Where AI SEO statistics actually come from

The five provenance checks

These are the checks used to grade every figure in this audit, and they are the grading scheme applied to every statistic published anywhere on this site. Each takes about two minutes. Together they catch nearly everything in the ledger below.

1. Click the citation, not the summary. Follow the actual link, one hop at a time, until you reach a primary source, a circular loop, or a page that cites nothing. If there is no link at all, stop there — you already have your answer.
2. Find the population. A percentage without a denominator is not a statistic. "US searches" and "queries that trigger an AI Overview" are different worlds, and conflating them is how the 58% collision below happens.
3. Find the date and the collection window. These products change monthly. An AI-search figure older than roughly six months may describe a product that no longer exists in that form, and an undated figure decays silently.
4. Check the method is disclosed well enough to disagree with. Sample size, technique, and how the measurement was taken. The real test of disclosure is whether a critic could attack it — not whether an organisation's name appears.
5. Ask what would falsify it. If no possible observation would contradict the claim, it is a position rather than a finding, however many decimal places it carries.

Grades follow directly from those checks. Traceable means the chain reaches a named study with sample, date and method disclosed. Partial means it reaches a named organisation whose underlying method or corpus is withheld. Broken chain means no primary source is reachable, or the sources contradict each other irreconcilably. And when a number passes all five, cite the primary source rather than the page you found it on — every extra hop is one more place the method can fall out.

An unchecked statistic is a liability in a specific way. When you quote a figure in a client deck or a strategy doc, you are implicitly vouching for what it measures. It may instead measure something adjacent — a different population, a different time window, a different definition of the same word. The recommendation built on top of it is then wrong in a way nobody can see until it fails.

• A budget reallocated on "AI traffic converts 23× better" — one of three unreconciled vendor claims.
• A crawler blocked on a crawl-to-referral ratio drawn from a different vertical entirely.
• A journalist repeating "AI Overviews cut clicks by 58%", a figure this audit found conflates two separate studies.

None of these require bad faith. They require only that a plausible-sounding number circulated widely enough, for long enough, that checking it stopped feeling necessary.

So this audit asks one question of twelve claims: can a reader get from the statistic to the thing that produced it? Not "is it true" — that is a much harder question. Just: is the path there open.

How we traced twelve claims

The method is deliberately dull, because the value is in doing it consistently rather than cleverly.

What we found

The headline is less damning and more interesting than "SEO statistics are made up." Nearly half trace cleanly. The problem is the other half — and specifically that a reader has no way to tell which half a given number belongs to, because both are presented in the same confident register.

Chain length is the practical signal. One or two hops is normal and healthy — a trade publication summarising a study is doing its job. Past three hops, the original method has almost always been stripped away, and what survives is a number and a vibe.

The full ledger

Every claim, its grade, and where the chain actually ends. This is the part worth bookmarking.

A live thirteenth case, not yet in the ledger, shows the same method working on a fresh claim. "Google AI Mode fans a query out into 8–16 sub-queries" is repeated across dozens of SEO articles from 2025 and 2026, almost always without a citation. Google's documentation confirms the fan-out mechanism exists, and a granted patent (US11663201B2) describes eight distinct sub-query types the system may generate. Nowhere in that patent, or in any Google publication located during this research, is a specific per-query count of "8 to 16" stated. Graded Broken chain — plausible, oft-repeated, untraceable to any primary count. Full write-up at the Query Fan-Out Corpus study.

One trace, in full: how "2,237 pages per referral" was checked

Ledgers are abstract. One trace, step by step, exactly as it happened.

1. The claim, as encountered: a marketing blog post stated "Anthropic's ClaudeBot crawls 2,237 pages for every visitor it sends back," with no link attached.
2. Hop one: a search for the exact phrase surfaced a second aggregator site repeating the same number, this time with a link to a third-party SEO-tools blog.
3. Hop two: that SEO-tools blog cited "Cloudflare data, July 2026" in prose, again without a direct link to Cloudflare Radar itself, and rounded the figure slightly differently (2,200-ish) than the first source.
4. Hop three: Cloudflare Radar's own public dashboard was checked directly. It exposes bot-traffic classification and category-level breakdowns. The specific per-crawler ratio, as phrased, was not found on Radar as a standalone published figure. A third party most likely derived it from Radar's underlying data rather than quoting a Cloudflare publication.
5. Verdict: graded Partial . The underlying network is almost certainly Cloudflare's, and the figure is plausible against the general pattern Cloudflare has described publicly. But nobody in the chain shows the derivation — which time window, which subset of traffic, how "referral" was defined. A reader repeating "2,237:1" is repeating a number nobody has shown their work for.

That is what "partial" looks like in practice: not a fabricated number, but one that has lost its method somewhere between the primary network data and the blog post that made it famous.

The 58% problem

This is the clearest finding in the audit, because it shows how a statistic can be simultaneously accurate and useless. Three separate, legitimate pieces of research produced numbers near 58%. They measure entirely different things. In circulation, they have collapsed into one figure people quote interchangeably. The CTR statistics page keeps each of the three attached to the population it actually describes, which is the only way the number stays usable.

