- Surfer's July 2026 study (roughly 5 million citation source URLs, 20,000 prompts, three months of data across AI Mode, AI Overviews, ChatGPT, and Perplexity) found domain-authority proxies correlating with AI citation frequency at close to zero, slightly negative in their measurement: a real, disclosed, honest null result.
- What actually circulates in its place is a specific, oddly precise cluster of numbers: a 0.18-versus-0.81 correlation between domain authority and something called "E-E-A-T," specific numeric "trust score" tiers with attached citation-rate percentages, a claimed "96% of citations from strong E-E-A-T sources" figure, none of which trace back to Surfer's actual published study, or to any other named, disclosed source located for this piece.
- E-E-A-T itself has no public documentation from any AI lab describing it as a measurable, scoreable input to citation or retrieval systems. It originates in Google's Search Quality Rater Guidelines, a framework for human evaluators, not a disclosed algorithmic signal.
- The pattern: a real, honest, "it doesn't correlate" finding gets replaced, across secondary marketing sites, by a fabricated-sounding precise alternative that reverses the actual conclusion, implying authority signals are highly measurable and predictive when the real study found the opposite.
- This is a different mechanism from every other stat-mythbusting piece on this site: those trace a real, narrow study that gets generalized past its scope. This one is a real null result getting inverted into invented precision.
Where "X% of people use AI search" stats actually come from traces five widely-repeated usage statistics back to real surveys that got generalized past what they actually measured. This is a different, in some ways more concerning pattern: a real study reached a plain, honest null conclusion, and what now circulates in its place isn't a generalization of that finding. It's a specific, precise-sounding set of numbers that doesn't trace back to it, or to anything else checkable.
What did the real study actually find?
Surfer SEO published a study in July 2026, "We Analyzed 5 Million AI Citations. Domain Authority Didn't Predict Any of Them.", pulling three months of AI Tracker data spanning 20,000 prompts and roughly 5 million unique citation source URLs across AI Mode, AI Overviews, ChatGPT, and Perplexity. The study measured how citation likelihood correlated with three separate domain-authority proxies: PageRank and Harmonic Centrality (both derived from Common Crawl data) and Surfer's own Domain Score (via DataForSEO). The finding: domain authority correlated negatively with AI citation frequency, but the relationship was close enough to zero to be noise, a real, disclosed, honest "this doesn't predict what you'd expect" result, with a named methodology and a specific dataset behind it.
That's a useful, credible finding on its own. It's also, on its face, a less exciting headline than most marketing content wants to run with. A plain "we checked, and it's not correlated" doesn't generate the same engagement as a specific, dramatic-sounding number.
What fake statistic replaced it?
Search for how domain authority relates to AI citations now, and a different, much more specific set of numbers dominates: a claimed correlation of 0.18 between domain authority and AI citation probability, against a claimed 0.81 correlation for "E-E-A-T", implying domain authority explains roughly 3% of citation variance while E-E-A-T explains something far larger. Alongside that: specific numeric "trust score" tiers (commonly presented as ranges like 0-19, 20-39, 40-59, 60-79, and 80-100), each one paired with an attached citation-rate percentage, plus a separate claim that 96% of Google AI Overview citations come from sources with "strong E-E-A-T signals."
None of these specific figures trace back to Surfer's actual published study, which measured domain-authority correlation and found it near zero, not 0.18, and never measured or reported an "E-E-A-T correlation" of 0.81 or anything resembling it. Nor do they trace to any other single, named, disclosed study located during research for this piece. They appear across multiple separate-looking marketing and SEO content sites, cross-referencing each other and attributing the numbers vaguely to "recent 2026 analyses" or "research" without ever naming a study, an author, a sample size, or a methodology that could actually be checked.
Why is this worth calling out specifically?
