There is no separate door
ChatGPT and Perplexity each run their own crawler, so with them you can be in one and out of the other. Google does not work that way. Its documentation says AI is built into Search, which is why robots.txt directives for Googlebot are the control for site owners. One crawler feeds the blue links and the AI answers alike.
Practically, that means eligibility for AI Overviews is the same eligibility you already have for Search. If Google can find, crawl and index your page, it can also pull it into an AI answer. If it cannot, no amount of AEO work changes anything.
Google-Extended is not the crawler people think it is
This one confuses almost everybody, including people who write about it. Google-Extended is not a crawler at all. In Google's own words it has no separate user agent string, and crawling is done with the existing Google user agents. It is a switch you flip in robots.txt to say whether your content may be used to train future Gemini models.
The line that settles the argument is Google's: Google-Extended does not impact a site's inclusion in Google Search, nor is it used as a ranking signal. Block it if you would rather not feed model training. You will still be eligible for AI Overviews the next morning.
The settings that do take you out
Four controls genuinely reduce or remove what Google can show from your page, and every one of them is something you or your builder switched on:
- noindex removes the page from Search entirely, and with it any chance of an AI Overview. It can hide in a meta tag or in an HTTP header called X-Robots-Tag, where you will never see it by reading your page.
- nosnippet stops Google showing any text extract from the page. No extract means nothing to build an answer from.
- data-nosnippet does the same for the specific chunks of a page you wrap with it.
- max-snippet limits the extract to a set number of characters, which can leave too little to be useful.
- Disallowing Googlebot in robots.txt cuts off everything at once, ordinary results included.
What is left is the work you already know
Once you are eligible, Google does not publish how it picks which sites to credit, and anyone selling you a guaranteed method is guessing. What can be said honestly is that the pages showing up in AI answers are the ones ranking well for that query and answering the question cleanly, close to the top of the page.
So the list is short and dull: get indexed, cover the question properly, put the answer in the first couple of sentences under a heading that matches how people ask it, and keep the page current. There is no AI Overviews checkbox to find, which is either disappointing or a relief, depending on how much you were dreading it.
Check where you stand
Search Google for site: followed by your domain, with no space. If your pages come back, you are indexed and eligible. If nothing comes back, indexing is the problem to solve before anything else, and it is usually a noindex somebody left on after a redesign.
Since June 2026 Google also gives you a direct reading. Search Console, its free dashboard for site owners, has a Generative AI performance report under Performance that counts how many times your pages appeared inside AI Overviews and AI Mode, split by page, country, device and date. It shows impressions only, never clicks, and Google opened it to every site worldwide on 31 August 2026. If you are indexed and that report stays empty, you are eligible but not yet being picked, or not often enough to register, and the work is on the page rather than on access.
HowFindable scans your site for free and flags a noindex whether it sits in your HTML or in the HTTP header, along with the structure and answer-first signals that decide whether there is anything quotable on the page. If you want the fixes, we generate the corrected files ready to upload.