Don't Be Fooled by AI Search Stats

Most stats about AI search are directionally right but misleading if taken at face value. The study claiming 'brands cited in AI Overviews get 35% more clicks' is qualified by its own authors as 'correlation, not causation.' The 'CTR is halved' figure actually ranges from roughly 34% to 61% across studies — a rounded number gone rogue. And the very Google Search Console you use to verify all this hides most of your queries through anonymization. Don't get swept up in numbers: trace sources to their origin, know the limits of your own measurement, and recognize that what you should do still converges on 'good SEO.' That is the only honest way to face the numbers in the AI search era.

The Google I/O 2026 'Hype'

On May 19, 2026, Google announced a sweeping overhaul of Search at I/O 2026. Per Google, the new AI search box is its 'biggest upgrade in over 25 years,' and figures underscoring its scale were disclosed.

AI Mode surpassed 1 billion monthly users within a year of launch; AI Overviews exceeded 2.5 billion monthly users (CEO Sundar Pichai, I/O 2026 keynote).

The global default model for AI Mode and Search switched to Gemini 3.5 Flash (May 19, 2026, Google).

Source: Google, “Search at I/O 2026” official announcement

When a tectonic shift of this magnitude hits, social feeds and SEO blogs flood with eye-catching numbers: 'citations give you 35% more clicks,' 'CTR is halved.' But should you actually trust them? A principal's job isn't to hand out numbers — it's to judge their origin and limits. Let's dismantle three representative 'number traps' by tracing them to their sources.

Trap 1: '+35% from Citations' Is Not Causation

The most-cited figure: 'brands cited in AI Overviews get 35% more organic clicks.' The source is a Seer Interactive study (3,119 search terms, 42 organizations, June 2024–September 2025, published November 2025). Indeed, cited brands saw a 0.70% organic CTR versus 0.52% when not cited — a gap of about 35%.

But the source itself qualifies it

Seer explicitly notes: 'We cannot definitively prove citation causes higher CTRs; it's equally possible that brands with stronger authority and higher baseline CTRs are simply more likely to be cited by Google's AI.' In other words, this study cannot separate 'citation raised clicks' from 'a strong brand gets both cited and clicked.'

Reading	Resulting Action	Validity
Read as causation (citation → more clicks)	Chase 'citation hacks'	Risky (source denies it)
Read as correlation (strong → also cited)	Raise authority itself	Sound (= good SEO)

Many secondary blog posts strip away this caveat and cite +35% as causal proof that 'getting cited increases clicks.' But if the causal arrow points the other way, what you should pursue isn't citation tricks — it's the authority that earns the citation, i.e., good SEO. Same number, but mistaking the direction of causation flips your strategy entirely.

Source: Seer Interactive, “AIO Impact on Google CTR (September 2025 Update)”

Trap 2: The Rounded 'CTR Halved' Figure

You'll also often hear 'CTR is halved on informational queries with AI Overviews' or 'drops over 50%.' Directionally true. But no primary study matches that rounded 'halved' / 'over 50%' figure exactly. The measured values differ by study.

Seer Interactive

On informational queries showing AI Overviews, organic CTR dropped 61% (1.76% → 0.61%; June 2024–September 2025, 3,119 search terms).

Ahrefs

A 58% drop at position 1 (300K-keyword analysis, updated December 2025). Their earlier version reported -34.5% — different scope and timing move the number.

Multiple large independent studies agree on the direction — CTR falls (roughly 34–61%). But treating a single rounded number as an absolute misreads reality. Hold numbers as a range, not a point: who measured, on which queries, and when all move the figure for the same phenomenon.

Source: Seer Interactive (AIO CTR study)Source: Ahrefs, “AI Overviews reduce clicks (updated)”

Trap 3: Your Own GSC Doesn't Show the Truth Either

It's not only external studies that are distorted. The instrument you use to verify them — Google Search Console — has a structural limit of its own. GSC withholds queries searched by only a few people from its query report to protect privacy (anonymization). This is a behavior Google officially acknowledges.

The query report shows only part of the truth

As a result, clicks summed by query come out smaller than the same totals viewed by page. The difference is the anonymized queries. 'Not in the query report' does not mean 'not searched' — often it just means 'not shown.'

Smaller sites are hit harder: with fewer searches per query, more of them get anonymized, and it's easy to misread the blanks in the query report as 'no traffic.' In the AI search era, phrasing diversifies and grows more long-tail, so the share of anonymized queries trends upward. Lean too hard on query-level analysis and you lose sight of your site's real shape.

The fix isn't hard: don't over-trust query-level figures — read by page using impressions, clicks, and average position. We cover how to read GSC correctly in a dedicated article. the GSC verification guide.

