AI Search Statistics 2026: The Data Behind the Shift

This article brings together three different kinds of evidence: historical studies of Google clicks, vendor and analyst estimates of AI referral traffic, and a vendor survey of enterprise marketing budgets. Our own directory observations are kept in a separate section with their original populations. They are not interchangeable measurements of the same traffic loss or proof that GEO spending pays off. Where this article relied on trade reporting rather than an original dataset, that distinction is explicit below.

The short version

SparkToro’s Similarweb-based analysis put zero-click U.S. Google searches at 68.01% in January–April 2026. Pew found traditional-result clicks on 8% of visits to pages classified as having an AI summary, versus 15% without one, using March 2025 browsing data and search results retrieved in April. Those are observational comparisons, not by themselves proof of causation. Separately, Conductor reported roughly 1% AI referral traffic in its enterprise-domain sample from May–September 2025 and surveyed enterprise leaders about planned 2026 spending. Each finding has a different denominator; none is a measurement of every website today.

The zero-click shift

Metric	Figure	Source	Period
Share of U.S. Google searches ending with no click	68.01%	SparkToro (Similarweb clickstream data)	Jan–Apr 2026
Same figure, roughly a decade earlier	~45%	SparkToro	~2016 baseline
Mobile-only zero-click rate	Not established by the primary text/chart reviewed	The earlier ~77% figure is not retained as a verified finding	Jan–Apr 2026 claim
Desktop-only zero-click rate	Not established by the primary text/chart reviewed	The earlier ~50% figure is not retained as a verified finding	Jan–Apr 2026 claim

SparkToro explicitly warns that the decade comparison mixes Jumpshot, Datos and Similarweb panels: the roughly 45% historical figure and 68.01% in 2026 are not a like-for-like longitudinal sample. Its 2026 calculation weights browser-based searches two-thirds mobile and one-third desktop and excludes Google’s mobile search app. SparkToro argues AI Overviews contributed to the rise, but its near-60% CTR reference links to a separate Ahrefs study, not an AI-Overview treatment effect measured in the Similarweb panel. We previously merged those sources and overstated the causal conclusion.

Click behavior: observational comparisons and a randomized experiment

These studies ask related but different questions. Pew observes behavior; Ahrefs compares keyword-level desktop CTR against an estimated counterfactual; Agarwal and Sen randomly alter the search interface. Their results should not be ranked by institutional reputation or pooled into one effect size. The negative click findings remain important, but the unit, study window and design travel with each number.

Study	Finding	Method	Date
Pew Research	Traditional-result click in 8% of visits with a summary vs. 15% without; summary-source click in 1% of visits with a summary	900 U.S. adults; 68,879 unique Google searches; observational browsing panel	March 1–31, 2025 browsing; results re-fetched April 7–17; published July 22
Agarwal and Sen working paper, via April 27 reporting	April reporting: 38% fewer organic outbound clicks on triggered queries; zero-click 54% without vs. 72% with AIO	Randomized Chrome-extension study; 1,065 U.S. desktop participants in reported analysis	Two weeks per participant, Jan–Feb 2026; April working-paper estimate, not a current revised estimate
Ahrefs	Estimated 58% lower average CTR for the position-one page relative to its modeled no-AIO counterfactual	300,000 keywords (150,000 AIO / 150,000 informational non-AIO); aggregated desktop GSC CTR	December 2023 vs. December 2025; published February 4, 2026

Pew also reported sessions ending on 26% of pages with a summary versus 16% without. Its definition was exiting the browser for at least five seconds; this does not establish that a user read, trusted or was satisfied by the summary. The April re-fetch means the study did not capture the exact March result page at the moment of every visit. The AEA protocol independently documents Agarwal and Sen’s three-arm randomized design and January–February 2026 window, with about 1,500 planned participants. That planned sample is not the 1,065 analyzed participants in April reporting. The SSRN text was inaccessible during this correction, so the original 38%/54%/72% figures remain expressly attributed to that dated reporting, not claimed as freshly checked paper results or peer-reviewed conclusions.

