Why is Google AI search engagement decreasing?

Ask three people why Google AI search engagement is decreasing and you will get three incompatible answers, all of them citing real data. That is not because the data is bad. It is because two completely different declines are being described with the same words.

Two declines, one label

Decline one: people click less. When an AI Overview appears, a smaller share of searchers click through to any website. This is thoroughly documented and not seriously disputed. It is a decline in engagement with websites.

Decline two: people used the AI surfaces less than expected. Google’s dedicated AI Mode tab took much longer to find an audience than its launch coverage implied. That is a shortfall in engagement with the AI product itself.

These have opposite implications. The first says AI search is working so well it is eating the click. The second said users were slower to adopt a chat-shaped search box than Google hoped. Both were true at different moments, and only one of them is still true.

Decline one: the click collapse, and how big it really is

The most careful public work here is Pew Research’s, because it observed real behaviour rather than modelling it. Tracking roughly 68,000 real search queries, Pew found that users clicked a result in 8% of visits where an AI summary appeared, against 15% where one did not — a relative decline of about 47%.

Ahrefs, working from a much larger but differently constructed dataset, published the trajectory over time. In April 2025 it measured a 34.5% CTR decline for the top-ranking result when an AI Overview was present. By February 2026 the same analysis put the figure at 58%.

Whichever number you prefer, the direction is consistent and the magnitude is large enough to reorganise a content strategy around. The publisher-side data agrees: reporting on the first half of 2025 put the median publisher’s year-over-year traffic decline at around 10%, with non-news content sites down about 14%.

The click did not disappear. It got more expensive to earn, and it now goes to fewer sites per query.

Two structural facts sit underneath that. First, an AI Overview physically occupies the space the first organic result used to hold, so on mobile position one can now sit below the fold — the rank did not change, the geography did. Second, a results page offers ten organic slots while a generated answer typically names three to five sources. That is a reduction in how many sites can win a query that is independent of quality.

Decline two: the AI Mode adoption curve

The second story is more interesting because it reversed, and most commentary has not caught up.

When Google launched AI Mode as a dedicated tab, early measurement was underwhelming. Analysis by Moz’s Tom Capper found AI Mode capturing roughly 1–2% of total US search traffic, and Similarweb data showed usage of the AI Mode tab on Google.com in the US dipping after initial growth to sit at just over 1%. Google’s VP of Search, Nick Fox, put AI Mode at 75 million users in December 2025.

Then it accelerated. By May 2026 Google was publicly reporting 2.5 billion monthly active users for AI Overviews and 1 billion for AI Mode. AI Mode roughly quadrupled between May and November 2025 and doubled again over the following six months.

So the honest summary is: the AI surfaces are growing fast, the dedicated chat tab was slow out of the gate and then caught up, and the thing genuinely and durably decreasing is the number of clicks a website receives per search.

Which one is happening to you

If you are searching for why your AI search engagement is down, decline one is almost always what you are experiencing. But “AI took the clicks” is true without being actionable — there are eight distinct mechanisms underneath it, and they call for different responses. We break each one down in the reasons behind the decline.

One mechanism deserves flagging here because it is invisible in Google’s data entirely: some queries never reach Google at all now. They are typed into ChatGPT, Perplexity, Claude or Gemini directly. No Google-side metric can show you this, because the search did not happen there. Measuring only Google systematically understates the problem — which is the core argument for tracking prompts across engines rather than keywords in one.

Working out your own exposure

Averages are close to useless at the site level, because AI Overview exposure is wildly uneven by query type. Three factors determine yours:

• Intent mix. Informational queries — definitions, how-tos, comparisons — attract AI Overviews at far higher rates than transactional and navigational ones. A site living on informational traffic is structurally more exposed than a store.
• Query length and specificity. Longer, conversational queries are more likely to trigger a generated answer, which is precisely the long-tail territory many content programmes were built on.
• Answer completeness. If your page’s value is a fact that fits in two sentences, an AI Overview substitutes for it entirely. If your value is a tool, a dataset, a community or a decision that needs your interface, it cannot.

You can measure this in an afternoon. Export your top 200 queries by impressions from Search Console for the last 16 months, tag each by intent, check AI Overview presence for a sample of each group logged out, then compare CTR year over year within each group. If informational CTR fell while transactional held, you have confirmed AI Overview exposure rather than a general ranking problem. A decline concentrated at positions one to three is a particularly strong signature, because that is where displacement bites.

If that pattern does not appear, your decline has a different cause — an algorithm update, a technical regression, seasonality, or a competitor. Establishing that distinction is worth doing before you rebuild a content strategy around the wrong villain.

What to actually do about it

The practical move is to stop asking “how much did AI Overviews cost the industry” and start asking “which of my pages are substitutable by a paragraph.” Those are the pages bleeding, and they are usually a minority of the site producing a majority of the loss.

For that minority, three responses work:

1. Move up the value chain. Stop competing on facts that fit in a paragraph. Compete on judgement, original data, tooling and decisions that require your product.
2. Optimise to be cited rather than to rank below. Being the source inside the overview is worth more than being first underneath it. That inverts the usual target and is what generative engine optimization is for.
3. Measure across engines, not just Google. If a meaningful share of your category’s questions now go straight to an assistant, Google analytics will never show you the gap. Tracked prompts will.

There is also a genuine opportunity hiding in the data. High-profile AI answer errors have made a segment of users sceptical of generated summaries, and some scroll past them deliberately. Being the cited source under a summary users distrust is worth more than it looks — the sceptical clicker is a high-intent clicker.

Sources

• Pew Research Center — study of approximately 68,000 real search queries; clicks fell from 15% to 8% of visits when an AI summary was present.
• Ahrefs — CTR impact of AI Overviews; 34.5% decline for position one (April 2025), approximately 58% (February 2026).
• Google — 2.5 billion monthly active users for AI Overviews and 1 billion for AI Mode, stated 19 May 2026.
• Moz (Tom Capper) and Similarweb — early AI Mode adoption at roughly 1–2% of US search traffic; Nick Fox cited 75 million AI Mode users in December 2025.
• Publisher traffic reporting, H1 2025 — median publisher down ~10% year over year; non-news content sites down ~14%.

Frequently asked questions