AI Overviews are the answers Google writes with AI above the search results, with links to the sources it used. They went live in Spain on March 25, 2025, where they're called "vistas creadas con IA." There's nothing to switch on to be eligible.
This guide covers four things we couldn't find in the pages currently ranking for this query: the exact date and conditions of the European rollout, what Google's own documentation says you don't need to do, a free and repeatable way to check whether a query triggers the block (with three queries measured on the same day), and what Search Console can actually tell you about clicks.
What are AI Overviews, and how do they differ from a featured snippet?
An AI Overview is a block of text generated by Google's models that summarizes the answer to a query and links to the pages that answer came from. It doesn't lift a paragraph from one site. It writes a new one out of several.
That's the difference from the featured snippet, the box Google has been showing for years and that plenty of guides still treat as the same thing.
Short version: a featured snippet pulls a verbatim passage from one page and you can predict it by looking at who holds the top 10; an AI Overview writes new text from several sources, varies between users, and can't be predicted from your position. You lose a featured snippet when someone takes it from you. You don't "win" an AI Overview. You're one of its sources, or you aren't.
The old name was SGE (Search Generative Experience), and it still shows up in articles from 2024. As a search term it's dead: it went from 590 monthly searches in Spain in January 2024 to 10 in July 2026.
When did AI Overviews reach Europe?
On March 25, 2025. That day Google announced the rollout to nine European countries: Austria, Belgium, Germany, Ireland, Italy, Poland, Portugal, Spain and Switzerland. In Spain they launched in both Spanish and English.
The fine print in that announcement matters, because it changes what you see when you check a query: the block is shown to signed-in users aged 18 and over. If you check a query from a private window without signing in and you don't see the block, that doesn't mean the query doesn't have one.
How does Google pick which pages it cites?
Google wrote the answer down in its guidance for site owners. It's shorter than the industry would like:
"To be eligible to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements."
And then the sentence that dismantles a fair number of billable services: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary."
The same page gives you five denials worth keeping on hand the next time someone pitches you an "AI optimization" package:
- You don't need to create machine-readable files or AI text files. That includes
llms.txt. - You don't need special markup for these features.
- There's no schema.org structured data you have to add for them.
- You don't need to chunk your content into fragments for the AI.
- You don't need to rewrite your style for the models.
Our read, after working on generative engine optimization across several accounts: eligibility is a floor that almost any healthy site already clears, which is exactly why nothing gets decided there. What gets decided is whether an answer can be pulled out of your page. Being eligible to be cited and being cited aren't the same thing, and Google's documentation only covers the first one.
How do you check whether a query shows an AI Overview?
No paid tool required, and repeatable. Two ways, and you want both.
By hand. Search while signed in, in the country and language of your audience. Note whether the block appears, where it sits on the page, and which sources it links. Do it again on a different day: the block isn't stable.
With data. Any tool that returns SERP features flags the block as one more element, with its position. We use OpenSEO. The useful reading isn't "yes or no," it's where it sits: a block in first position takes the screen; a block in seventh, below the video and the images, barely exists.
Three queries measured on the same day, September 8, 2026, on Google Spain:
The numbers: "ai overviews" shows the block in position 1; "impermeabilizar terraza sin obras" (a Spanish home-improvement query) shows it in position 7, behind a video, the related questions, three organic results and the images; and "que es el generative engine optimization" shows no block at all, even though it's a pure definition question, exactly the kind of query everyone assumes triggers one.
That third row is the one that changes decisions. Before you rewrite a page "for AI Overviews," check whether its queries have any. You may be optimizing for a block that never shows.
Seven criteria for getting cited
None of these come from a Google announcement. They come from crossing what Google publishes about eligibility with what we measure on pages that end up inside the block.
In brief: answer the question in the first two or three sentences of each section; turn your H2s into the questions people actually type; make every section stand on its own without the ones before it; use figures with a source and a date; confirm that AI crawlers get a real 200 from your server; keep the technical SEO healthy (indexing, canonicals, unique titles), because eligibility depends on it; and don't duplicate the same answer across three of your own pages.
The first three are about form, not substance, and they're the ones most sites get wrong. An article can hold the best information in its industry and still be useless to a model if that information shows up in paragraph fourteen.
A worked example: two pages from the same query
Both rank for "ai overviews" in Spain. Both are well written. Their shapes are opposites.
