Make the page eligible for Google Search with a snippet, answer the buyer’s question plainly and early, and add something first-hand that other pages do not have. Google says there is no special markup, file or separate optimisation for AI Overviews. Pages are pulled from the same index, so ordinary search quality decides it.
That answer disappoints people looking for a trick, and it is meant to. Much of what is sold as AI Overview optimisation is either standard good practice under a new name or a claim nobody can check. Google has published its own guidance, and it is short and specific.
This note summarises what Google says, turns it into the page formats that tend to work well as supporting links, lists what does not matter, and ends with a checklist you can run against any page, plus how to measure the result in Search Console.
What does Google say about getting into AI Overviews?
Google’s own guidance is short: no special requirements, no special markup.
Google’s Search Central page on AI features states that there are no additional requirements to appear in AI Overviews or AI Mode, and no other special optimisations necessary. To be shown as a supporting link, a page needs to be indexed and eligible to be shown in Google Search with a snippet. It also says you do not need to create new machine-readable files, AI text files or markup to appear in these features.
Google describes a technique it calls query fan-out: the AI feature may issue several related searches across subtopics to build its answer. That is why an overview can link to pages that would not rank for the exact words the searcher typed. A page that answers one subtopic clearly can be pulled in for a broader question.
Google’s separate guide to optimising for generative AI search adds the content side. It favours first-hand experience and what it calls non-commodity content, pages with a unique expert or experienced take, over pages that restate what is already available elsewhere. It says there is no requirement to break content into tiny pieces, no need to write in a special way for AI, and that structured data is not required for generative AI search, though it remains useful for rich results.
It is worth reading both pages yourself, because they are the only statements on this topic that come from the system itself. Google also notes in the guide that third-party tools do not have access to its internal ranking or AI systems.
Which page formats tend to work as supporting links?
Formats that answer one question well, and say something the others do not.
Google does not publish a list of preferred formats, so treat what follows as practical reasoning from its guidance rather than a rule. An overview is built from pages that answer parts of a question. A page is easier to use for that when its answer is stated plainly, in text, near the top, and when it adds something the other candidate pages lack.
Five formats fit that description well for a B2B company.
- The direct answer page: one buyer question as the heading, a two or three sentence answer underneath, then the detail. The first paragraph can stand on its own.
- The comparison: two or more options side by side, in a real HTML table, with the criteria a buyer would use. Fair to the options you are not.
- The how-to with steps: a numbered process, each step one action, written by someone who has done it.
- The first-hand case or data: what you did, what you measured, what you found, with the date and the sample size. This is the non-commodity content Google’s guide describes.
- The definition: what a term means, in one sentence, followed by what it includes and excludes. Useful for questions buyers ask early.
Notice what these have in common. None is written for AI. Each is written for a buyer who wants the answer quickly, and each makes the answer easy to find on the page. That is the same thing.
None is written for AI. Each is written for a buyer who wants the answer quickly.
What does not matter, according to Google?
Several popular tactics are explicitly not required.
Based on Google’s published guidance, you do not need:
- An llms.txt file or any other AI-specific text file.
- Special schema.org markup for AI Overviews. Structured data remains useful for rich results, but it is not an AI Overview requirement.
- Content chopped into small chunks for AI to parse.
- A separate version of your content written for AI.
None of these is harmful in itself. The risk is spending your limited time on them instead of on the page’s actual answer. If someone sells you a tactic that Google says is not needed, ask them what evidence they have that it changed anything, measured on a site like yours.
What should you check on every page?
