How to show up in AI search, and what the evidence says does not work
Google and OpenAI documentation contradicts much of the advice being sold about AI search. What the primary sources say, and the checks worth running.
To appear as a supporting link in Google’s AI Overviews or AI Mode, a page has to be indexed and eligible to show in Google Search with a snippet. Google says there are no extra technical requirements, and that you do not need new machine-readable files, AI text files, markup or special schema.org structured data to appear in these features (Google Search Central).
That is the whole entry test for Google. ChatGPT search is a separate system with its own crawler, and there the documented requirement is that your robots.txt does not block OAI-SearchBot (OpenAI). Much of what is sold on top of those two points is either unsupported or contradicted in writing by the companies running the systems. Below are the recommendations that turn up most often in AI visibility pitches, each checked against the primary source rather than against another agency blog repeating it.
The short version
- Google says a page needs to be indexed and snippet-eligible to be a supporting link in AI Overviews or AI Mode, with no extra technical requirements and no special structured data.
- Google’s own AI optimisation guide says Google Search ignores llms.txt files, and an Ahrefs study of 137,210 domains found 97 percent of valid llms.txt files received no requests in May 2026.
- An Ahrefs study of 863,000 keyword results pages and 4 million AI Overview URLs found 37.1 percent of cited pages also ranked in the organic top ten for the same query.
- OpenAI says sites that block OAI-SearchBot will not be shown in ChatGPT search answers, and that OAI-SearchBot and the training crawler GPTBot can be allowed or blocked independently.
- Google’s
nosnippetandmax-snippetrules also limit how much of a page can be used as direct input for AI Overviews and AI Mode. An old snippet limit can hold a page back in both places. - Since 31 August 2026 Search Console’s generative AI performance reports are available to all sites, showing impressions by page, country, device and date.
What does Google say a page needs to appear in AI Overviews?
A page needs to be indexed and eligible to be shown in Google Search with a snippet. Google’s documentation states that there are no extra technical requirements beyond that (Google Search Central).
Three terms are worth separating first, because much of the confusion lives in the gap between them. AI Overviews is the generated summary that appears above the ordinary results on a Google search. AI Mode is the separate conversational surface inside Google Search. ChatGPT search is OpenAI’s own live retrieval, a different system with different crawlers and different rules. Advice that works on one does not automatically apply to the others, and many pitches treat all three as one thing.
For the two Google surfaces, the best practices Google lists are the ordinary ones. Allow crawling, link internally, keep important content in text and give a good page experience. Make sure any structured data matches what is visible on the page (Google Search Central). Google’s separate guide to generative AI search says the same work applies and names the tactics you can ignore (Google Search Central).
The awkward part for anyone selling an AI package is that this is the same work as ranking a page in normal search. There is no second track.
Which AI search tactics do the primary sources contradict?
Five recommendations turn up again and again. Here is what the documentation says about each.
| The claim | What the primary source says | Verdict | Source |
|---|---|---|---|
| Add llms.txt so AI can read your site | Google Search ignores these files, and they neither help nor harm visibility | Contradicted for Google | Google Search Central |
| Add schema markup to get into AI Overviews | No special schema.org structured data is needed for AI Overviews or AI Mode | Contradicted | Google Search Central |
| Rewrite or chunk pages in an AI-friendly style | You do not need to write in a specific way or break content into tiny pieces for generative AI search | Contradicted | Google Search Central |
| Do not block the AI search crawler | Sites that block OAI-SearchBot will not be shown in ChatGPT search answers | Supported | OpenAI |
| Cite sources and use concrete figures | Adding quotations, statistics and citations were among the strongest methods in the GEO study | Supported, with limits | Aggarwal et al., KDD 2024 |
Read the table as a spending guide rather than a scorecard. The two supported rows are cheap and take an afternoon. The three contradicted rows are what many AI visibility retainers are billing for.
