AI Overviews and Traffic Loss Claims

What changed in the data, and when

This is the most important fact on the page and it is recent enough that most published commentary predates it. On 3 June 2026 Google introduced generative AI performance reports in Search Console — one for Search and one for Discover. Before that date there was no first-party reporting of AI surfaces at all. AI Overview impressions and clicks were folded into ordinary web-search totals with no way to segment them.

That produces a hard line through every matter in this category. For any damages period ending before roughly mid-2026, no first-party Google data segments AI Overview exposure. Analysis of that period rests entirely on third-party SERP-tracking data, with the foundation questions that carries: who captured it, from what location, on what device, at what frequency, and whether the capture was contemporaneous or reconstructed. That is a predictable line of cross-examination, and it belongs in the report rather than in a deposition.

Note also what Google's ordinary performance definitions do with AI surfaces. Clicking a link to an external page in AI Mode counts as a click. An AI Overview occupies a single position, and every link inside it shares that position value — so position data for a query returning an AI Overview does not mean what it means elsewhere.

What the generative AI report contains, and what it does not

Read the specification carefully, because four of its properties are dispositive of what an opinion can say.

The report offers impressions only — how many times links to the site were shown to a user in a generative AI feature. There are no clicks, no click-through rate, no average position. Dimensions are pages, aggregated by canonical URL after redirects; countries; dates, daily, weekly or monthly in Pacific Time; and devices. Impressions are deduplicated: two results from the same site in one generative AI feature count once. Search Labs data is excluded, the thousand-row table limit applies, the newest data may be preliminary, and not all properties have access to the report at all.

The four consequences:

1. AI Overviews and AI Mode are combined. A claim attributing loss specifically to AI Overviews cannot be isolated from AI Mode using Search Console.
2. There is no query dimension. You cannot determine from Search Console which queries produced AI impressions.
3. There is no click metric. An AI-feature click-through rate cannot be computed from first-party Google data. Any figure presented as one came from somewhere else.
4. The data is additive, not carved out. Google states these impressions are also counted in the main Performance report. A before-and-after model that subtracts AI impressions from total impressions is double-counting.

That fourth point is the one I would check first in any opposing analysis, because the error is easy to make and it is invisible in a summary chart.

The public statistics, and how to use them without overstating

Three measurements have published methodologies. Everything else circulating on this subject should be treated as unsupported until the underlying study is produced.

Zero-click share. SparkToro's analysis of Similarweb clickstream panel data found that 68.01% of US Google searches ended without a click between January and April 2026. The caveats travel with the number. It is a desktop and mobile web browser panel, US only, for that four-month window; panel weighting assumed a two-to-one mobile-to-desktop ratio; sample size was not disclosed. Most importantly, the panel excludes the Google mobile app, covering only mobile searches made in a browser — SparkToro's own view is that the true rate is therefore likely higher, not lower. SparkToro also states that comparisons against earlier years are not directly equivalent, because the underlying data providers changed, so this figure should be presented as a level for a stated window and not as a measured change over time.

Click-through impact. The citable figure is Ahrefs': AI Overviews are associated with a 34.5% lower click-through rate for the top-ranking organic result. Methodology: 300,000 informational-intent keywords, half with AI Overviews and half without, using aggregated Search Console desktop click-through at position one, comparing March 2024 with March 2025, published 17 April 2025. Ahrefs states the analysis is correlational rather than causal, that Search Console does not expose AI Overview click-through directly, and that it expects the figure to be the highest the rate will be. Use it with those caveats intact.

Independent corroboration. Pew Research Center found that in Google visits where an AI summary appeared, users clicked a traditional result link in 8% of visits, against 15% where none appeared; only 1% clicked a link inside the summary. The panel was 900 US adults sharing browsing activity through March 2025, covering 68,879 searches, 12,593 of which produced an AI summary. Google has publicly disputed the study, and a report is stronger for saying so.

Google's contrary position, and why both accounts can be true

Google's public position, stated by the head of Search on 6 August 2025, is that total organic click volume from Google Search to websites has been relatively stable year over year, that average click quality has increased — a quality click being one where users do not quickly click back — and that this contrasts with third-party reports which, in Google's characterization, inaccurately suggest dramatic aggregate declines based on flawed methodologies or isolated examples.

