party tool has access to its internal ranking or AI systems.

Also be wary of AI-specific markup. Google says it does not use special machine-readable files or markup for AI features. If a vendor sells you an "AI schema" package, ask which engine has documented using it.

Finally, avoid publishing AI-written pages at volume. Google's guidance on generative AI content says AI-assisted content must still meet its spam policies and should be fact-checked before publishing. A page that repeats what a model could already say gives that model no reason to cite you.

Where should a B2B team start?

Start with access and indexing, because nothing else works without them. Check robots.txt, verify the site in Bing Webmaster Tools, and confirm Bing has indexed your key pages. Then rewrite your three most important service or insight pages answer-first, with sources and one piece of original evidence each.

For most UK B2B firms, the pages most worth citing are the ones that answer buying questions: what something costs, how long it takes, how to choose between options, and what the regulations say. Those are the questions buyers now put to ChatGPT before they ever reach a search results page.

We design and build sites with this built in from the first release, through our platform build and product design teams. If you want help getting your site cited, talk to us.