The llms.txt case is the clearest. Google’s guide says it is fine to keep llms.txt files for other services that use them. It says doing so will neither harm nor help your visibility in Google Search, because Google Search ignores them (Google Search Central).
Then somebody measured what happens in practice. Ahrefs looked at 137,210 domains, of which 38,360 had a valid llms.txt file, and found that 97 percent of those files received no requests at all in May 2026 (Ahrefs). Of the requests that did arrive, 96 percent came from bots, AI retrieval bots made up about 1.1 percent, and SEO audit tools made up 21.7 percent. The largest single group checking the file was the SEO industry itself.
Photo: Kampus Production on Pexels.
Do you need to rank in the top ten to be cited in AI Overviews?
No. In Ahrefs’ study, most cited pages did not rank in the top ten. Ahrefs’ March 2026 study covered 863,000 keyword results pages and 4 million AI Overview URLs. It found that 37.1 percent of cited pages also ranked in the organic top ten for the same query, counting standard blue links only (Ahrefs). Another 26.2 percent ranked between 11 and 100, and 36.7 percent did not appear in the top 100 at all.
The same article compares that with about 76 percent in Ahrefs’ July 2025 study. Two cautions before anyone reprioritises a quarter on the change. Ahrefs says it improved its own parsing methodology between the two studies, so the waves are not perfectly comparable. And AI Overviews moved to a newer Gemini model in between, which Ahrefs suggests leans more on the related searches Google runs behind the scenes. Google confirms that AI Overviews and AI Mode may use this “query fan-out” technique, issuing several related searches across subtopics to build a response (Google Search Central).
What it means in practice: a page can be cited without ranking for the exact query, and a page ranking fourth can be left out. Chasing one position is not the whole job any more. Abandoning rankings is not the answer either, because more than a third of citations still came from the top ten.
What does the research say actually moves AI visibility?
The best public evidence is still the GEO paper by Aggarwal and colleagues, presented at KDD 2024 and built on a benchmark of 10,000 queries (Aggarwal et al.). The researchers tested nine ways of rewriting a source and measured how much of the source appeared in generated answers.
The top performers were adding quotations, adding statistics and citing sources. The paper reports its best method improving on the baseline by 41 percent on its main word-count measure, and describes those three methods as reaching a relative improvement of 30 to 40 percent. Keyword stuffing, the old SEO habit, scored below the unmodified baseline on that same measure.
The instruction that falls out of this is plain. Name your sources in the sentence carrying the claim. Use exact figures rather than rounded ones. Quote the people who did the work. That makes a sentence easy to lift into an answer, and it is the same discipline that makes a page worth reading. It is also the core of any serious content strategy rather than a bolt-on.
One limit on this evidence. The paper is from 2024, the systems have changed a great deal since, and it measured generated answers from a test setup rather than business outcomes. Treat it as the best available signal, not as a law.
Which technical settings can quietly remove a site from AI answers?
Two settings can do it: a snippet limit in your robots meta tags, and a robots.txt rule that blocks OpenAI’s search crawler. Both are free to check and take under an hour.
Check your snippet controls. Google’s robots meta tag documentation says nosnippet also prevents content from being used as a direct input for AI Overviews and AI Mode. It says max-snippet also limits how much of the content may be used as a direct input for them (Google Search Central). Plenty of sites set a snippet limit years ago for reasons nobody remembers. If yours has one, it is holding back two things at once, and removing it is a one-line change. This sits in technical SEO rather than content.