Read those rows as a set. The first describes search behaviour overall. The second is a conditional effect on one subset of queries. The third is a modelled business impact from self-reported survey data. Quoting "58%" without saying which one you mean lets a reader assume whichever is most alarming — usually the second, applied to the population of the first.

The same collapse has happened to the zero-click figure, which is quoted for a population it does not cover — traced separately here.

The conversion spread

The second broken chain is more consequential commercially, because it is the number people use to justify budget. Published claims about how much better AI-referred traffic converts range from 4.4× to 23×, with one study reporting 14.2% against 2.8% across 312 B2B firms. These are not small disagreements. They are different orders of magnitude, and they cannot all describe the same phenomenon.

The conversion benchmarks page lists every published claim in that range side by side, with what each one counted.

The likely explanation is not dishonesty. It is that "conversion" is not one thing:

• Some datasets count any goal completion; others count only revenue events.
• Some compare AI referrals to all organic; others to non-brand organic only.
• AI referral volume is tiny — around 1% of traffic in one large panel — so small absolute numbers produce wildly unstable ratios.
• Sites that show up in AI answers skew toward considered, high-intent purchases in the first place, which inflates the comparison before anyone measures anything.

That last point is the one nobody controls for, and it may account for most of the effect. Until somebody publishes a like-for-like definition, the honest version of this statistic is: AI referral traffic appears to convert better, by an amount nobody has credibly established.

The unverifiable tier

Four claims sit in an awkward middle. They come from named organisations doing real work at genuine scale — a reported 680 million citations is not a number anyone invents casually — but the corpus, the query set and the extraction method are not public.

This is not an accusation. It is a structural observation, and the same one that motivates the AI Citation Index: a company whose product is proprietary citation data cannot open that data without dismantling its own business. The incentive runs the wrong way, permanently. So these figures should be quoted the way you would quote a well-sourced press claim — attributed, dated, and flagged as unverified. Not laundered into a bullet point that reads like established fact.

Why bad statistics spread, and who benefits

The mechanism is structural rather than a matter of a few careless writers. A number with a caveat attached — "roughly," "in one non-public analysis," "for a specific subset of queries" — is less shareable than the same number stated flatly. Every hop has a mild incentive to drop a qualifier, because the flatter version reads more authoritatively. Repeat that across three or four hops and a hedged, population-specific finding becomes an unqualified, universal-sounding fact — not through any single act of dishonesty, but through many small simplifications, each individually reasonable.

The first row is not the villain. In every traceable claim in the ledger, the original research was disclosed properly — the degradation happens downstream, at the aggregation and repetition stages, exactly where a reader has the least visibility into what changed.

One propagation vector is specific to this moment. A statistic that appears across enough pages has a real chance of being absorbed into a language model's training data, then repeated on request. It carries the same false confidence as the pages it learned from, now delivered as if it were the model's own synthesis. That is a fifth hop, and the "source" a curious reader is shown may simply restate the same unsourced chain. The implication is unchanged, with higher stakes: trace before you trust, whether the number came from a blog, a slide deck, or a chat window.

What good provenance looks like

That last row deserves particular credit. Publishing a revision that undercuts your own earlier headline is the single strongest signal of good faith available to a research publisher, and it costs real attention to do.

This site's own publishing rule

This audit is not commentary. It is the operating rule for every other page here. Any statistic quoted on this site carries the same three-tier grade used in the ledger. A number graded Broken chain is either qualified explicitly in the text or left out, never smoothed into a confident bullet point. Where this site's own research produces a figure — as with the Citation Index — the underlying data is published, so someone else could in principle run this same audit against it. Where every other statistic here sits on this scale is recorded in the master AI SEO statistics index, and a first-party example of the disclosure standard argued for here is the llms.txt log test.

Limitations

An audit about provenance has an obligation to be explicit about its own.

• Twelve claims is a small sample, and it was not randomly drawn. Claims were selected for prominence by judgement. A different selection would produce different proportions, and the 5/4/3 split should be read as a pattern, not a population estimate.
• Tracing was done through public web search. A chain that looks broken may have a primary source behind a paywall, in a PDF, or on a page that has since moved.
• "Broken" is a statement about the chain, not the claim. An untraceable statistic may still be perfectly accurate. This audit measures checkability only.
• It is a snapshot. Pages get edited and studies get published. Grades are true as of the first week of September 2026 and will be re-run twice a year.
• This page is itself a source that could be mis-cited. If you quote the 5/4/3 split, quote the sample size with it.

What the next edition adds

This is v1 of a recurring audit. The next edition is scheduled alongside the first full run of the Citation Index, and adds three things. It expands the traced-claim set from twelve to at least thirty. It adds a fourth grade tier separating "traceable but superseded" from "traceable and current", because a correctly sourced figure from 2024 can still mislead when quoted as though it describes 2027. And it begins tracking how long a corrected figure, like Ahrefs' 76%-to-38% update, takes to displace the outdated version in circulation. That lag, once measured, will itself be a statistic worth tracing.