AI referral traffic is growing fast, from a genuinely small base

Metric	Figure	Source	Period
AI referrals as a share of traffic in Conductor’s enterprise-domain sample	1.08% (roughly 1%)	Conductor benchmark	U.S., May–Sep 2025
ChatGPT’s average share of AI referrals across those industries	87.4%	Conductor benchmark	U.S., May–Sep 2025
ChatGPT referral-traffic growth	+52% YoY	Similarweb, via Digiday	Sep–Nov 2025 vs. Sep–Nov 2024
Gemini referral-traffic growth (same window)	+388% YoY	Similarweb, via Digiday	Sep–Nov 2025 vs. Sep–Nov 2024
ChatGPT monthly active users	~810M (+5%)	Sensor Tower, via Digiday	Aug–Nov 2025
Gemini monthly active users	~346M (+30%)	Sensor Tower, via Digiday	Aug–Nov 2025

Conductor’s traffic analysis used anonymized, aggregated traffic across 1,215 enterprise-customer domains, not a census of the web. Its 13,770-domain citation corpus is a different population. AI Mode and AI Overviews were not separated into its AI-referral channel. The Similarweb growth and Sensor Tower user figures above were supplied to Digiday and remain attributed to its 22 December 2025 reporting; we did not obtain their underlying panels. Referral growth of 388% means about 4.88 times the earlier level, not a 388-percentage-point gain in all traffic. None of these samples quantifies whether the same publishers recovered their lost Google clicks. Citation visibility and referral traffic therefore need separate measurement.

Enterprise leaders reported spending and plans, not verified returns

Metric	Figure	Source	Period
Average share of 2025 digital marketing budget allocated to GEO/AEO	12%	Conductor survey	2025
Digital marketing leaders reporting high/significant GEO investment in 2025	56%	Conductor survey	2025
Digital marketing leaders planning to increase GEO/AEO spend in 2026	94%	Conductor survey	Plans for 2026; exact field dates not stated in accessible methodology

Conductor surveyed over 250 U.S. enterprise leaders at organizations with 500+ employees across 12+ industries; 73% described their AEO programs as advanced or very advanced. The 12%, 56% and 94% figures describe that vendor-recruited audience, not all marketers or audited budget transfers. Planned increases do not prove realized spending or return on investment. Our companion vendor pricing and feature drift article documents supply-side changes, but those changes do not establish that customer budgets caused them.

What we see in our own index

The article’s retained August records describe a 72-tool directory cut: API status determined for 64, with 43 recorded yes and eight unknown; the self-serve examples were summarized at $14.99–$99/month. A separate 2026-08-13 correction below uses a 71-tool trial population, with 36 advertised trials (33 stated lengths and three unstated). We preserve those original, different denominators rather than silently rebasing them. Warehouse collector run 8116, observed 13 August 2026 at 03:15 UTC, retains a 72-tool taxonomy population, but does not by itself establish the article’s API, price or 71-tool trial classifications. An immutable source-to-row mapping for those classifications was not recovered in this review. Treat them as archived editorial observations, not independently reproduced statistics or proof of a market response to the budget survey.

How to read these numbers together

None of the four sections above are measuring the same thing, and conflating them is the most common mistake in how this data gets cited. Zero-click and AI-Overview-CTR studies measure what happens to Google search traffic. AI referral-traffic studies measure a separate, much smaller channel: visits that originate from ChatGPT, Gemini, or Perplexity, not organic Google results. Marketer-budget surveys measure intent and spend, not outcomes. A brand losing organic clicks to AI Overviews is not the same brand necessarily gaining AI referral traffic, and a rising GEO budget doesn’t guarantee either metric moves in that brand’s favor. It’s a bet, not a result. If you’re trying to answer “is this working for us specifically,” the numbers above are the market context; our guide to choosing an AI-visibility platform covers how to measure your own brand’s citation rate rather than infer it from industry averages.

2026-08-13 correction: the index figures in “What we see in our own index” previously said 33 of the 71 software tools advertise a free trial, 30 publishing its length. Both numbers were taken from the free-tier census but from two different cuts of it. Its “30” counts tools whose only free option is a published-length trial, which excludes Ubersuggest, CiteLens and SEORCE because each also runs a standing free tier; adding the three vendors that advertise a trial without stating a length therefore produced a total that omitted those three. Re-counted against the corpus on 2026-08-13: 33 tools publish a trial length and three more advertise a trial of unstated length, so 36 of the 71 advertise one. The free-tier census itself was correct and is unchanged; only this article’s summary of it was wrong.

5 September 2026 editorial correction: distinguished Ahrefs CTR analysis from SparkToro clickstream data, added Pew’s re-fetch limitation, corrected Conductor’s observation window and survey population, and qualified unproven causal and directory claims. The original publication and correction dates are preserved. This was a source review, not a new traffic measurement, vendor survey or directory census.