Check OAI-SearchBot in robots.txt. OpenAI runs separate crawlers for separate jobs. GPTBot collects content that may be used to train its models. OAI-SearchBot is used to surface websites in ChatGPT’s search features, and OpenAI says sites that block it will not be shown in ChatGPT search answers, though they can still appear as navigational links (OpenAI). Each setting is independent, so a site can allow OAI-SearchBot and still disallow GPTBot. OpenAI says it can take about 24 hours for its search systems to pick up a robots.txt change.
| Setting | Where to look | Pass condition | Source |
|---|---|---|---|
nosnippet robots rule | Page source and HTTP headers of key templates | Absent, unless you deliberately want no snippet | |
max-snippet limit | Page source and HTTP headers | Absent, or a value you can justify | |
| OAI-SearchBot | robots.txt | Not disallowed, if you want to appear in ChatGPT search | OpenAI |
| GPTBot | robots.txt | Your choice. It controls training use, not ChatGPT search | OpenAI |
| Indexing | Search Console page indexing report | Commercial pages indexed |
A firewall or CDN rule that blocks unfamiliar bots can have the same effect as a robots.txt block, so if a page is allowed in robots.txt and still never appears, check that layer too.
Photo: Sora Shimazaki on Pexels.
How do you measure AI visibility in Search Console?
Open the generative AI performance reports in Search Console. Google announced them on 3 June 2026 and added a note that as of 31 August 2026 they had rolled out to all websites worldwide (Google Search Central Blog). They show impressions in generative AI features such as AI Overviews and AI Mode, and in Discover, broken down by page, country, device and date.
What the announcement leaves out matters as much. It lists impressions, not clicks, and no query data. You can see that a page appeared in an AI feature. You cannot see what the appearance was worth, or what somebody typed to trigger it. Google says it is working with site owners on adding more metrics over time, without saying which or when. AI feature traffic is also counted inside the ordinary Performance report under the Web search type (Google Search Central).
Use it as a baseline rather than a scoreboard. Record where your AI impressions sit today, note the date, and compare in ninety days. Anyone quoting you an AI visibility figure more precise than that is using a third-party estimate, and it is fair to ask which one and how it was built.
What none of this can tell you
Four limits, and the last one decides whether any of this is worth your time.
Nobody outside Google and OpenAI knows the selection logic. The studies here, including the ones that support the advice, are correlations measured from the outside.
Appearing in an AI answer is not the same as being chosen. The Search Console report counts impressions, and an impression inside a summary that answers the question completely may send nobody at all.
The measurement window is short. AI features change faster than studies about them can be published, and a finding from March describes a system that has since moved.
And for a typical small business, AI visibility sits downstream of ordinary visibility. If a site is not indexed properly, is not ranking for its commercial terms and has no pages answering real questions, there is nothing for a generated answer to cite. The work that fixes that is the same work as before, and it runs on the same timeline as any SEO work, usually measured in months.
What should you do first if you only have an hour?
Start with the two checks that can remove a site from AI answers, then set a baseline. In order of risk removed for effort spent:
- Open robots.txt and confirm OAI-SearchBot is not disallowed.
- Search your templates and headers for
nosnippetandmax-snippet, and remove any limit you cannot justify. - Open Search Console, go to the generative AI performance report, and write down today’s impressions with the date.
- Take the three pages that matter most commercially and add a named source and an exact figure to the claims already on them.
- If generating llms.txt is costing anybody time, stop, since Google Search ignores it.
That is the evidence-backed list. It is short because the evidence is thin, and a short list you finish is worth more than a long one somebody sells you.
If you want a second pair of eyes on it, an SEO review from us covers the crawler and snippet settings. It also covers what your pages are currently cited and ranked for, and where the gap between the two is.
The bottom line
Google and OpenAI have both written down what their AI search features need, and it is short: be indexed, be allowed to show a snippet, and do not block the search crawler. Past that, the work that earns a citation is the work that earns a ranking. Pages with clear claims, named sources and exact figures give any system, and any reader, something worth quoting.
Questions people ask about this
Do I need to add llms.txt to my website?
Not for Google. Google’s AI optimisation guide says Google Search ignores llms.txt files, so they neither help nor harm your visibility there. An Ahrefs study across 137,210 domains found 97 percent of valid files received no requests at all in May 2026. Keep one if another service you use reads it. Otherwise it is time better spent elsewhere.
Does schema markup get my pages into AI Overviews?
