AI Overviews
AI Overviews is an AI-generated summary Google shows above the organic results. It is assembled from several sources, links to them, and the user never asked for it: it simply appears. For a website that means you can be visible in first place and still receive no clicks at all.
In short
| What it is | A model-generated answer block above the classic search results |
| When it triggers | On informational and comparison queries with several available sources. Rarely on navigational queries, increasingly on purchase queries, including product overviews |
| What it does to traffic | It cuts clicks on queries where the answer is enough. The share of impressions without a visit grows |
| How to get into it | A clear statement right under a heading, a structure the model can take apart, and data nobody else has |
| How it is measured | Indirectly: the impression-to-click ratio on informational queries, plus citation reports such as AI Performance |
How an AI Overview is built
Google evaluates the query, picks a set of sources from its own index and lets a model assemble the answer. It is not a new ranking algorithm, it is a layer on top of one. Pages already holding visibility get into the candidate set, but the order inside the block does not match the order of the links. It is common for a cited page to sit on page two of the organic results.
The model looks for statements, not pages
What enters the answer is a specific sentence or paragraph that answers a sub-question completely, without the surrounding context. That is the fundamental difference from classic optimization: it is not enough for the page to be about the topic, it has to contain a formulation that can be lifted and used on its own.
Sources rotate faster than rankings
The composition of the block changes between queries and over time, because the model re-selects on every query. Tracking a single control phrase therefore tells you very little. You have to measure a sample.
From our own practice: positions without clicks
On informational queries the same picture repeats: a page holds a top position on a long query and receives practically no clicks from it, while the same page sits deep on a short head term.
A second view of the same thing: search impressions grow much faster than clicks. The widening gap between impressions and clicks is not an error, it is what this layer does.
The practical conclusion: on informational queries the click stops being the primary metric. If a client sets the goal as organic traffic, AI Overviews will always look like a loss.
Common mistakes
- Responding by pulling the content. Blocking the model means disappearing from the answer without gaining a single extra click, because the query gets answered from other sources.
- Optimizing for rank only. Position one in the links does not guarantee a citation in the block, and the reverse holds too.
- Writing long introductions. Text that only reaches the answer after three paragraphs gets skipped.
- Measuring success by traffic alone. The report then shows a decline exactly where visibility is in fact growing.
How to respond
- Separate the queries. Measure informational queries on impressions and position, transactional ones on clicks and conversions. Mixing them into one chart makes no sense.
- Put the answer on the first screen. Two or three sentences under the heading that work standalone.
- Add your own numbers. Original data is the strongest reason to cite you rather than someone else.
- Deploy structured data. FAQPage and Organization help the model understand what is on the page.
- Watch technical availability. A site a crawler cannot reach never enters the candidate set.
Related terms
See also AI Mode, zero-click search, AEO, GEO, RAG and citation share.
Frequently asked questions
How do I know my page appeared in an AI Overview?
There is no direct report yet. The closest signal is a combination of high position, growing impressions and flat clicks on informational queries, backed by a manual check of a sample.
Can you opt out of AI Overviews?
You can, through content preview settings, but in practice it means losing visibility with nothing in return. The query gets answered from competing sources.
Does AI Overviews cost everyone traffic?
Not evenly. The biggest losers are pages that answered a simple factual question. Pages where the user needs a tool, pricing, a comparison or a contact feel far less impact.
Does schema help?
It aids comprehension but guarantees nothing. Without a clear answer in the text, structured data has nothing to point at.
How we can help
Visibility in AI search is one of our services, from technical availability through structured data to content that can actually be cited. Details are on the AI visibility agency page.