This piece looks at four studies of whether Google's AI summaries cost clicks: Pew, Ahrefs, Seer and Semrush. They do not measure the same thing, so their numbers should not be set side by side without care. Here is what each one did.
Pew: real browsing behavior
Pew analyzed browsing data from 900 U.S. adults who agreed to share their online activity.1 For each respondent it received a list of all URLs visited on a tracked device during March 1-31, 2025.1 The total dataset contains 68,879 unique Google searches.1 Some 12,593 of these searches produced an AI summary.1
Pew re-ran each search through a third-party scraping service, with results collected April 7-17, 2025.1 That means the summary it analyzed is how the page looked weeks later, not necessarily what the person saw. Pew says the AI summary for a given search may change over time.1
Users who saw an AI summary clicked a traditional result in 8% of visits.1 Those who did not see one clicked in 15% of visits.1 Clicks on a link inside the summary happened in 1% of visits to pages with one.1 Users ended their browsing session on 26% of pages with a summary, against 16% of pages without.1
What it can show: individual behavior, including what people did next. What it cannot show: anything beyond Google, since the analysis includes only Google searches.1 Searches with a summary also differ from those without. Longer searches and questions produce summaries more often.1
Ahrefs: a before-and-after estimate
Ahrefs compared 150,000 keywords with an AI Overview present against 150,000 keywords with informational intent and no AI Overview present.2 It used aggregated Search Console data for each keyword's average desktop CTR per month, comparing March 2024 (before the US rollout) with March 2025.2
For the AI Overview keywords, position one CTR fell from 0.073 to 0.026.2 For informational keywords generally, it fell from 0.056 to 0.031.2 Ahrefs explains the method: "We can use the drop in CTR for informational keywords to predict the CTR for our sample of AI Overview keywords, assuming AI Overviews had not rolled out."2 Comparing the actual figure with that forecast suggests a reduction of about 34.5% for position one.2
What it can show: a modeled estimate for position one on desktop. What it cannot show: anything about lower positions, mobile, or a measured counterfactual. The "without AI Overviews" line is a calculation. Ahrefs describes its headline result as a correlation with a lower average CTR.2
Seer: client data and a causation warning
Seer's update analyzed informational and educational queries from June 2024 through September 2025.3 The set was 3,119 search terms across 42 client organizations, with 25.1M organic impressions and 1.1M paid impressions.3 Data came from Google Search Console and Google Ads, among other sources.3
In Q3 2025, organic CTR was 0.52% where an AI Overview appeared and the brand was not cited.3 It was 0.70% where the brand was cited, and 1.45% where no AI Overview appeared.3 For queries with AIOs, organic CTR fell 61% from June 2024 to September 2025.3
The most useful part is a correction. Queries that triggered an AI Overview in September 2025 showed lower CTRs than non-AIO queries in every month of the range.3 In June 2024 the organic figures were 1.76% with AIOs and 2.74% without.3 Seer reads this as AIOs surfacing on query types that were already generating fewer clicks.3 It also found that organic CTR for non-AIO queries fell 41% over the period.3
Seer categorizes queries by September 2025 AIO presence and calls this "a retrospective cohort analysis, not a month-by-month correlation study."3 It also says it cannot definitively prove that citation causes higher CTRs.3 Seer says the paid impression sample is significantly smaller than the organic one.3
Semrush: same keywords, before and after
Semrush worked with Datos, a Semrush company, on over 200,000 keywords with search volume over 100, between January 2025 and October 2025.4 It also tracked keywords that did not trigger AI Overviews in May but did in October.4
For those same keywords, the zero-click rate decreased from 33.75% to 31.53%.4 Semrush concluded that AI Overviews don't automatically increase zero-click behavior.4 Keywords with AI Overviews do tend to have higher zero-click rates on average, which Semrush says may reflect informational intent rather than causation.4
What it can show: a within-keyword comparison, the closest of the four to a controlled test. What it cannot show: clicks to any one result. Zero-click rate is a different metric from CTR. Semrush also reported that the share of AIO-triggering keywords with commercial, transactional or navigational intent grew, which changes the mix of keywords over time.4
Why the numbers differ
| Study | Unit | Metric | Period |
|---|---|---|---|
| Pew | 900 adults, 68,879 searches | Next action after a search | March 2025 |
| Ahrefs | 300,000 keywords | Position one CTR, desktop | March 2024 vs March 2025 |
| Seer | 3,119 terms, 42 clients | Organic and paid CTR | June 2024 to September 2025 |
| Semrush | 200,000+ keywords | Zero-click rate | January to October 2025 |
Three differences matter most, and the table above shows each of them: the metric a study counts, the comparison it makes, and who is in its sample. Read each number with those three in mind.
The queries that trigger summaries are also not typical. Pew found that 60% of searches beginning with question words produced a summary, while just 8% of one- or two-word searches did.1 Seer's reading that AIOs surface on query types already generating fewer clicks points the same way.3 Any raw gap between "with" and "without" mixes the effect of the summary with the type of query.
My read
Pew found fewer clicks on result links when a summary is present, 8% against 15% of visits.1 Ahrefs concludes that "the presence of AI Overviews reduces the click-through rate for position 1 by ~34.5%".2 Both are consistent with a real drop. Seer says AIOs surface on query types that were already generating fewer clicks.3 Semrush's same-keyword test found the zero-click rate decreased from 33.75% to 31.53%.4
I would not pick one number. I would treat the range as the honest answer, and check your own Search Console data by query type before and after a summary appears. Compare like with like: same queries, same device, same period.
Comments 0
No comments yet.
Sign in with GitHub or Google to comment. Reading needs no account.