AI Overview Traffic
- Search Console reports AI-feature impressions, not AI-feature clicks.
- AI Overview traffic is an estimate: the change in clicks on affected pages, judged against comparable pages.
- Judge the change against comparable pages over the same period — the shared decline is not the page's own loss.
Consensus definition
In practice the term covers two things that are often mixed up: the clicks a site receives from links shown inside an AI Overview or in AI Mode, and the clicks a site loses on queries where an AI answer sits above the classic results. Google states that sites appearing in AI features are included in the overall search traffic in Search Console and reported in the Performance report within the "Web" search type1 — there is no separate search type for them. A Generative AI performance report, announced on 3 June 2026 and rolled out to all websites worldwide by 31 August 2026, adds a dedicated view; the information it lists is impressions by page, country, device and date, and the list contains no clicks, queries or click-through rate2. Google mentions "adding additional metrics over time" as a possibility, not as a commitment2. AI Overview clicks are therefore not observable as a separate number; they can only be estimated. Independent studies describe the click side from outside Search Console. In a browsing panel of 900 US adults in March 2025, Pew Research Center found that users clicked a traditional result on 8 % of visits to search pages with an AI summary and on 15 % of visits without one, and clicked a link inside the summary on 1 %3. A Seer Interactive analysis of Search Console data on 3,119 informational queries measured an organic click-through rate in Q3 2025 of 1.45 % where no AI Overview appeared, against 0.52 % where one appeared without citing the site and 0.70 % where it cited the site4. For Google search in the US, SparkToro's clickstream estimate puts searches that end without a click at 68.01 % for January to April 2026, up from 60.45 % in 2024, with the caveat that the panel provider changed between the two measurements5.
rhinegold operator refinement
Rhinegold's refinement: AI Overview traffic is an estimate, and an estimate only means something against a comparison. Clicks and click-through rates on queries with AI answers move for reasons that every site in the market shares — more AI answers, more zero-click layouts, seasonal demand. A before/after on one page books all of that as the page's loss. In our own comparisons of treated and untreated pages, both groups lost clicks over the same year; the difference between them, not the level, is what we read. The estimate therefore runs in three steps: identify the pages whose links are shown in AI features (the report's impressions); track clicks and click-through rate on those pages and their queries over time; and compare that change with comparable pages of similar topic, intent and baseline traffic that show few or no AI-feature impressions, over the same weeks. This is difference-in-differences applied to search data, with the comparison pages standing in for the counterfactual. It assumes that, without AI answers, affected and comparison pages would have moved alike; the placebo test checks that. The market-wide change is shared; only the difference is yours. That difference is the change beyond what comparable pages also experienced — the incremental effect of higher AI exposure, not the total effect of AI answers.
Operational use
Start from the Generative AI performance report to list the pages with AI-feature impressions, then pull clicks and impressions for the same pages from the Performance report. Build a comparison set before looking at results: pages matched on topic, search intent and pre-period traffic, with little AI-feature exposure. Compare the change in clicks between the two groups across the same window; read click-through rate only next to impressions, because impressions in AI features count in the denominator, and state whether changes are compared as ratios or in percentage points. Check the setup with a placebo test on a period in which nothing changed. Report the result as a range with its comparison set named, not as a single loss figure. Read it next to LLM referral traffic, which covers visits from chat assistants outside Google.
Measurement boundary
Clicks from links inside an AI Overview cannot be separated from clicks on the classic results of the same query: both are counted in the "Web" search type1. The Generative AI report lists impressions by page, not by query, and its listed dimensions do not split AI Overviews from AI Mode2; query-level exposure needs third-party results-page tracking, which samples a query set. Pages whose queries show an AI answer that does not link them have no impressions in the report; without results-page tracking they end up in the comparison group and shrink the difference. Comparison pages may carry AI answers too: where AI Overviews also appear on their queries, the measured difference usually shrinks; it can also grow when AI Overviews change which queries remain on the comparison pages (see Spillover & Contamination). The Pew and SparkToro figures cover US users, and the three studies use different data bases (browsing panel, clickstream, Search Console), so their figures are neither comparable with each other nor transferable to other markets. None of these methods measures what a user took from an answer without clicking.
Distinct from
From zero-click search: zero-click is the market-wide outcome pattern; AI Overview traffic is the page-level estimate of what that pattern added to or removed from one site. From LLM referral traffic: chat assistants such as ChatGPT or Perplexity send visits from their own domains, which analytics can partly identify by referrer, while AI Overview and AI Mode clicks arrive as ordinary Google organic traffic. From AI Overviews vs AI Mode: that entry separates the two surfaces; for traffic measurement Google reports them together, so the estimate covers both. From AI Mode: the surface itself, not the traffic it generates.
Why a before/after can overstate the effect
Common mistakes
- Reading the "Web" totals in Search Console as organic traffic without AI influence — AI Overview and AI Mode clicks are inside them.
- Treating the Generative AI report as a traffic report: it shows impressions, not clicks or queries.
- Calculating AI Overview loss as a before/after on affected pages, which books the market-wide decline as the page's own loss.
- Applying published click-through rates from panel or vendor studies to one's own pages instead of measuring against comparable pages.
- Choosing the comparison pages after seeing the results, which makes the estimate say what was expected.
Where consensus is missing
There is no standard method for estimating AI Overview clicks. Published studies use user panels, clickstream data or Search Console exports with different definitions of a click and of AI exposure, and Google names additional metrics in its AI report only as a possibility2.
Sources & deeper reading
- 1Google Search Central — "AI features and your website" (AI Overviews and AI Mode in the Performance report, "Web" search type)
- 2Google Search Central Blog — "Introducing Search Generative AI performance reports in Search Console" (3 June 2026, all websites from 31 August 2026; impressions, pages, countries, devices, dates)
- 3Pew Research Center — "Google users are less likely to click on links when an AI summary appears in the results" (22 July 2025; browsing panel, 900 US adults, March 2025)
- 4Seer Interactive (Tracy McDonald) — "AIO Impact on Google CTR: September 2025 Update" (4 November 2025; Search Console data, 3,119 informational queries, June 2024 – September 2025, grouped by AI Overview status in September 2025)
- 6Seer Interactive — "AIO Impact on Google CTR: 2026 Update" (24 April 2026; Search Console data, 53 brands, 5.47M queries, January 2025 – February 2026)