Page A (a marketing platform's glossary entry, the fourth element on the results page on September 8) opens with the full definition in its first two sentences, about 320 characters. Its H2s are literal questions. It has a comparison table against the featured snippet and a nine-question FAQ. Roughly 3,300 words.
Page B (an agency guide, the fourteenth element on the same page) runs about 4,700 words, with an anchored table of contents and more technical depth than A. Its H2s mix questions and statements. There's no table and no FAQ, and the definition of its central concept lands after about 380 characters of context.
Here's where we have to be straight, because almost no guide is: we can't tell you which of the two the block cites. The SERP tools we use return that the block exists and where it sits, not which pages its text came from. And the block varies by user and by day, so a manual check is only good for the day you ran it. Anyone showing you a table of "cited pages" without telling you how they built it is showing you an estimate.
What we can tell you is the shape. Page A can be broken apart: each section answers one question and holds up alone. Page B has to be read. If the mechanism is pulling out passages that answer on their own, A hands the model finished material and B makes it do the assembly. That part is under your control.
Do you lose clicks if you show up?
What Google states in writing is specific, and worth reading slowly: "when people click from search results pages with AI Overviews, these clicks are higher quality (meaning, users are more likely to spend more time on the site)."
That's a claim about the quality of the click, not the volume. Google doesn't say there are more clicks.
On the measurement side there's a limitation almost nobody declares: Search Console doesn't break out AI Overviews. Google's documentation says so plainly — traffic from AI features is folded into the overall Performance report, under the "Web" search type. There's no AI Overviews filter. Anyone presenting you with "your clicks from AI Overviews" has estimated them.
What you can read is the gap between impressions and clicks. One B2B construction account we run, August 2026 against July of the same year:
The numbers: impressions went from 36,488 to 49,221 (+34.9%), clicks from 1,014 to 1,183 (+16.7%), and CTR fell from 2.78% to 2.40% (−13.5%). Impressions grew at twice the rate of clicks. That's the pattern the industry attributes to AI Overviews.
We don't attribute it. These numbers can't establish the cause. New content shipped in that window, average position moved, and Search Console doesn't separate the block. What they're good for is narrower and still worth having: if your CTR falls while impressions rise, comparing clicks against last month stops measuring what you think it measures. Change the metric before you draw the conclusion.
FAQ
Can I stop Google from using my content in AI Overviews?
You can use nosnippet, max-snippet or data-nosnippet to limit the snippet, but eligibility for AI Overviews is tied to eligibility to appear with a snippet. Opting out of the block means opting out of regular search snippets too.
Does llms.txt help me show up?
Not for AI Overviews. Google states in writing that AI text files aren't needed. We publish one on auroraglobalgroup.com because it takes five minutes and other agents read it, but it isn't a lever for Google's block.
Does structured data get me in?
There's no specific schema that turns the feature on. Structured data still earns its place for other things (rich results, entity clarity), which is why we ship it, but it isn't a requirement here.
How many queries show an AI Overview?
It depends on the sector and it shifts week to week, so any global percentage ages badly. What's useful is measuring it across your own twenty or thirty commercial queries and repeating the measurement monthly.
Is AI Mode the same as AI Overviews?
No. AI Overviews are a block inside the standard results page. AI Mode is a separate conversational tab. They share eligibility requirements and they're measured together in Search Console.
Is showing up worth anything if nobody clicks?
It works as a brand mention in front of someone who's deciding, the same way a featured snippet gets read without a click. What doesn't work is measuring it in clicks and expecting them to rise.
Google, "We're bringing the helpfulness of AI Overviews to more countries in Europe," blog.google, March 25, 2025 — date and countries of the European rollout.
Google Search Central, "AI Features and Your Website" — eligibility, what isn't required, and how it's reported in Search Console.
OpenSEO (DataForSEO data for Google Spain), retrieved September 8, 2026 — volumes, difficulty and SERP features for the three queries.
Google Search Console, property of an Aurora client account, July and August 2026 — impressions, clicks and CTR.
What we do with this
We start by measuring: which of your commercial queries show the block today, where it sits, and which sources it links. Then we work on the shape of the pages competing in those queries, and we measure again the following month with the same queries. Without that first measurement, everything after it is opinion.
If you want that first measurement on your domain, it's part of our GEO audit.
More on this: SEO for AI: how to get ChatGPT to cite you and Generative Engine Optimization (GEO): the complete guide.