Key takeaways
- Google states literally that there are no additional requirements or special optimisations to appear in AI Overviews or AI Mode.
- It explicitly says you do not need to create machine-readable files, AI text files or new markup.
- Traffic from these features appears in Search Console under the «Web» search type, with no separate breakdown.
- All of that applies to Google. ChatGPT, Perplexity and Claude choose their sources independently and have published nothing equivalent.
What is Google AI Mode?
Google AI Mode is the conversational version of the search engine. Instead of a list of links it returns a written answer with links to the sources it used. It sits alongside AI Overviews, which appear above the ordinary results.
For anybody publishing, there is one question: what do I have to do to appear there. In May 2026 Google published official guidance answering it.
What Google says exactly
The central line of that documentation is blunter than the industry would like. It reads: «there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary».
And in case any doubt remained, it adds: «you don't need to create new machine readable files, AI text files, or markup to appear in these features».
That second sentence has an audience. For two years, audits have been sold for implementing llms.txt-style files and AI-specific markup. Google says they do nothing for its AI Mode. We already said so in SEO for AI search engines, and now the source has written it down.
So what does it recommend?
Google's list is, deliberately, the usual one. Eight points anybody will recognise.
- Meet the technical requirements for search.
- Create helpful, reliable content written for people.
- Allow crawling in robots.txt.
- Make content easy to find through internal links.
- Look after page experience.
- Keep important content in text, not only in images or video.
- Support it with good images and video.
- Make structured data match what is visible on the page.
The penultimate point is the one most people breach without noticing. Key content tucked inside an image or a video, where no system is going to read it.
The crawling point is worth checking rather than assuming. Our analyser goes through the AI crawlers that matter one by one and tells you which ones your site shuts out.
We see it in three places above all. Prices inside a rate-card image, product specifications in an infographic, and opening hours or conditions buried in a PDF. All three are invisible to the system deciding whether you appear.
The last point carries its own trap. Structured data must match what the page shows. Marking up a price or a rating that never appears on screen is what Google classifies as misleading markup.
Why is Google saying this now?
Because it has spent two years watching a parallel industry grow around its AI features. New acronyms, specific audits and file formats proposed as standards have appeared, and Google has adopted none of them.
It is worth noting how many of those proposals came from vendors who then sold the implementation. Not illegitimate, but it explains why the official documentation took so long and why, when it arrived, it was so blunt.
There is a self-interested reading of that refusal, and it deserves saying: it suits Google for the industry to keep playing its own game, where it holds a twenty-year lead. But the claim is verifiable on the point that matters: no markup or file exists that Google recognises specifically for these features.
That leaves the work where it was. If a page answers a question well and the system can read it, it is a candidate. If not, no file will rescue it. The same reasoning applies to structured data: it describes what is already on the page rather than replacing it.
How do you measure whether you appear?
Here comes the bad news. Traffic arriving from AI Overviews and AI Mode is grouped in Search Console under the «Web» search type, mixed with everything else. There is no separate breakdown.
Google does make a claim about the quality of those clicks. It says clicks from pages with AI Overviews are higher quality, in the sense that users spend more time on the site. That is Google's own statement with no published methodology, so treat it as such.
It also says these features display a wider and more diverse set of links than classic search, creating opportunities for a broader range of sites. It sounds good and cannot be verified from outside, because the breakdown that would allow measuring it does not exist.
To know whether you really appear, it has to be measured externally: a panel of queries rerun monthly and checked by hand. That is what we do in AI visibility audits.
The caveat Google does not cover
All of the above applies to Google. And Google is no longer the only place people ask.
ChatGPT, Perplexity and Claude select sources with their own systems, and none has published equivalent guidance. Anthropic has put in writing what it considers manipulating AI sources, but not how it picks its own. What we know about them comes from observation and experiments rather than documentation. Differences do matter there: which crawler each uses, whether it reads in real time or from its own training, and which content format it reuses most readily.
So the honest answer to «do I need anything special for AI?» has two halves. For Google, nothing beyond ordinary SEO done properly. For the other assistants there is specific work, and it starts with checking their crawlers can get in, which many sites block without realising.
That blocking is rarely deliberate. It usually comes from a firewall, a hosting provider rule or a robots.txt template copied years ago. The check is quick and worth running first, because until the crawler gets in, nothing else has any effect.
What actually blocks your appearance
If Google says nothing special is needed, the useful question changes. It becomes what is getting in the way.
- Blocked crawlers. A robots.txt or firewall rule keeping them out. The most common block and the quietest.
- Content that only exists in JavaScript. If the text appears after code runs, many systems never see it.
- Orphan pages. Content nothing links to from within the site, so nobody finds it.
- Buried answers. The answer to the question turns up in paragraph twelve, after an introduction that does not answer it.
All four are fixed without new tools and without extra line items. And all four do more than any file proposed as a standard.
What to do with this
If somebody offers you a specific optimisation for Google AI Mode, show them the documentation. No such thing exists, and paying for it means paying twice for the same SEO.
What does pay off is the dull stuff. Content in text, answering a specific question properly, crawlers able to get in, and structured data saying the same thing as the page. That holds for classic search, for AI Mode and, with caveats, for the other assistants. Google is even testing a programme for paying publishers whose content holds up AI answers, which suggests it tells sources apart.
Frequently asked questions
Is anything special needed to appear in Google AI Mode?
Google says no: there are no additional requirements or special optimisations, and no need to create machine-readable files or new markup.
Is an llms.txt file any use?
For Google AI Mode, according to its documentation, no. Other assistants are a separate matter: they have published no criteria and the evidence there is observational.
How do I see how much traffic comes from AI Mode?
There is no separate breakdown. Search Console groups it under the «Web» search type alongside the rest of search traffic.
Does the same apply to ChatGPT or Perplexity?
Not necessarily. Each assistant selects sources with its own systems and none has published equivalent guidance. The first step there is checking their crawlers can access your site.
Are clicks from AI better?
Google states they are higher quality because users spend more time on the site. That is its own claim with no published methodology, so treat it accordingly.



