E-E-A-T and AI citation: what our research actually found
E-E-A-T was Google's framework for human quality raters. It is not the rubric AI assistants use when they decide which page to quote. Here is what our three studies showed — and what to do about it.
Introduction
If you have done any SEO in the last five years, you know E-E-A-T: Experience, Expertise, Authoritativeness, Trustworthiness. It is the rubric Google's human search quality raters use when grading pages, and over time the SEO industry collapsed it into a set of visible artifacts — author bylines, "reviewed by" headers, credential sidebars, editorial-policy pages, trust badges.
When generative engines arrived, the natural assumption was that the same signals would transfer. They did not.
Across our research line — the original GEO citation study, an E-E-A-T-focused follow-up, and an ecommerce recommendation-survival study — the visible E-E-A-T signals did not positively predict whether AI assistants cite a page. In the ecommerce study, brands that leaned hard on E-E-A-T styling were recommended less, not more.
Key findings
- E-E-A-T was built for human reviewers, not AI assistants. The framework was designed so that contractors rating Google search results would converge on similar judgments. AI assistants choosing which source to quote do not run the same checklist.
- Visible E-E-A-T signals did not move AI citation rates in our research. Author bylines, "Reviewed by Dr. X" labels, credential boxes, and trust badges showed no positive relationship with whether a page was cited.
- In ecommerce, heavy E-E-A-T optimization was actively counterproductive. The brands AI assistants kept recommending through a shopping conversation were the ones with simple, direct pages — not the ones plastered with editorial credentials.
- Credibility still matters — but AI infers it from content, not signaling. Clear definitions, direct answers, and references to authoritative external sources are how AI assistants gauge whether a page is worth quoting.
Why E-E-A-T does not transfer to AI search
E-E-A-T is a human rubric. It was written so that a contractor rating "is this page trustworthy?" would notice the same things a thoughtful reader would: who wrote this, what are their qualifications, does the site take its own publishing seriously, has anyone vouched for the information.
That is a reasonable set of questions for a human evaluator. It is not how a generative engine assembles an answer. AI assistants are looking for passages that are directly relevant to the user's query, cleanly stated, internally consistent, and corroborated by other sources the model has already weighted as credible. None of that requires reading a byline.
The result: a page can be loaded with editorial signals and still be invisible to AI, and a page with no byline at all can be the one quoted — because what AI cares about is whether the answer is on the page, not whether the page looks trustworthy.
What did not move AI citation
These are the artifacts the SEO industry built up around E-E-A-T. In our research, none of them positively predicted whether a page was cited by AI assistants:
- Author byline blocks with credentials
- "Reviewed by Dr. X" headers
- "About the author" sidebars and bio cards
- Trust badges, certifications, and award logos
- Editorial-policy and fact-checking disclosure pages
- Schema markup for author and reviewer roles
This is not a claim that these elements hurt — for most informational content they were neutral. The point is that adding them is not how you increase AI citation rate. Investing more in this layer is investing in the wrong layer.
Ecommerce is the exception that is not really an exception. In our ecommerce study, heavy E-E-A-T styling on product and category pages correlated with fewer recommendations through a shopping conversation. The pages that survived the funnel were the ones that read like a clear product description, not a journalistic feature with editorial credits. Shoppers — and the AI assistants helping them — wanted product facts, not bylines.
What did predict AI citation
Three on-page signals showed up as positive predictors across our studies. They are content-level signals, not signaling artifacts — they live in the words on the page, not in a sidebar or a badge.
Answerability
Direct, declarative content that answers the question without making the reader dig for it.
Citation Quality
Pointing to authoritative external sources from inside your own content, with enough context that AI can verify the claim.
Definitions
Explicit "X is Y" statements near the top of the page, in the same vocabulary the reader is likely to use.
Two other factors mattered more than any individual page-level signal — and neither is something a content team can fully control:
- Brand recognition. Well-known brands got cited even with thin pages; lesser-known brands struggled even with strong content.
- Query phrasing. The same page could be cited or invisible depending on how the user's question was worded. This was the largest single factor we measured for informational queries.
The full picture — including why citation rate alone is the wrong top-line metric and why platforms agree more than they disagree — is in our cross-study synthesis.
What to do with this
If you already invested in E-E-A-T for Google
Leave it. Author bios, editorial policies, and reviewer credits still play a role in traditional search and they give human readers something to evaluate. The mistake would be doubling down on them as your AI citation strategy. Treat the existing layer as a hygiene investment, not a growth lever.
If you are planning new content
Skip the credential theater. Spend your effort on the parts of the page AI assistants actually read: a clear definition near the top, a direct answer to the question the page exists to answer, and references to authoritative external sources where claims need backing. That is where the citation lift comes from in our research.
If you sell things
Be careful about importing editorial conventions onto product pages. The brands that survived a multi-turn shopping conversation in our ecommerce study were the ones that read like clean product pages, not magazine articles. State what the product is, who it is for, and how it compares — not who reviewed the page.
Key takeaway
E-E-A-T is a framework for human reviewers, not a recipe for AI citation.
AI assistants reward clarity, direct answers, and good sourcing — not bylines, badges, or credential boxes.