- The clearest tell of a fake or unverifiable metric is precision with no disclosure: a specific decimal, a numbered tier, a clean percentage attached to nothing checkable. This site traced one real example: a fabricated 0.81 "E-E-A-T correlation" that replaced a real study's honest null result.
- A blended, cross-platform score isn't fake, but it can hide the most useful part of the picture. Citation rate has been measured varying by up to 615x for the same brand across platforms.
- Funding and market size are unrelated to methodology disclosure. A direct vendor comparison found the largest-dataset vendor published the least technical detail of the three checked.
- A score's day-to-day volatility (up to 65% source turnover, per a 45-day academic study) means a single reading is a snapshot, not a fact, and nobody has yet published a study showing these scores predict anything about the future either.
- IAB's own published framework states the rule plainly: where a provider can't or won't disclose against a required methodology item, treat that refusal as the answer.
AI visibility score covers what a score is and why a single reading is unreliable. This piece is the practical companion, a checklist built directly from real cases this site has traced back to their actual sources, for spotting a fake or misleading metric before it reaches a board deck or a budget decision.
Is the number too precise for what's actually disclosed?
The most repeated fake statistic in AI search marketing traces a real case: Surfer's July 2026 study of roughly 5 million citation URLs across 20,000 prompts found domain-authority proxies correlating with AI citation frequency at close to zero: an honest, disclosed null result. What now circulates in its place is a specific 0.18-versus-0.81 correlation between domain authority and something called "E-E-A-T," plus numbered "trust score" tiers each paired with a citation-rate percentage. None of it traces to Surfer's actual study, or to any other named, disclosed source. E-E-A-T itself has no public documentation from any AI lab describing it as a measurable, scoreable input to citation or retrieval at all.
The pattern worth internalizing: a decimal correlation coefficient, a numbered tier system, and a clean percentage all read as more credible than a vague claim, precisely because specificity usually correlates with real measurement. That correlation breaks down completely once the specificity isn't attached to a named source, sample, and methodology someone could actually check.
Is the score blended across platforms that behave nothing alike?
A blended score isn't fabricated the way the E-E-A-T numbers are, but it can hide the most useful signal available. Superlines' March 2026 analysis of 62 brands across 10 platforms found citation rate varying by up to 615x for the identical brand depending on the engine: in their data, Grok cited sources at 27.01% of responses, Perplexity at 13.05%, ChatGPT at 0.59%, and Claude at 0%. Average those together into one number and a brand that completely dominates one engine while being nearly invisible on another can come out looking like an unremarkable middle-of-the-pack score, technically accurate as an average, and useless for deciding where to actually focus.
Does the vendor's funding or size stand in for real disclosure?
What vendors won't tell you about their own AI-visibility methodology compares three real, funded companies directly, and the result cuts against the intuitive assumption: Ahrefs' Brand Radar, with a stated 421-million-plus prompt corpus, discloses per-engine volumes and makes its raw corpus searchable. Profound, the best-funded of the three at $154.5 million raised and the one with the largest stated dataset (1.5 billion-plus conversations), does not publish comparable technical documentation of how that data becomes a score. Dataset size, funding, and market position turned out to be unrelated to methodology transparency in that comparison: not correlated, not anti-correlated, just separate axes entirely, and a large or well-funded name is not itself evidence of disclosure.
Is the number a single snapshot dressed up as a stable reading?
The University of St. Gallen's 45-day academic study found cited sources turning over roughly 65% from one day to the next, and identical prompts run simultaneously overlapping only 32-43%, see AI visibility score and how many prompts do you actually need for the full convergence-analysis numbers. A dashboard reporting a single day's reading as the number, with no stated run count or measurement window, is reporting one frame of a genuinely noisy signal as if it were a fixed fact.
Is the score being sold as predictive of something it's never been shown to predict?
Do AI visibility audits actually predict anything? asks a narrower, sharper question than whether a score is accurate: does a score measured today predict citation volume, traffic, or revenue at a later date? No study meeting basic disclosure standards was found testing that relationship directly. A vendor pitching a score as a validated predictor of future business outcomes is making a claim this site could not find support for anywhere; a vendor offering the same score as a directional trend to track over weeks is making a more modest, better-supported claim.
What's the actual five-question checklist?
| Ask this | Why it catches a fake or misleading metric |
|---|---|
| Is the number precise with no disclosed source, sample, or methodology? | Precision without disclosure is the E-E-A-T pattern, a fabricated-sounding replacement for a real, honest null result. |
| Is it a single blended figure across platforms? | Citation rate has been measured varying up to 615x between platforms for the identical brand, a blend can hide the most actionable part of the picture. |
| Does the vendor's size or funding stand in for actual disclosure? | Unrelated axes: the best-funded, largest-dataset vendor in one direct comparison disclosed the least technical detail. |
| Is it a single-day or single-run reading presented as stable? | A 45-day academic study found up to 65% day-to-day turnover, one reading is a sample, not a fact. |
| Is it sold as predicting a future outcome? | No disclosed-methodology study was found establishing that any AI visibility score predicts future citations, traffic, or revenue. |
IAB's own published framework states the operating rule for all five, formally rather than informally: where a provider can't or won't disclose against a required methodology item, that absence should itself be treated as a signal, not a gap to extend the benefit of the doubt on.
Frequently asked questions
What's the single biggest red flag in an AI visibility metric?
Precision without disclosure: a specific decimal number, a numbered scoring tier, or a clean percentage with no named source, no stated sample size, and no described methodology behind it. Specificity usually signals real measurement, which is exactly why it works as camouflage when it isn't attached to anything checkable.
Is a blended, cross-platform AI visibility score automatically fake?
Not fake, but structurally weaker than it looks. Superlines' analysis of 62 brands across 10 platforms found citation rate varying by up to 615x for the identical brand depending on the engine. A single blended number can average out a brand that dominates one platform and is nearly invisible on another into an unremarkable middle score, hiding the most actionable part of the picture without technically being wrong.
Does a big, well-funded vendor automatically mean trustworthy methodology?
No: the two are unrelated. A direct comparison of three funded AI-visibility vendors found the one with the largest stated dataset (1.5 billion-plus conversations) published the least technical detail about how that data becomes a score, while a smaller competitor's disclosure was the most complete of the three. Funding size and market position say nothing about methodology transparency.
What should I actually do when I'm handed a number I can't verify?
Ask the same five questions this piece walks through, and treat a provider's inability or unwillingness to answer them as the answer itself. That's IAB's own stated rule in its published measurement framework, not just an informal suspicion. A number nobody can explain the source of doesn't get partial credit for looking authoritative.
Sources referenced
- Surfer SEO: We Analyzed 5 Million AI Citations. Domain Authority Didn't Predict Any of Them. (Jul 2026)
- Superlines: AI Search Statistics 2026 (March 2026, 62 brands × 10 platforms)
- Schulte, Bleeker & Kaufmann, "Don't Measure Once: Measuring Visibility in AI Search (GEO)" (University of St. Gallen)
- IAB: Measuring Visibility in the AI Era (Aug 2026)
This piece is the practical checklist built from the volatility research in AI visibility score, the fabricated-statistic case in the most repeated fake statistic in AI search marketing, the vendor comparison in what vendors won't tell you, and the predictive-validity question in do AI visibility audits actually predict anything?.