Not on its own. Google states in its AI features documentation that there is no special schema.org structured data needed to appear in AI Overviews or AI Mode. Structured data still helps pages qualify for rich results in ordinary search, which is a good enough reason to have it. The claim that it is the route into AI answers is not supported by the documentation.
Can I see whether my site appears in AI answers?
Partly. Google says its generative AI performance reports in Search Console had rolled out to all websites by 31 August 2026. They show impressions from AI Overviews, AI Mode and Discover AI features by page, country, device and date. The announcement lists no clicks or queries, so the report tells you that you appeared and not what it was worth.
Should I block GPTBot but allow OAI-SearchBot?
That is the setting many businesses want. OpenAI says GPTBot crawls content that may be used for training, and that each crawler setting is independent. OAI-SearchBot is the crawler that surfaces pages in ChatGPT search, and OpenAI says sites that block it will not be shown in ChatGPT search answers. Open your robots.txt, check each user agent line separately, and allow about a day for changes to register.
What does a max-snippet limit do to my pages?
It caps how much of your text Google may show in a snippet. Google’s robots meta tag documentation says the same cap limits how much of the content may be used as direct input to AI Overviews and AI Mode. A limit set years ago for a reason nobody remembers is now holding back both. Search your templates for max-snippet and nosnippet, remove any you cannot justify, and check the page source after the next deploy.
Will rewriting pages in an AI-friendly style help?
Google’s guide to generative AI search says you do not need to write in a specific way for it, or break content into tiny pieces. What the GEO research measured was the effect of quotations, statistics and named sources, not a writing style. So rewrite a page only when it is unclear or unsupported. Add the source and the figure to the claim already there, and leave a page that reads well alone.
How do I add sources without padding the page?
Put the source in the sentence that carries the claim, not in a list at the bottom. Replace a rounded figure with the exact one from the study. Quote the person who did the work in a single line. That adds a few words per claim and nothing else. Start with the three pages that matter most commercially, and check each claim still says what the source says.
Does ranking on page one still matter for AI answers?
Yes, though less than it did. Ahrefs’ March 2026 study put the share of AI Overview citations that also ranked in the organic top ten at 37.1 percent. That is down from about 76 percent in its July 2025 study, with a change of method in between. A page can be cited without ranking, and a fourth-place page can be left out. Keep working on rankings for commercial terms, and treat citations as a second outcome of the same work.
Is an AI visibility retainer worth paying for?
Ask what it does that ordinary SEO does not. If the answer is llms.txt, schema for AI, or rewriting in an AI style, Google’s own documentation says it does not need any of the three. The two supported items, crawler access and sourced claims, take an afternoon. Pay for measurement and for content that answers real questions. Google says there is no separate set of requirements to pay for.
What if AI impressions rise but traffic does not?
That is a pattern to expect for now, and it does not mean the work failed. An appearance inside a summary that answers the question completely may send nobody. Search Console shows impressions for these features but, as announced, no clicks. Compare organic clicks on the same pages over the same period, and watch whether enquiries hold. If both are steady, the summary is taking the easy queries and leaving the buyers.
How is ChatGPT search different from Google AI Mode?
They are separate systems with separate crawlers and separate rules. Google’s AI features draw on its ordinary index, so eligibility is indexing plus snippet eligibility. ChatGPT search uses OpenAI’s own crawler, OAI-SearchBot, so access is controlled in robots.txt. Check both, because a site can be fine for one and invisible to the other, and advice written for one does not carry over.
Can a small local business appear in AI answers?
Yes, if the pages are indexed and answer a question somebody asks. Google also lists keeping Business Profile information current among its best practices for AI features, so the profile matters as much as the site. For everything else, a page that states your prices, your coverage area and a named source for any claim is as eligible as anything a national brand publishes. Check Search Console after ninety days.
Free review
Want this looked at on your own site?
Send the address. A person reads it and writes back with the two or three things that would move your enquiry number first. Free, and yours to keep whether you hire us or not.
