Google AI Overviews are the generated summaries that appear at the top of some search results, assembled from a handful of sources and shown before the ordinary links. Sometimes they add context. Sometimes they answer the question completely, and the searcher never clicks anything.

For a small business, that is the uncomfortable part: a good ranking can coexist with a fall in clicks, because the answer was given above you. But the response is not to abandon search. It is to understand what the Overview draws on and make sure you are one of the sources it draws on, not one of the pages it replaces.

What an AI Overview actually is

An Overview is assembled at the moment of the query. Google’s systems gather a set of sources that appear to answer the question, generate a short synthesis, and cite a few of them. The citation set is small — usually a handful of links, often including pages ranking well but not exclusively the number one.

This is the same mechanism as the stand-alone assistants, inside Google. And the same behaviour shows up in the baseline: citation is not the same as ranking. The sources an answer leans on are those that state the answer cleanly enough to be summarised. That is a different quality from “optimised for a keyword”.

When an Overview removes the click — and when it does not

Overviews are not uniformly destructive. The behaviour depends on the intent behind the query.

  • Informational queries (“what is”, “how do I”, “does X need Y”) are the ones an Overview can answer outright, and the ones where the click is most at risk. This is where being the cited source matters most.
  • Commercial and local queries (“best X in Y”, “X near me”, “X cost in the UK”) are harder to finish in a paragraph. They tend to generate follow-up, comparison and a choice, which means clicks still happen. Here an Overview can be a shop window rather than a wall.
  • Branded queries are mostly unaffected; the searcher already knows who you are.

So the first useful thing is to sort your priority questions into “answerable in a paragraph” and “needs a decision”. The first group is where you must be cited. The second is where you must be present and easy to choose.

What to do about them

The work overlaps almost entirely with being citable by any AI surface. The specifics:

  1. Answer the question in the first paragraph. Not a preamble, not a definition of the industry — the answer, in plain sentences, with the number or the recommendation in it. That is what gets lifted and what earns the citation link.
  2. Give the overview a reason to cite you rather than a competitor. Specifics: real prices, real timeframes, a named method. Generic pages are easy to summarise without attribution; distinct ones are worth linking.
  3. Use structure a machine can read. Headings that match the question, lists, tables, FAQ markup, and facts in structured data. Overviews are assembled from extractable pieces.
  4. Keep the page fresh. A visible reviewed date and a real update cycle help for a surface that favours current sources.
  5. Keep the pages you already earn from. Before changing anything, know which URLs bring clicks and links. An Overview is not a reason to rebuild your site; a drop in clicks with no research behind the response is how good pages get lost.

What not to do

The panic response is expensive and usually wrong: rewriting the whole site around “AI”, stuffing FAQs, or paying for an “AI Overview optimisation” package with no method. The Overview is sourced from pages that answer things well. The way to be in it is to answer things well, in a form that can be quoted, on pages that are findable and current.

It is also worth remembering that Overviews do not appear on every query, and their presence varies by topic and over time. Optimising as though every search now begins with one would distort a good site to chase a surface that is neither universal nor stable.

How to tell whether it is working

Count, don’t guess. Put your priority questions to Google AI Mode (and the assistants) on a schedule, and record whether you are cited and whether clicks to the answering pages hold. The same question set that measures AI visibility measures Overviews, because AI Mode is one of the four surfaces in the baseline.1 If your citations rise while your clicks hold, you are gaining a surface rather than trading one away.

If you want the shortlist of questions worth tracking, that is what GroundWork produces. Keeping the answers current and the citations moving is Coverage. The mechanism behind all of it is the main guide: how AI assistants choose which businesses to recommend.

  1. KOODOS AI-visibility baseline, 17 September 2026: 15 questions × Perplexity, ChatGPT, Gemini and Google AI Mode, 60 completed calls. Raw data in docs/research/ai-visibility-2026-09/. ↩