AI Overviews and the 68% Problem: What the Click Data Shows

68.0% of US Google searches now end without a click. In 2024 that figure was 60.45%. In the first four months of 2026 it was 68.01% – a rise of 7.6 points in two years, and AI Overviews are the main reason.

Seven points sounds survivable. It is the wrong number to look at, and this piece explains why, using click data rather than opinion.

Look at the clicks you get, not the ones you lose

What matters to a business is not the share of searches that end without a click. It is the share that ends with one.

That fell from 39.55% to 31.99%.

Read as a decline, that is 19.1% fewer clicks sent anywhere. Roughly one click in five, gone, in two years.

The framing genuinely changes the conversation. "Zero-click went from 60% to 68%" sounds like a drift. "A fifth of your click supply disappeared" is the same fact, and it is the version that belongs in front of a board.

Why AI Overviews cause it: the Pew evidence

Panel data tells you what happened. It does not tell you why. A Pew Research Center study using real browsing data does.

Across 68,879 searches by 900 US adults, a results page carrying an AI summary produced a click on a normal result 8% of the time. Without a summary, it was 15%.

That is a 46.7% reduction in click-through when an AI Overview is present.

Now the number that undercuts the usual reassurance. Clicks on the sources inside the summary ran at 1%.

The trade being offered to publishers is: you lose the click, but you gain the citation. At a 1% source-click rate, that trade does not pay in traffic. It pays in brand exposure, which is real and worth something, but it is not a traffic channel and should not be budgeted as one. The difference between being quoted and being visited is worth understanding properly, and why citation share is not traffic sets out the three architectures behind it.

The number nobody quotes

Here is a figure that cuts the other way, included because a piece that only reports the alarming numbers is not analysis.

Google's AI Mode reports one billion monthly users. In the same panel period, only 0.34% of searches actually moved into it.

Exposure and use are different things. A billion people being shown a product is not a billion people using one, and most AI adoption statistics live in that gap. Apply the scepticism in both directions: to the vendor claiming a billion users, and to the agency claiming AI has eaten search.

There is also something else the missing clicks did. The share of searches leading to another Google search rose 7.6 points over the same period. People are not giving up. They are refining, inside Google, without leaving.

How AI Overviews change what ranking means

The old results page was an index. Its job was to send you elsewhere.

The new one is a destination. Its job is to end the session, and it is doing that well.

That one change flows into everything downstream:

• Ranking used to mean being picked from a list. It now means being summarised, often without a visible link.
• Search volume used to imply traffic waiting to be captured. Two thirds of those searches now answer themselves.
• Position one used to be the prize. There is no position two inside a paragraph.

Google's own guidance is worth reading here, because it is more useful than most commentary about it. Google Search Central's documentation on AI features states plainly that there are no special requirements or optimisations for appearing in AI Overviews or AI Mode: normal indexing, snippet eligibility and helpful content rules apply, and traffic from AI features appears in Search Console under the standard "Web" search type. The structural consequences of that, and what breaks in a conventional programme, are covered in the five shifts AI Overviews force on a search programme.

The damage is surgical, not general

If AI Overviews had shaved a flat percentage off everything, this would be a budgeting problem. They have not.

I took six keywords and sorted them into two groups before pulling any trend data. That order matters: choosing the groups afterwards would prove nothing.

• Group A: questions an assistant can finish inside the tool you are already in. A syntax lookup, a formula, a definition.
• Group B: physical, local and comparison questions, where the answer is not the deliverable.

Group A fell 64.2% from its peak. Group B fell 4.5%. Same index, same window, same scoring rule. A 14.2-fold difference.

The steepest single fall was 74.7%, and no competitor was involved. Nobody out-ranked that page. The question simply stopped being asked.

Six keywords describe a mechanism, not a market. I would not build a business case on them, and neither should you. But you can test the same mechanism on your own data in an afternoon.

Why Google is fine and publishers are not

One last piece of context, because without it this reads as search collapsing, and search is not collapsing.

In the same window that clicks to the open web fell 19.1%, Google's search revenue grew 17% year on year. Alphabet also reported its model APIs processing 22 billion tokens a minute, up 37.5% in a single quarter.

Both facts are true at once, and the reason is structural. Google is not losing money because people stopped clicking. It is making more because they stopped, since an answer on the results page keeps the session, the ad slot and the data in one place.

The engine and the open web were never the same business. They were a bargain: the engine got attention, the web got traffic. Generative answers let one side keep its half while quietly ending the other. There is no version of this where the traffic returns because Google decides to send it.

What to do about AI Overviews on Monday

AI Overviews are not something you can opt out of without also opting out of Google, so the work is adjusting what you measure and where you invest. Four steps, none of which need new software.

1. Re-baseline on clicks, not impressions. Impressions and average position have held up. Clicks have not. Put clicks per search in front of your leadership before someone else puts sessions there.

2. Sort your top 200 queries by one question. Can a model finish this user's task without them leaving the tool they are in? Yes goes in the forecast-a-decline pile. No is where your traffic still lives in 2028.

3. Stop counting citations as traffic. At a 1% source-click rate, a citation is a brand impression. Report it as one.

4. Check what your tools can still see. Since late August 2026, Google routes result clicks through a redirect that blocks results-page scraping. Rankings are unchanged; what rank trackers observe is not. Clickstream panels are unaffected. Those two get quoted in the same slide constantly.

Questions people ask about AI Overviews

These are the questions searched most often alongside AI Overviews, answered from the click data above.

What are AI Overviews? They are AI-generated summaries that appear at the top of Google results for many queries, assembled from web sources and shown instead of, or above, the traditional list of links.

How do Google AI Overviews work? Google fans a query out into related subtopics, retrieves pages that are already indexed and snippet-eligible, and assembles an answer from them. Google states there are no separate requirements or markup to appear.

How do I rank in AI Overviews? There is no separate ranking system. Per Google's own documentation, standard indexing and helpful-content rules apply. What the citation data adds is a format instruction: dense, canonical, checkable pages get quoted repeatedly, and pages that publish real figures get quoted more than pages that gate them.

How do I track AI Overviews? Traffic from AI features appears in Search Console's Performance report under the "Web" search type, with no separate breakdown. Third-party trackers scrape the results page, which the goto redirect now interferes with, so check which method your tool uses.

Do AI Overviews reduce clicks? Yes, measurably. In Pew's browsing data, click-through on a traditional result fell from 15% to 8% when a summary was present – a 46.7% reduction. Clicks on sources inside the summary ran at 1%.

---

Zero-click figures: SparkToro analysis of Similarweb desktop and mobile panel data; 2026 covers January to April, and searches inside Google's mobile app are excluded. Click-through figures: Pew Research Center, 68,879 searches by 900 US adults. Cohort figures: Ahrefs volume history pulled live on 7 September 2026. Full method in the report, How AI Changed the Way People Search.