The Referee · Issue 1 · July 2026
We checked the AI-visibility indices against each other
✓ We publish only discrepancies we can document — with both sources
Everyone is racing to publish "the AI-visibility index." Profound, Semrush, 5W, Similarweb, 2X, Conductor — new numbers every month, quoted like gospel. Nobody is refereeing them. So we took the seat. The rule is simple and unbreakable: we only report a discrepancy we can show with both sources. Here's what didn't line up.
1 · There is no "top source for AI" — the engines don't agree
A 7-month, 1,056-data-point study found each engine has its own editorial identity: ChatGPT anchors on Wikipedia; Perplexity, Google AI Overviews, and Gemini lean YouTube; and Claude never surfaced YouTube, Wikipedia, or Reddit — not once. Even Google's own three engines disagree: AI Overviews cite YouTube for Purchase queries while AI Mode routes users back to Google's own properties.
Referee call: "the #1 source for AI" is a category error. Any index that names one winner is measuring one engine, one intent, one month.
Conductor, "How AI Engines Choose and Cite Sources: A 7-Month Analysis" (Sep 2025–Mar 2026) —
conductor.com2 · The schema stat everyone quotes as proof — that a controlled test kills
"53% of AI-cited pages use schema markup" gets passed around as proof that schema drives AI citations. But when Ahrefs actually tracked 1,885 pages that added JSON-LD against matched controls, the effect was statistically indistinguishable from zero (AI Overviews −4.6%, AI Mode +2.4%, ChatGPT +2.2%). The 53% is correlation; the causation isn't there.
Referee call: "X% of cited pages do Y" is not evidence Y causes citations. Demand a controlled test before you re-tool around a stat.
Ahrefs, "We Tracked 1,885 Pages Adding Schema. AI Citations Barely Moved," May 2026 —
ahrefs.com3 · "AI Overviews appear on X% of searches" is meaningless without the sample
One widely-cited tracker put AI Overviews on ~48% of Google queries by March 2026. Real number — but it moves with the keyword sample, the country, and the month. A different query set gives a very different figure. Quoted bare, "48%" (or anyone else's number) is noise.
Referee call: any "AI Overviews appear on X% of searches" stat needs its query set and date attached, or it means nothing.
4 · Adoption is not trust — the inevitability pitch skips half the data
The loudest AI-visibility marketing sells inevitability: "everyone trusts AI answers now, optimize or vanish." The data disagrees — 53% of consumers distrust or lack confidence in AI-powered search results, and 41% say AI overviews make search more frustrating. People are using it and doubting it at the same time.
Referee call: rising usage is not rising trust. The opening isn't to shout louder — it's to be the checkable, sourced answer.
5 · The "AI adoption exploded" jump partly measures the survey, not the world
Local AI-recommendation use was reported at ~45% in 2026 — a real, verified figure. But the year-over-year "it exploded from single digits" framing that travels with it often reflects a changed survey method as much as changed behavior.
Referee call: a giant year-over-year jump is a prompt to check whether the ruler changed, not just the thing measured.
The Referee's creed
- We publish a discrepancy only when we can link both sources. No "we heard."
- Every number carries its date, engine, and method — or it doesn't run.
- We referee claims, including our own. Our Citation Ledger is public and dated.
- Being the fifth index is a race. Being the auditor of the category is a lane — and it earns links without ad spend.
None of this says the indices are worthless — Profound, Semrush, Ahrefs Brand Radar and the rest do real work. It says a single index number, quoted without its date and method, is hype in a lab coat. The receipts-first move is to treat every one as a dated snapshot and check before you spend.
Want your own dated read — not someone's index?
Run the free multi-engine check. You get the raw transcript from each engine, dated. Referee it yourself.
Run my free check
See the ladder