How do you appear in Google AI Overviews?
Google publishes the answer. Be indexed and eligible for a snippet, keep your content crawlable, use semantic HTML, give the page a clear structure, and write something that is not a commodity. It also lists what does not help, and llms.txt is on that list.
Published · Last updated
By Stiven Hernandez, founder, UPAI · Updated 2026-09-25
Key takeaways
The short version of this page, in the order the sections below develop it. Each point is stated here in full so it can be read without the rest of the page, and the section that argues it gives the measurement and the source behind it.
- Publish content with a point of view, structure it with real headings, keep it crawlable, use semantic HTML, and make sure the page is eligible to show with a snippet.
- Five things, named directly in Google's documentation (2026): llms.txt and other special files, chunking content into fragments, rewriting pages for…
- The most common technical reason a page is never read: it only exists after JavaScript runs.
The failure Google's guide does not cover is the common one. Our own site shipped 133 characters of text to a crawler that does not run JavaScript, while showing 8,368 to a browser — the entire argument was invisible to the assistants it was written about.
Our own measurement: measured on upai.lat before and after prerendering, September 2026.
What does Google say to do?
Google says to publish content with a point of view, structure it with real headings, keep it crawlable, use semantic HTML, and make sure the page is eligible to show with a snippet. It also points at JavaScript SEO practice, which matters more than the sentence suggests — see the last section.
- Open your page with JavaScript disabled. If the main text disappears, nothing below this line matters yet.
- Search Google for a distinctive sentence from the page in quotation marks. No result means the page is not indexed.
- Check the page is eligible for a snippet: no
nosnippet, nodata-nosnippetaround the part you want quoted, nomax-snippet:0. - Read the raw HTML and confirm the headings are real
h2andh3elements, not styleddivs. - Ask whether the page says something a competitor's page does not. If it does not, the first four steps will not save it.
For commerce and local businesses it adds two more: keep Merchant Center feeds accurate and keep your Business Profile current. And it points at Search Console's generative AI performance report as the place to measure — first-party, free, and limited to Google.
What does Google say not to bother with?
Google says not to bother with five things, named directly in its documentation (2026): llms.txt and other special files, chunking content into fragments, rewriting pages for machines, overfocusing on structured data, and pursuing inauthentic mentions across the web. An industry analysis of 137,210 domains found 97% of them received no requests at all for their llms.txt file, which is consistent with Google's position that Search ignores it. The table below is Google's own list, with what to do instead.
| What people do | What Google says | What to do instead |
|---|---|---|
| Publish an llms.txt | Search ignores it; neither harms nor helps | Check the page is readable without JavaScript |
| Break content into small chunks | Not required | Write self-contained paragraphs under each heading |
| Rewrite pages specifically for AI | Not required | Answer the question a buyer actually types |
| Add structured data everywhere | Not required for generative AI search | Keep it as hygiene, not as a lever |
| Buy or seed mentions | Listed as something not to do | Earn real listings and correct wrong facts |
| What people do | What Google says |
|---|---|
| Publish an llms.txt | Search ignores it; it will neither harm nor help |
| Break content into small chunks | Not required |
| Rewrite pages specifically for AI | Not required |
| Add structured data everywhere | Not required for generative AI search |
| Buy or seed mentions across the web | Listed as something not to do |
"You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn't use them."
What does the guide leave out?
The guide leaves out the most common technical reason a page is never read: it only exists after JavaScript runs. Googlebot executes JavaScript, so Google's own guidance can afford to treat this as solved. The other assistants' crawlers are widely reported not to. We found this on our own site. A request without a browser returned 133 characters of text — an empty shell — while a browser saw 8,368. Every figure, every source and every argument existed only after the page had run its JavaScript. Our robots.txt explicitly invited fifteen AI crawlers into a room with nothing in it.
Check yours before changing anything else: request your own page without a browser and count the text that comes back. If the answer is a few hundred characters, nothing else on this list matters yet.
Questions people ask next
These are the follow-ups we are asked most often once the answer above has landed. The same rule applies to each: where a figure appears, its source and year sit in the same sentence, and where we do not know something we say so rather than estimate it.
- Does structured data help?
- Google says not to overfocus on it and that it is not required for generative AI search. It does not hurt and it helps other surfaces, so it is worth having and it is not the lever.
- Where do I measure this?
- Search Console has a generative AI performance report. It is free and first-party, and it covers Google only — which is the catch, because Google is one of four surfaces your buyers use.
- Does any of this apply to ChatGPT?
- No, and be careful with anyone who says it does. Google's guidance is about Google Search. The other assistants publish far less about how they choose sources, which is exactly why measuring all four is the only way to know where you stand.
Related
Other questions in this library that cover the ground next to this one. Each is answered the same way — the direct answer first, then the measurement behind it — and each carries its own sources, so you can check any of them on its own page without taking this one on trust.
Sources
Every figure on this page comes from one of the sources below, each linked so you can open it and check the number in its original context rather than against our summary. Where a figure is our own measurement, its method and sample size are stated with it.