The SEO Stats Everyone Quotes Are Mostly Vendor Marketing — Here’s What’s Actually Measured

By Andrej Ruckij · June 7, 2026

TL;DR: If you’ve read anything about AI search in 2026, you’ve seen the numbers: 64.82% of Google searches are zero-click, AI Overviews cut clicks by up to 58%, 96% of AI citations come from “strong-E-E-A-T” sources, brand mentions correlate 3× more than backlinks. Almost all of these come from a single SEO vendor’s blog. Exactly one of the big claims is backed by primary, independent research — and it’s the one most worth acting on. Here’s how to tell measurement from marketing, so you can quote the first kind to your boss and discount the second.

Why this matters

SEO strategy in 2026 runs on statistics, and most of those statistics trace back to vendor content marketing: SEO tools publishing impressive numbers that happen to support buying SEO tools. The numbers aren’t necessarily wrong. But “a vendor’s correlation study found X” and “a research institution measured X in real user behavior” sit at very different levels of confidence. Treat them the same, and you end up betting budget on a decimal point nobody ever measured.

So we went hunting for the primary sources behind the five most-quoted AI-search stats. Here’s the scorecard.

The one claim that’s actually anchored: AI Overviews cut your clicks

The single best piece of evidence in this space is the Pew Research Center’s July 2025 study. Crucially, it’s clickstream data, not a survey: Pew tracked the real browsing behavior of 900 US adults across 68,879 Google searches. What they found:

Users clicked a traditional search result on 8% of pages with an AI summary, versus 15% without — roughly half the click-through rate.
They clicked the AI summary’s own cited source links on just 1% of visits. Being cited inside an AI answer almost never sends you a visitor.
Sessions ended outright on 26% of AI-summary pages versus 16% without.

This is the claim to take to the bank: a named research institution measuring actual behavior. (Google publicly disputed the study’s methodology, but that’s the company whose product is implicated, so weigh it accordingly. The direction is corroborated by independent panels.)

Notice what this doesn’t say. It doesn’t validate the popular “up to 58% position-1 CTR drop” figure; that’s a vendor number. Pew’s ~47% halving is the real, institutional version. Same direction, measured honestly.

The four claims that are vendor-only

The stat you’ve seen	Where it actually comes from
64.82% of searches are zero-click	Similarweb (a vendor; it does publish methodology). Pew’s data supports a “roughly two-thirds browsed elsewhere or left” order of magnitude — but the precise decimal is one company’s estimate.
AI Overviews cut CTR up to 58%	Ahrefs (vendor). The measured version is Pew’s ~47%.
96% of AI citations come from strong-E-E-A-T sources	Vendor-only. We found no primary source for this figure.
Brand mentions correlate 3× more than backlinks (0.664 vs 0.218)	A vendor correlation study. No independent source for the coefficients.
Google AI Mode: 75M users, 92–94% zero-click	Vendor/unverified. No institutional or panel source surfaced.

That doesn’t make them useless. It makes them directional. A reasonable rule: cite the direction with confidence; cite the decimals with a caveat. “AI Overviews substantially reduce clicks, roughly halving them in measured data” is defensible. “AI Overviews cut CTR by exactly 58% and 96% of citations need E-E-A-T” is repeating marketing copy.

The self-report trap

Here’s a worked example of why measurement beats assertion. The Reuters Institute at Oxford surveyed people and found 33% say they “always or often” click links inside AI summaries. Encouraging, on its face.

But Pew measured 8% (and 1% for the AI’s own citations). The survey number runs four times rosier than the behavior. People overstate how much they click. That’s the whole reason behavioral data beats stated intent: when you can watch what users actually do, trust it over what they say. Treat self-reported AI-engagement numbers with suspicion.

What the evidence does support

Strip away the unverified decimals and a robust, defensible picture remains:

AI answers are materially reducing clicks to the open web. This is measured. Plan for it. On-SERP and in-answer presence is now a marketing outcome in itself, not just a path to a click.
Earned, third-party authority beats brand-owned content in AI search. The specific “3×” coefficient is vendor, but the direction has independent experimental support: a 2025 University of Toronto study (Chen et al.) found AI search engines show “a systematic and overwhelming bias towards earned media over brand-owned and social content.” The strategic move — get cited by authoritative third parties, the heart of generative engine optimization (GEO) — holds even though the headline correlation isn’t independently confirmed.
Measure visibility, not just clicks. When two-thirds of searches don’t produce a click, “organic sessions” is no longer the whole scoreboard.

The meta-point

This article isn’t here to dunk on vendors. Their data is often the only data we have, and some of it — Similarweb’s clickstream — is genuinely useful. The point is that knowing which number is which is a professional skill. Say “the click-reduction effect is real and institutionally measured; the 96% figure is an unverified vendor estimate,” and you’re giving sharper advice than the agency reciting the full listicle as gospel. In a field flooded with confident statistics, calibration is the edge.

Key takeaways

Most quoted 2026 AI-search stats are vendor estimates, not independent measurement.
The AI-Overview click-reduction claim is anchored — by Pew Research clickstream data (~47% drop measured; ~half the clicks).
64.82% zero-click, 58% CTR drop, 96% E-E-A-T, 3× brand-mentions, and 75M AI Mode users are vendor-sourced — directionally plausible, not independently verified.
People overstate clicking in surveys (Reuters self-report 33% vs Pew measured 8%) — trust behavior over self-report.
The defensible strategy survives anyway: plan for fewer clicks, build earned third-party authority, measure visibility not just traffic.

new-seo-playbook-ai-search — The 8-practice playbook these numbers underwrite (and which figures to trust acting on)
can-you-trust-ai-overviews — Companion piece: how often AI Overviews are wrong, and whether you can trust what they say about your brand
seo/zero-click-strategy — The operating model for a high-zero-click world, with the full primary-vs-vendor evidence grading
e-e-a-t-ai-search — Deep dive on the biggest vendor figure here: E-E-A-T as the AI-citation gate, and which of its numbers are real
seo/ai-visibility — Measuring brand presence in AI answers (where the vendor E-E-A-T / brand-mention numbers live)
glossary/share-of-model — The competitive metric for AI-mediated discovery
when-can-you-trust-ai — The general trust framework these search-specific numbers sit inside

Sources

Pew Research Center (July 2025), “Google users are less likely to click on links when an AI summary appears” — clickstream, 900 adults / 68,879 searches.
Reuters Institute, “Generative AI and News Report 2025” — self-reported AI-overview click-through (the contrast case).
Chen, Wang, Chen & Koudas (Univ. of Toronto, 2025), arXiv:2509.08919 — earned-media bias in AI search (preprint; directional support).
Vendor sources (Similarweb, Ahrefs) — the origin of the 64.82% / 58% / 96% / 3× figures.
Full evidence grading: seo/zero-click-strategy § “How solid are these numbers?”