Optimising for Google AI Overviews

How AI Overviews Assemble an Answer

An AI Overview is not a ranking of one page. Google retrieves a set of candidate passages from multiple documents, generates a synthesised answer, and links to the sources those passages came from.

The practical consequence is that the unit of retrieval is the passage, not the page. A page ranking eighth organically can supply a cited passage if that specific passage answers the question cleanly, while a page ranking first can be omitted if its relevant content is buried in a long preamble.

This is why the optimisation target shifts from "rank the page" to "make individual passages independently answerable".

Which Queries Trigger Them

Triggering is uneven and changes frequently, but the pattern is reasonably stable. Overviews appear most on informational and explanatory queries — how something works, what something means, comparisons, and multi-part questions.

They appear far less on navigational queries, on transactional queries where Google prefers to show shopping or ads, and on queries in YMYL categories where Google is more cautious — though medical and financial overviews do appear with heavier caveating.

Check your own query set rather than relying on general figures. Trigger rates vary enormously by industry, and the only reliable data is your own.

What Makes a Passage Quotable

Content that gets cited tends to share specific structural traits:

• A direct answer stated early, before elaboration. A definition in the first two sentences under a heading is far more extractable than the same definition reached after three paragraphs of context.
• Self-contained passages. A paragraph that depends on the previous three to make sense cannot be lifted.
• Question-shaped headings that match how people actually phrase the query.
• Specific, checkable detail — figures, named steps, dates, concrete examples. Generic assertion is not worth citing.
• Clear entity language. Name the thing rather than relying on pronouns and implied subjects.

The Click-Through Reality

AI Overviews reduce clicks on the queries where they appear. This is not disputable and it is the honest starting point for any strategy discussion.

Two things soften it. First, the clicks that do occur tend to be from users who wanted more than the summary gave them, which is a more qualified visitor. Second, citation itself has brand value — appearing as a named source in an answer millions of people see is a visibility asset even without the click.

What this argues for is shifting effort toward queries where a summary cannot substitute for the page: anything requiring a tool, a calculator, a download, a demonstration, pricing, or a decision that depends on the user's specific situation.

Measuring It

Search Console does not separate AI Overview impressions from standard impressions, which makes direct measurement awkward. Workable proxies:

• Track a fixed set of target queries manually or with a rank tracker that records Overview presence
• Watch for the signature pattern — impressions flat or rising while clicks fall on informational queries
• Segment performance by query intent; informational and transactional queries should be trending differently

Do not diagnose a general ranking problem from a change that is actually confined to Overview-triggering queries. Segment before concluding.

Reading the Traffic Change Correctly

The pattern reported most often is impressions holding while clicks fall, and interpreting it correctly determines whether you respond usefully or expensively.

Segment before concluding anything. The effect is heavily concentrated in informational queries with short factual answers. Transactional and navigational performance is often unchanged, and a blended figure conceals both.

Compare query by query, not in aggregate. Identify the specific queries where an overview now appears and measure those against themselves over time. That is the only comparison that isolates the cause.

Do not respond by rewriting titles. A falling CTR on a query whose answer is now displayed is not a title problem, and title work is where teams spend the first month for no return.

Watch branded search and direct traffic as the indirect signal that appearing without a click still does something.

And be honest in the reporting: some of this traffic is not coming back, and a plan built on recovering it will keep failing.

Where the Effort Actually Pays

Two responses have evidence behind them and one does not.

Be extractable. Answer in the first sentence under a heading, keep paragraphs self-contained, state definitions plainly, and use specific checkable claims. This is what gets a page used as a source, and it is better writing regardless.

Shift the content mix. Toward questions that cannot be answered in a box — comparisons with trade-offs, judgement calls, tools that need the user's own numbers, anything where currency matters and a cached answer may be stale. This is the strategic response and it is slower than anyone wants.

What does not pay: trying to be excluded. Blocking the feature generally means losing the organic result alongside it, which trades a reduced click rate for none.

And measure the right thing. If you are being cited without being visited, the metric that matters is whether people arrive later by name — so branded search volume and direct traffic become the scoreboard, not sessions on the page.

Sources

What each claim on this page rests on. Entries are typed so you can see which are primary.