A report that presents only the panel studies, or only Google's statement, will be handed the other one. The useful observation is that they are measuring different quantities and are not strictly inconsistent. Pew and the clickstream panels measure per-search click behavior on samples of users. Google reports aggregate click volume across all queries and all sites. Aggregate stability is entirely consistent with severe redistribution — some sites gaining while others lose — and with per-search click rates falling while total query volume rises.

Which leads to the point that matters in a case: none of these statistics say anything about the claimant. They describe a market. Presenting a market-level figure as though it measured a party's loss is the error most likely to draw a challenge, and it is avoidable.

The pending litigation, and what it is and is not

Several public filings frame this dispute, and they are worth knowing because opposing counsel will know them. They are pleadings, not findings, and every allegation in them remains an allegation.

Chegg, Inc. v. Google LLC and Alphabet Inc., No. 1:25-cv-00543-APM (D.D.C.), was filed 24 February 2025. Its counts are reciprocal dealing under Sherman Act § 1, reciprocal dealing under § 2, monopoly maintenance under § 2, and common-law unjust enrichment. The theory is that Google conditions access to search referral traffic on publishers supplying content for AI training, featured snippets and retrieval-augmented generation without compensation.

Penske Media Corporation v. Google LLC, No. 1:25-cv-03192 (D.D.C.), was filed 12 September 2025 by Penske and its subsidiary publishers, asserting Sherman Act claims and unjust enrichment. Two allegations bear on methodology: that from late 2024 into early 2025 the share of searches returning both links to its sites and an AI Overview rose to roughly 20%, and that opting out through Google-Extended does not exclude a publisher from AI Overviews, so exclusion is practically unavailable short of leaving the index. A second, separate Penske action, No. 1:26-cv-00225 (S.D.N.Y.), concerns advertising technology — “Penske v. Google” is now ambiguous and should be cited by docket.

In Europe, the European Publishers Council has filed a formal antitrust complaint over AI Overviews and AI Mode, and the European Commission was reported on 9 December 2025 to have opened an investigation into Google's use of publisher content for AI products without appropriate compensation.

The isolation problem

This is the core of the page. A claim that AI answer features caused a specific traffic loss has to survive at least six confounds, and they all run concurrently.

• Core and spam updates. Google's Search Status Dashboard records each with a start date and duration, and the record shows updates running in essentially every quarter in which AI features expanded. There is no clean control period, which is a stronger statement than “this is difficult.”
• Result-page layout changes unrelated to AI — shopping modules, expansion of related-question blocks, video carousels, and sponsored-result density.
• Query mix shift. Users phrase searches differently when they expect a generated answer, and longer conversational queries change which pages are eligible at all.
• Substitution to assistants entirely off Google, which leaves no trace in Google's data.
• Seasonality and category demand. Google's own traffic-drop guidance names seasonality as a category and directs analysts to view the full sixteen-month window specifically to check whether the drop recurs annually.
• The party's own changes — migrations, redesigns, paywalls, content reductions, and paid media shifts.

There is a preservation consequence too. Search Console retains sixteen months of performance data. Where the “before” period predates that window at the time of analysis, the first-party baseline no longer exists unless it was exported contemporaneously, and no request to Google will recover it.

What an expert can and cannot defensibly do

The available methods are real but narrower than the claims usually are.

Defensible. Segment the claimant's own query set into queries that do and do not trigger AI Overviews, using SERP-tracking data with dated snapshots, and compare the two cohorts' click trajectories. That is a difference-in-differences design and it is the strongest approach available, provided the cohorts are matched on intent and prior volume and the matching is disclosed. Use the generative AI report's page-level impressions, for periods after June 2026, to identify which of the party's URLs appear in AI surfaces at all. Overlay dated update windows on the traffic series and state whether the inflection aligns with an update or with AI rollout. Benchmark against a set of comparable sites where one can be constructed.

Not defensible on current data. Computing an AI Overview click-through rate from first-party Google data. Separating AI Overview effects from AI Mode effects using Search Console. Establishing query-level AI presence from Search Console. Applying a market-wide percentage to a claimant's revenue and calling the product a damages figure.

That last one deserves emphasis because it is the most common. A published market average multiplied by a party's traffic is arithmetic, not evidence of what happened to that party. In my experience an opinion built that way does not survive a serious challenge, and the underlying claim may have been provable by the cohort method instead. The primary references are Google's generative AI performance report documentation, its ranking update history, and the Pew study.

Frequently Asked Questions

Published 2026-08-11