How to get cited in Google AI Overviews (and ChatGPT)

To get cited in Google AI Overviews, a page first has to be indexed and eligible to show a snippet in normal Search. Google says there are no other requirements: no special schema, no llms.txt, no "chunking" and no separate way of writing for AI. Beyond that, nobody controls which sources get cited. What you can do is make pages worth citing (original facts, clear answers, a known author or brand), keep them technically clean, and measure with the new Generative AI report in Search Console. That is also most of what people call generative engine optimization (GEO), and the same habits help in ChatGPT search and Copilot.

#What AI Overviews are

An AI Overview is a generated summary at the top of some Google results, with links to the pages it drew from. Google shows it only when its systems judge it adds something to the classic results, so many queries never get one. AI Mode is the longer, conversational version for questions that need comparison or exploration; it also shows supporting links.

Both can use what Google calls query fan-out: running several related searches across subtopics to build one answer. In practice a page can be cited for a sub-question of the query, not only for the exact words someone typed. A section that cleanly answers "how much does X cost" can be cited in an overview for "is X worth it".

#What Google documents, and what is speculation

Most GEO advice online mixes Google's own statements with one agency's observations. Google has two short pages on this: AI features and your website and a longer guide, Optimizing your website for generative AI features on Google Search, which includes a "mythbusting" list. Sort what you read into three buckets:

Google also warns to be wary of third-party tools that promise ranking success or claim access to internal Google metrics: no third-party tool has access to its ranking or AI systems. That applies to any tool that reports AI citations, Serpkin included. Those tools observe results; they don't see inside the system.

Your controls are the normal snippet controls:

• nosnippet, data-nosnippet and max-snippet limit what can be shown, including in AI features. noindex removes the page from Search.
• Google-Extended in robots.txt limits use of your content for training and grounding in some of Google's other AI systems. It is listed separately from the Search controls and does not remove you from AI Overviews.

#Do you need llms.txt?

Not for Google. Google's generative AI guide says you don't need llms.txt or other AI-specific files, that Google Search doesn't use them, and that they won't help or hurt your visibility in Search. Crawling a file doesn't mean it gets special treatment.

Two caveats keep it from being a simple "no". First, it isn't aimed only at Google: some documentation sites publish one so AI tools and coding agents can find pages quickly. Second, Chrome's Lighthouse has an "Agentic browsing" audit category that calls llms.txt an emerging convention. That audit only flags a server error when fetching the file; a missing file (404) is marked not applicable. It is separate from the SEO category. So: skip it if you only care about Google Search. If you add one, make sure the URL returns 200 or a clean 404, never a 500.

#Practical steps to become citable

1. Be indexable and snippet-eligible

Check the live HTML and headers, not your source code:

# Robots headers and meta tags on the live page
curl -sI https://your-site.com/guide | grep -i x-robots-tag
curl -s https://your-site.com/guide | grep -io '<meta[^>]*robots[^>]*>'

Then confirm the page is indexed with URL Inspection in Search Console. If it shows "Discovered" or "Crawled, currently not indexed", fix that first; an unindexed page can't be cited. Use data-nosnippet on elements you don't want quoted (pricing tables you change weekly, legal boilerplate) instead of blocking the whole page. Main content should be in the server HTML: Google's guide points to its JavaScript SEO best practices, and server HTML is the safest bet for crawlers that may not run your scripts.

2. Publish something original

This is the step Google stresses most. Overviews summarize what many pages agree on. Pages that add something (your own benchmark, a price survey, real screenshots, a worked example with numbers, a clear position from experience) give the system a reason to link to you rather than to the tenth page that says the same thing. Date the data and say how you got it. Google's guide also recommends high-quality images and video.

3. Answer first, then explain

Put a direct answer in the first one or two sentences under the H1 and under each H2, then the detail, caveats and examples. Google says you don't need to write in a special way for AI, and this isn't that: it is ordinary good writing that readers skimming a page also want. It just happens to make a clean, quotable statement easy to find.

4. Use headings that match real sub-questions

Because of fan-out, each section can be a candidate on its own. Write H2s and H3s as the questions people actually ask (Search Console queries and "People also ask" are good sources), one topic per section, in real HTML heading tags. You don't need to split one useful page into ten thin ones: Google says its systems understand pages that cover several topics.

5. Make entities clear

Say plainly who you are, what the product is and what each page is about. Use the same names everywhere. An About page, author names on articles, and Organization markup on the homepage with sameAs links to your official profiles help search systems connect the page to a known entity. Structured data isn't required for AI features, so treat it as a description of the visible page, not a citation lever. Only mark up what is visible.

const org = {
  "@context": "https://schema.org",
  "@type": "Organization",
  name: "Acme",
  url: "https://acme.com",
  logo: "https://acme.com/logo.png",
  sameAs: ["https://github.com/acme", "https://www.linkedin.com/company/acme"],
};

6. Keep it fresh, honestly

Update facts, prices, screenshots and examples when they change, and change dateModified only when the content really changed. Bumping a date without new content doesn't make a page more useful.

