An AI Overview is a generated summary that appears above Google's regular search results, written by a model that assembles the answer from pages in Google's index and links a handful of them as sources.
That sentence contains the two things people most often get wrong. It is generated, not extracted. And the sources it links are not the same thing as the brands it names.
What an AI Overview actually is, mechanically
Google's AI features documentation describes generative features as rooted in the core Search ranking and quality systems, assembling answers from the Search index using retrieval-augmented generation.
In practice: you type a query, Google retrieves a set of pages, a model reads them and writes new prose, and a few of those pages appear as source links attached to the answer.
Three consequences follow from "writes new prose," and they matter more than anything else on this page.
It varies. Ask the same query twice an hour apart and you can get different wording, a different brand order, a different set of brands entirely, with nothing having changed in the world. A single screenshot is not evidence of anything.
Nobody "wins" it. There is no position zero to occupy. There are contributors, and being one of them is a matter of degree measured over many checks.
Cited and named come apart. Your page can be in the source list while a competitor is in the sentence someone reads. That is not a bug in the measurement. It is the most common surprising finding in this whole category.
AI Overview versus featured snippet
These get conflated constantly and they are mechanically different.
A featured snippet is an extract. Google lifts a passage verbatim from one page and shows it above the results with a link. One page wins. You can point at it. The old position-zero playbook applies: answer in the first forty words, match the query phrasing, use a clean definition paragraph or list.
An AI Overview is a synthesis. A model writes text that exists nowhere on any single page, drawing on several. There is no one winner, the wording is not yours, and your brand can appear in the prose without your page being cited, or your page can be cited without your brand appearing.
If you optimize for Overviews using snippet tactics you will not be wasting your time, because clean passage-level answers help both. You will just be measuring the wrong thing when you check whether it worked.
AI Overview versus AI Mode
Also not the same, and this one costs people money.
Google launched AI Mode as a separate tab and has continued extending it. It is a conversational surface with its own retrieval. Both it and AI Overviews draw on Google's index. Both are Google.
They still disagree. We have measured the two naming different brands for the same query on the same day. Reporting them as one number describes neither, which is why we keep every engine separate and treat those two as separate engines rather than one Google line item.
What they do to traffic
This is the part that moved AI Overviews from a curiosity to a budget line.
Pew Research Center tracked real browsing behavior in March 2025 and found that users who encountered an AI summary clicked a traditional search result in 8% of visits, compared with 15% for users who did not. Roughly half as often. They clicked a link inside the summary itself in 1% of visits, and 26% of pages with an AI summary ended the browsing session against 16% without.
Ahrefs approached it from the ranking side. Across 300,000 keywords they measured a 34.5% lower average click-through rate for the top-ranking page when an AI Overview was present.
Two independent methods, the same direction. The answer is increasingly where the query ends.
Four outcomes, not one ranking
Because there is no position to hold, every check on a tracked query lands in one of four states, and collapsing them loses the diagnosis.
No Overview appeared. Worth recording separately. Overviews do not show for every query, and a query without one is not a query you are losing. Counting it as an absence makes your numbers worse than reality.
Named and cited. You are in the prose and your page is in the sources. The outcome to protect.
Cited, not named. Your page was read and a competitor got the sentence. Retrieval works; framing is losing. Cheaper to fix than it looks.
Neither. Not in the answer, not in the sources. Check reachability before touching the copy.
Keeping these apart is the entire reason our AI Overview tracker reports four states rather than one percentage.
How brands get into them
There is no separate lever, and that is the honest answer however much the industry dislikes it.
Google states plainly that structured data is not required for generative AI search and there is no special schema.org markup you need to add. There is no AEO schema. There is no Overview-specific ranking system.
What there is:
An eligibility gate. A page must be indexed and eligible to appear in Search with a snippet. Google's guidance is that nosnippet, data-nosnippet, max-snippet and noindex restrict how your content is featured in AI experiences. One legacy directive on a template can disqualify a whole section of a site.
Query fan-out. Google generates related queries alongside the one typed and retrieves for all of them, so you compete for questions nobody asked. A page answering one narrow question thoroughly can surface for an adjacent one it never targeted.
Ordinary passage quality. Answer the question in prose a model can lift, and make sure the third-party pages that get cited in your category describe you accurately. Those two do most of the work.
Where AI Overviews show up most
Not evenly. In our measurements and in the published research, Overviews cluster on informational and comparative queries, the "what is", "best", "how do I" shapes, and appear far less often on navigational or transactional ones.
That distribution is awkward for B2B specifically, because the informational queries where Overviews dominate are exactly the top-of-funnel questions buyers ask before they know any vendor names. Those are the queries where being named matters most and where you have historically been least able to see what happened.
It also means a tracked set weighted toward branded or bottom-funnel queries will show almost no Overviews and tell you almost nothing. Choosing which queries to track is a methodological decision, not a setup step, which is why prompt tracking shows search volume beside each suggestion rather than leaving you to guess.
How to actually check any of this
Not by looking. By repeating.
Because Overviews are generated fresh, you need enough checks on a fixed query to separate a real change from ordinary variance, which is roughly thirty per query. Every rate needs the count behind it, and below five checks a fraction is more honest than a percentage. Rewriting a tracked query starts a new series, because a trend across two different wordings is not a trend. That is our methodology, in public.
For the mechanics end to end, see how to track Google AI Overviews. For what moves the result once you can see it, see how to rank in AI Overviews.