Eligibility first, then the answer, then what makes it yours.
| Check | How to verify | Why it matters |
|---|---|---|
| Site is included in Search generative AI features | Search Console, Settings, Search generative AI (Include is the default) | Excluded sites do not appear in AI Overviews or AI Mode |
| Page is indexed | Search Console, URL Inspection | Only indexed pages can be supporting links |
| Page can show a snippet | No nosnippet, restrictive max-snippet or noindex on the page; no data-nosnippet on the answer text | Google names these as the controls that limit appearance in AI features |
| Googlebot can reach it | robots.txt, firewall and CDN rules; URL Inspection live test | A blocked page cannot be crawled or indexed |
| The answer is in the first paragraph | Read the first 60 words out of context | A self-contained answer is easy to use for a subtopic |
| Key facts are in text | Prices, steps and comparisons are not only in images or scripts | Readable text is what gets understood |
| There is something first-hand | A case, a measurement, a dated example, an expert view | Google’s guide favours non-commodity content |
| Comparisons use a real table | An HTML table, not a screenshot | Structured information is easier to read and reuse |
| Headings match buyer questions | Compare headings to the questions in your sales calls | Helps the page match the subtopics a fan-out search covers |
| Images and video are useful and described | Alt text, captions, standard image and video practice | Google suggests them as further ways to appear |
| Page is linked internally | At least one link from a crawled, relevant page | Helps discovery and shows the page matters |
Run the checklist on your five most important buyer-question pages first. Most of the fixes are small, and the first three rows catch the pages that could never have appeared in the first place.
How do you find which of your queries show AI Overviews?
Look at the results page for your buyer questions, and write down what you see.
Not every search shows an AI Overview, and which ones do can change. Take the buyer questions you already use for AI assistant checks, as described in Does ChatGPT recommend my company? A one-hour check you can run today, and search each in Google, signed out. Record whether an overview appears, which pages it links to, and whether you are among them.
The linked pages are the ones Google considered useful for that question today. Read them. They show what a supporting link looks like in your niche: which formats, which depth, which sites. Our sibling note, How to get your business recommended by ChatGPT: find the sources it trusts in your niche, then earn a place there, applies the same approach to ChatGPT, and the source map in it works for Google too.
Results vary by location, device and time, so treat one search as a snapshot. Run the same questions monthly and compare.
How do you measure whether it worked?
Search Console now reports AI feature impressions separately.
Google added generative AI performance reports to Search Console in 2026. The Search version shows impressions from generative AI features such as AI Overviews and AI Mode, and lets you group them by page, country and date. Use it to see which pages are being shown in AI features and how that changes after you edit them. Clicks from AI features also count in the standard Performance report under the Web search type, combined with other search traffic.
Impressions tell you a page was shown. They do not tell you whether it mattered. The question worth answering is whether the people who arrived from search went on to book a call or buy, and that sits in your own analytics and CRM, not in Search Console.
There is no AI Overview trick. There is a page that answers the question better than the others, and a way to check that Google agrees.
See the use case for b2B health and life-science suppliers: How does a supplier to hospitals, clinics or biotech get named when a clinical or quality lead asks an assistant what to use?
See the check page for Gemini: Does Gemini name you when buyers ask?
Questions people ask next
Do I need special schema markup for AI Overviews?
No. Google says no special schema.org markup is needed for its AI features. Structured data can still help with rich results in ordinary search.
Can I submit my page to AI Overviews?
No. Pages are drawn from Google’s normal index. A page needs to be indexed and eligible to show with a snippet, and your site needs to be included in Search generative AI features in Search Console, which it is by default.
Does llms.txt help with Google AI Overviews?
Google says you do not need AI text files to appear in its AI features, and that such files neither help nor harm visibility in Google Search.
How do I track AI Overview traffic?
Use the generative AI performance report in Search Console for impressions by page, country and date. Clicks are also included in the standard Performance report under the Web search type.
How can I stop my content appearing in AI Overviews?
Google lists the controls: nosnippet, data-nosnippet, max-snippet and noindex on pages, and a site-level Search generative AI setting in Search Console.
Can Throughline track whether my pages appear in Google AI Overviews?
Yes, for the questions your buyers ask. The Lab records, with dates, whether an overview appears and which pages it links, and Throughline reads Search Console and your CRM so you can see what those visitors did next.