The mechanism here is different from, and arguably worse than, the pattern in where AI search stats actually come from. In that piece, every distorted stat traced back to a real survey that got generalized past its actual scope; the underlying research existed and was checkable, even when the popular version overstated it. Here, the real study reached an honest, checkable null conclusion, and what circulates instead isn't an exaggeration of that finding. It's a specific, invented-sounding replacement that asserts the opposite: that authority signals are, in fact, precisely measurable and strongly predictive, with decimal-point correlations and numbered scoring tiers attached. A reader encountering only the second version would reasonably conclude the opposite of what the one real, disclosed study on this exact question actually found.
Why doesn't E-E-A-T specifically hold up as a measurable AI-citation input?
Part of what makes the fabricated cluster convincing is that it's built around "E-E-A-T" (Experience, Expertise, Authoritativeness, Trustworthiness), a term that sounds official because it is a real Google concept. But E-E-A-T originates in Google's Search Quality Rater Guidelines, a framework written for human quality raters manually evaluating search results, not a disclosed algorithmic score. No public documentation from Google, OpenAI, Anthropic, or Perplexity describes E-E-A-T as a measured, numeric input that feeds into which sources get cited in an AI-generated answer. Attaching a specific correlation coefficient (0.81) to a concept no lab has confirmed measuring at all is exactly the kind of unverifiable precision worth treating with the same skepticism as any other unsourced number.
What pattern is worth checking before repeating a number like this?
A specific decimal correlation, a numbered tier system, a clean percentage: all of these read as more credible than a vague claim, precisely because specificity usually correlates with real measurement. That correlation breaks down entirely once the specificity isn't attached to anything checkable. Before repeating a number like "E-E-A-T correlates at 0.81 with AI citations," the same question that applies to every other statistic on this site applies here: who measured this, what was the sample, and can the original source actually be found, or is the number just precise-sounding enough that nobody's checked.
Frequently asked questions
Does domain authority predict whether AI models cite you?
On the one disclosed, named study located for this piece, no. Surfer's July 2026 analysis of roughly 5 million citation source URLs across 20,000 prompts (AI Mode, AI Overviews, ChatGPT, and Perplexity, three months of data) found domain-authority proxies correlating with citation frequency at close to zero, slightly negative in their data. That's a genuine null result from a real, disclosed methodology, not evidence that nothing predicts citations, just that domain authority specifically doesn't appear to.
What is the fake statistic, specifically?
A cluster of unusually precise numbers that circulate across multiple marketing and SEO content sites: a claimed 0.18 correlation between domain authority and AI citations versus a claimed 0.81 correlation for something called "E-E-A-T," specific numeric "trust score" tiers (commonly cited ranges like 0-19, 20-39, 40-59, 60-79, 80-100) each paired with a specific citation-rate percentage, and a claim that 96% of AI Overview citations come from sources with "strong E-E-A-T signals." None of these specific figures were traceable to a named, disclosed study during research for this piece.
Why do these specific fake numbers sound so credible?
Because they mimic the surface features of real statistics (a specific decimal correlation coefficient, a numbered scoring scale, a clean-sounding percentage) without the underlying disclosure (who ran it, what sample, what methodology) that would let anyone check them. That's the exact pattern this site has flagged before in a different context: precision is not the same thing as verification, and a number with more decimal places isn't automatically a more credible one.
Is E-E-A-T even something AI models can measure directly?
No public documentation from Google, OpenAI, Anthropic, or Perplexity describes E-E-A-T as a measurable, scoreable input to an AI citation or retrieval system. E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness) originates in Google's Search Quality Rater Guidelines, a framework for human quality raters evaluating search results, not a disclosed algorithmic score any lab has confirmed feeding into AI-generated answers specifically.
Sources referenced
- Surfer SEO: We Analyzed 5 Million AI Citations. Domain Authority Didn't Predict Any of Them. (Jul 2026, 20,000 prompts, ~5M citation URLs)
- Google: General Guidelines (Search Quality Rater Guidelines), the source document defining E-E-A-T for human quality raters
This piece follows the same source-tracing discipline as where "X% of people use AI search" stats actually come from, applied to a citation-ranking-factor claim rather than a usage-survey claim. For the underlying vendor-methodology comparison this connects to, see what vendors won't tell you about their own AI-visibility methodology.
