Most of what gets written about AI Overviews traffic is anecdote: one publisher’s graph, one screenshot of a summary that paraphrases somebody’s article. Anecdotes are easy to find and impossible to generalise from. The most useful independent measurement so far comes from a browsing panel, not a publisher, and it says something narrower and more actionable than “search is dying”.
The short version: summaries cut clicks hard on the searches where they appear, they appear on a minority of searches, and the links inside them barely get clicked at all. Each of those three facts changes what you should do about it.
What the Pew study measured
Pew Research Center tracked the browsing of 900 US adults for the whole of March 2025 and published the results on 22 July 2025. The panel ran 68,879 unique Google searches, and the researchers recorded what each results page showed and what the person did next. You can read the method in Pew’s write-up of the study.
| Measure (Pew, March 2025) | With an AI summary | Without one |
|---|---|---|
| Share of searches | About 18% (12,593 of 68,879) | About 82% |
| Visits with a click on a traditional result | 8% | 15% |
| Visits with a click on a link inside the summary | 1% | n/a |
| Pages where the browsing session ended | 26% | 16% |
Read the third row twice. The citation links that Google points to as its answer for publishers were clicked in 1% of visits. Being cited in a summary is not a traffic strategy; on these numbers it is closer to a brand impression.
Why the “AI Overviews killed SEO” reading is wrong
The headline number is a halving of clicks, and it is tempting to apply it to your whole site. That is the mistake. The halving applies to the roughly 18% of searches that triggered a summary in Pew’s sample. On the other four-fifths, the click behaviour looked like ordinary search.
So the AI Overview traffic drop is not a flat tax on organic search. It is a steep tax on a particular slice of queries, and whether your site feels it depends almost entirely on how much of your traffic sits in that slice. Two sites in the same niche can have completely different exposure.
The session-ending figure matters too. People ended their browsing on 26% of pages with a summary against 16% without. For a large share of those searches, the summary was the whole visit. Nobody was going to click through to read your 2,000 words on the same question.
Which sites carry the most AI Overviews traffic risk
Summaries do best on questions with a settled, short answer. Definitions, how-to steps, quick facts and simple explanations are all easy to compress into a paragraph. Pages that exist to answer exactly those questions are the ones whose clicks get absorbed.
Queries where the reader wants to compare, judge or buy are harder to compress. A summary can list three products; it cannot replace a hands-on comparison with prices, photos and a verdict the reader trusts. Those pages still need the click.
- High exposure: glossary pages, “what is” explainers, short how-tos, fact lookups.
- Moderate exposure: longer guides where the summary takes the first answer but not the detail.
- Lower exposure: reviews, comparisons, tools, local and transactional pages.
To see how this plays out across a niche rather than your own site, use the Google updates tracker alongside the explorer. In the explorer, filter to one niche, split the results by content type (informational blogs against review and comparison sites), and compare their before and after traffic across the most recent confirmed updates. That shows you which model is holding up in your niche, using estimated traffic rather than a guess.
How to track AI Overview traffic when no report shows it
There is no AI Overview filter in Search Console, and there is no setting in GA4 that isolates it either. Anyone selling you a dashboard that claims to read it directly from Google’s own data is overstating what that data contains.
What you can see is the signature. When a summary starts appearing on a query you rank for, impressions tend to hold steady, because you still appear on the page, while click-through rate falls. That pattern on informational queries, with your position roughly unchanged, is the closest thing to a measurement you will get.
- In Search Console’s performance report, compare two equal periods with clicks, impressions, CTR and position switched on.
- Filter queries by question words (“what”, “how”, “why”) to isolate the informational set.
- Flag queries where impressions held, position held and CTR fell. That is the likely summary group.
- Check those queries by hand in a clean browser to confirm a summary is showing.
Keep the flagged list in a spreadsheet and re-run it monthly. A single comparison can be distorted by seasonality or a ranking wobble; the same queries showing the same pattern three months running is a much stronger signal. Group them by page, too, because the decision you eventually make is about pages, not individual queries.
Your analytics referral data helps at the edges. Some visits from AI features arrive with distinguishable referrers, but treat them as a small, partial count rather than the total. The bigger effect is the click that never happened, and no referral report can record that.
What to change, and what not to
Do not rewrite your whole site for summaries. Most of your queries probably do not trigger one, and chasing citations in a box that converts at 1% is poor use of an editorial budget. Put the effort where clicks still happen.
In practice that means shifting new content towards comparisons, reviews and first-hand testing, and treating short informational pages as support for those rather than as traffic earners in their own right. It also means valuing traffic by what it does, not by its volume: a smaller number of commercial visits often earns more than a large informational audience.
The limitation is worth stating plainly. Pew measured US adults over a single month in early 2025, through a browsing panel. It tells you how people behave when a summary appears; it cannot tell you how often summaries appear on your queries today, or what share of your traffic sits in that group. Only your own Search Console data answers that.
More on how algorithm shifts show up in traffic sits in the algorithm archive and across the wider blog.
Frequently asked questions
How much do AI Overviews reduce clicks?
In Pew Research Center’s March 2025 study, people clicked a traditional result in 8% of visits when an AI summary appeared and 15% when it did not. Links inside the summary were clicked in 1% of visits.
Can I see AI Overview traffic in Search Console?
No. Search Console has no AI Overview filter and GA4 has no setting for it. Look for informational queries where impressions and position held while CTR fell, then check those queries by hand.
Which pages are safest from AI Overviews?
Pages the reader needs to click to get value from: reviews, comparisons, tools and transactional pages. Short definitions and quick how-tos are the easiest content for a summary to replace.
The takeaway AI summaries roughly halve clicks on the searches where they appear, and almost nobody clicks the cited links. Measure your exposure through CTR on informational queries, then build where the reader still has to click.