The Call: Measure the Limits of Your Own Measurement

What the three traps share: the numbers are distorted at the instrument level. External studies blur causation and correlation and drift through rounding; your own GSC hides part of the truth via anonymization. So how should you deal with numbers in the AI search era? Four principles.

Read external numbers down to the source's caveats

Don't stop at a secondary blog's summary — go to the study itself. Check whether the authors qualify causation, and what the sample and period are. The moment you read correlation as causation, your tactics miss the mark.

Hold rounded numbers as real values and ranges

Round figures like 'halved' or 'doubled' are catchy but treacherous. Line up real values from multiple studies and grasp the range. Simply refusing to treat one number as absolute prevents both overreaction and complacency.

Read your GSC by page, assuming anonymization

Neither over-trust nor misread the blanks in the query report. Anchor on page-level impressions, clicks, and position. Knowing the limits of your own instrument is the foundation for soberly judging anyone else's numbers.

What to do still converges on 'good SEO'

Once you read the numbers correctly, the moves that remain are the universal basics: E-E-A-T, structured data, and clear direct answers. Google's official guide (updated June 5, 2026) states that 'from Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO,' and reaffirms that llms.txt, special schema, and AI-specific rewriting are unnecessary.

Source: Google Search Central, “Optimizing for Generative AI Features”

Our consistent stance: 'GEO is just good SEO'

We have consistently argued that 'GEO is just good SEO.' Google now officially confirming this only reinforces that position. We examine the 'llms.txt is unnecessary' debate in detail in a separate article. Is llms.txt Really Unnecessary?

Summary: The Bigger the Shift, the More You Question the Numbers

Google I/O 2026 is a genuine tectonic shift. AI Mode at a billion users, AI Overviews at 2.5 billion — the premises of search are moving. But the bigger the shift, the more sensational numbers fly around, losing their sources and taking on lives of their own.

The +35% is a correlation its own authors disown; 'CTR halved' is a rounding of a measured 34–61% range; and your own GSC hides part of the truth through anonymization. Don't be fooled by the numbers. Trace sources, know the limits of measurement, and then concentrate on good SEO — even in the AI search era, that remains the most reliable judgment.

Before the Numbers, Free-Check Your Page's Foundation

Enter a URL to diagnose 45 items including E-E-A-T signals, structured data, and meta information. See how solid the 'good SEO' foundation that matters in the AI search era really is.

Written by

今井政和

SEO Director / Frontend Developer

SEO Director with 20+ years of web industry experience. Creator of CodeQuest.work SEO and the official WordPress plugin "ORECTIC SEO CHECK." Author of a book on web strategy inspired by Edo-era merchant principles.

@imai_director

FAQ

Do citations in AI Overviews really increase clicks?▾

A correlation is observed (in Seer Interactive's study, cited brands had a 0.70% organic CTR vs 0.52% when not cited — a ~35% gap). However, Seer itself qualifies this: 'we cannot prove citation causes higher CTR; brands with stronger authority and higher baseline CTR may simply be more likely to be cited.' Because the causal arrow may point the other way, raising authority itself is sounder than chasing 'citation hacks.'

How much does CTR drop with AI Overviews?▾

It varies by study. Seer Interactive reports a 61% drop on informational queries (1.76% → 0.61%); Ahrefs reports a 58% drop at position 1 (an earlier version showed -34.5%). The direction agrees — roughly a 34–61% decline — but no primary study matches a single rounded 'halved' or 'over 50%' figure exactly. Hold numbers as a range, not a point.

Why don't GSC's query-level numbers match the page-level totals?▾

Google Search Console withholds queries searched by only a few people from its query report to protect privacy (anonymization) — an officially acknowledged behavior. As a result, clicks summed by query come out smaller than the same totals viewed by page; the difference is the anonymized queries.

On a small site, does few queries showing in GSC mean no traffic?▾

Not necessarily. Smaller sites have fewer searches per query and are hit harder by anonymization. 'Not in the query report' usually means 'not shown,' not 'not searched.' It's safer to read by page — impressions, clicks, and average position — rather than by query.

So what should I actually do in the AI search era?▾

Once you read the numbers correctly, the moves that remain are the universal basics: E-E-A-T, structured data, and clear direct answers. Google's official guide (updated June 5, 2026) states that optimizing for generative AI search is optimizing for the search experience — and thus still SEO — and reaffirms that llms.txt, special schema, and AI-specific rewriting are unnecessary. GEO is just good SEO.

Search Console10 min

Diagnose the Foundation, Not the Hype

Enter a URL for a 45-item SEO diagnosis with auto-generated improvement code. See your 'good SEO' implementation at a glance for the AI search era.

Free SEO Check