AI Overview traffic decline: seven claims, four hold up

Ask how much AI Overviews have cut into search traffic and the honest answer is: it depends which of at least seven claims you mean, and they were not all measuring the same thing. We checked each one against its original source rather than repeating whichever number showed up first in a search. Four trace to distinct, credible, dated studies with real methodology behind them. Three did not hold up as independent measurements and are excluded below rather than repeated.

The figures that check out#

A randomized field experiment: outbound clicks down 39.8%. Saharsh Agarwal (Indian School of Business) and Ananya Sen (Carnegie Mellon University’s Heinz College) built a Chrome extension that randomly assigned 1,065 US-based participants to see AI Overviews or have them hidden, across 68,089 searches. Conditional on an AI Overview appearing, outbound organic clicks fell 39.8% and zero-click searches rose 34.5%, with no measurable gain in reported satisfaction or information quality. The paper was first posted to SSRN on 3 April 2026 and last revised 18 June 2026. It is the only figure in this report built on random assignment rather than a before-and-after comparison, which is what lets the authors call the effect causal rather than correlational.

An agency’s client data: organic click-through rate down 61%. Seer Interactive tracks 3,119 informational search terms across 42 client organizations, pulling Google Search Console and Google Ads data alongside third-party AI Overview presence tracking. Its September 2025 update found organic click-through rate on queries where an AI Overview appeared fell from 1.76% in June 2024 to 0.61%, which Seer reports as a 61% decline (the two endpoints alone work out to about 65%), against 25.1 million tracked organic impressions. This is not a random sample of the web; it is Seer’s own client accounts, weighted toward the query types most exposed to AI Overviews in the first place.

Industry-wide referral data: publisher traffic down about a third. The Reuters Institute for the Study of Journalism’s Journalism and Technology Trends and Predictions 2026 report, published 12 January 2026, cites Chartbeat data across more than 2,500 publisher sites showing global Google search referral traffic down roughly a third year over year to November 2025. A separate Chartbeat cut reported by Axios in March 2026 found the decline was uneven by publisher size: about 22% for large publishers, 47% for medium ones, and 60% for small ones, over two years. Neither figure isolates AI Overviews from the rest of how Google’s results page has changed; both describe referral traffic broadly.

One company’s own disclosed number: revenue down 24%. Chegg reported fourth-quarter 2024 revenue of $143.5 million, down 24% year over year, in its own 24 February 2025 earnings release, and has since sued Google, naming AI Overviews as a cause of its traffic and revenue decline. This is the only figure in the set that is a company’s own financial disclosure rather than a third-party traffic estimate, and it measures revenue, not clicks or visits.

Figures we could not verify#

“Business Insider down 55%, April 2022 to April 2025” does not trace cleanly to that number. The underlying reporting is real: the Wall Street Journal, via Similarweb data, reported in June 2025 that organic search traffic to Business Insider, HuffPost and the Washington Post was down roughly 50% over three years, and that the New York Times’ organic-search traffic share fell from 44% to 36.5% over the same period. The specific “55%, April 2022 to April 2025” framing that circulates in SEO blogs is a more precise restatement than the original reporting supports, so this report uses the traceable ~50%-over-three-years figure instead and does not repeat the sharper number.

CNN’s decline is Similarweb-via-press, not a named study. The New York Post reported in July 2025, citing Similarweb, that CNN.com traffic fell 28% year over year to 311.7 million visits. That figure is real and traceable. The wider “27 to 38%” range attached to CNN elsewhere in SEO content compiles several different comparison windows into one number without naming which is which, so this report cites the one figure with a clear source and period (28%, year over year to mid-2025) and drops the compiled range.

HubSpot’s 80% is an outside estimate, not HubSpot’s own figure. HubSpot’s own July 2026 blog post says a few companies posted data suggesting it lost 80% of its blog traffic. It does not confirm that number. It says traffic fell partly by design, because of a multi-year shift away from blog pageviews toward YouTube, podcasts and newsletters, and that no-click searches have also cannibalized search traffic. This report includes the claim because HubSpot’s own response is traceable, but treats it as the least reliable figure here: an outside estimate the subject does not confirm.

Why they do not compare#

They measure different metrics. A click count, a click-through rate, total site visits, and company revenue are four different numbers that can move independently. Click-through rate can fall even when raw clicks hold steady, if impressions rise faster; revenue can fall for reasons that have nothing to do with search, including layoffs, pricing changes or a shrinking market. Reading any one of these as a stand-in for “traffic” erases what it actually counted.

They cover different query sets and site types. The SSRN experiment used a general panel of US searchers across whatever queries they ran. Seer Interactive’s client accounts are weighted toward informational queries, the category most exposed to AI Overviews. Chegg is a single company in education technology; CNN and Business Insider are general news. A 39.8% average across all query types and an 80% estimate for one company’s blog are not describing the same population, even before either number’s own uncertainty is considered.

They cover different, overlapping time windows. The SSRN data was collected in early 2026; Seer’s comparison starts in June 2024; the Reuters Institute figure is year-over-year to November 2025; Chegg’s is a single quarter against the same quarter a year earlier. AI Overviews expanded in reach and query coverage across all of this period, so a study run later is measuring a more mature, further-rolled-out feature than one run in 2024, independent of any difference in method.

Why we care#

A single “AI Overviews cut traffic by X%” headline number is not a complete claim. Whether the right figure for a forecast is 24%, 39.8%, 61% or something closer to 80% depends entirely on whether you are modeling clicks, click-through rate, total visits or revenue, for what kind of query, over what window, and that dependency is exactly what most quoted versions of this stat strip out.

The rigor of the source matters more than the size of the number. A randomized experiment with a modest 39.8% effect is stronger evidence than an unverified 80% estimate the subject company disputes. Before using any AI Overview traffic figure in a deck or a forecast, trace it to its original study, note whether the design was randomized or observational, and check that its metric matches the one you actually need.