7. Earn references, don't buy mentions

Google's guide says seeking inauthentic mentions across the web isn't as helpful as it might seem. Real references (someone citing your benchmark, a docs page linking your tool) are still how people and search systems find and trust a page. Original data from step 2 is the most reliable way to earn them.

Prompts for your coding agent

Most of the checks above are code work. Two prompts that work in Claude Code, Codex or Cursor:

For every public route, curl the production URL and report:
X-Robots-Tag header, robots meta, whether the H1 and first paragraph are in the server HTML,
and any data-nosnippet or max-snippet values. Also read robots.txt and list rules for
Googlebot, Google-Extended, OAI-SearchBot, GPTBot and bingbot. Don't change anything.

For /guides/x, rewrite only the first paragraph under the H1 and under each H2
so the first sentence answers the heading's question. Keep every fact, number and caveat
that is already there. Don't add claims, statistics or sources that aren't in the page.
Show the before and after side by side.

The second prompt's last two lines matter. Agents asked to make text "more citable" tend to add confident numbers and drop qualifiers, which makes the page less accurate and less worth citing.

For ChatGPT search and Copilot

OpenAI says sites that opt out of OAI-SearchBot won't be shown in ChatGPT search answers. That crawler is separate from GPTBot, which is about model training. You can allow one and block the other:

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

User-agent: *
Allow: /
Sitemap: https://your-site.com/sitemap.xml

For Copilot and Bing's AI summaries, Microsoft says it respects robots.txt preferences and recommends IndexNow to keep content fresh across search and AI experiences, so don't block bingbot by accident. Also check that your CDN or firewall isn't blocking these crawlers, which robots.txt can't fix.

#How to measure it

There is now first-party data for Google, but it doesn't answer everything. Combine four sources:

• Search Console Generative AI performance report. Google launched it in June 2026 and says it reached all sites by August 31, 2026. It shows how often links to your site appeared in AI Overviews and AI Mode (impressions), broken down by page, country, device and date. It doesn't report clicks; those stay in the main Performance report under the "Web" search type, mixed with classic results, where all links in one overview share a single position. Sites with very few AI impressions may not see the report.
• GA4's AI Assistant channel. Google Analytics now puts visits referred by chat tools such as ChatGPT, Gemini and Copilot in a default "AI Assistant" channel (medium ai-assistant). AI Overviews and AI Mode are excluded: those visits arrive from Google Search and count as organic search. Other analytics tools: filter referrers such as chatgpt.com and perplexity.ai, and expect some AI visits to arrive with no referrer at all.
• Bing Webmaster Tools AI Performance. Shows how often your pages are cited in Copilot, Bing's AI summaries and some partner integrations, per page and over time, plus the "grounding queries" the AI used to retrieve them. It counts citations, not clicks, and Microsoft says the query data is a sample.
• Query checks. None of the reports above tell you which queries show an overview or who else is cited. Search your ten most important queries in a private window and note whether an overview appears, whether you are linked and who else is. Results vary by location, device, account and time, so one check is one observation. Repeat monthly, by hand or with a tool that records AI Overview presence and cited domains.

• The page is indexed (URL Inspection) and has no nosnippet or max-snippet:0.
• Main content is in the server HTML, not loaded by JavaScript after the page renders.
• The page has something original: data, a worked example or first-hand experience.
• The first paragraph under the H1 and each H2 answers its question directly.
• Structured data matches the visible text and validates.
• OAI-SearchBot, bingbot and Googlebot are allowed in robots.txt and at the CDN.
• You check the Generative AI report and the AI Assistant channel monthly, plus key queries by hand.

#How Serpkin tracks AI Overview citations

Search Console tells you how many AI impressions a page got, not which queries showed an overview or who else was cited. Serpkin fills that gap for a set of your important queries (top Search Console queries, mission targets and tracked keywords; how many depends on the plan). Once a month, at no credit cost, it records whether Google showed an AI Overview, whether your site was cited and which other domains were. You can also check up to 10 queries on demand at 1 credit per query (credit allowances are part of the proposed plans).

When an overview appears and you aren't cited, Serpkin opens a mission such as "Get cited in the AI answer for …", or "Competitors are cited, you aren't: …" when competitors you saved are among the sources. The mission names the page to improve and asks for an answer-first opening, sourced claims and Article markup. Your coding agent reads it through serpkin_get_ai_visibility and the mission tools (see SEO with a coding agent). Like any outside tool, Serpkin observes results and has no access to Google's systems. Each result is one observation that varies by location and device, and no citation or traffic change is guaranteed.