E-E-A-T and AI Search: There Is No Score to Improve

Google's guidance on helpful content states that E-E-A-T "isn't a specific ranking factor." It is a concept that describes what good content looks like. Google says its systems approximate it through other factors, with more weight on topics that affect health, money or safety.

The Search Quality Rater Guidelines (revision dated 11 September 2025) back this up. They say: "Using ratings to position results on the search results page would not be feasible, as humans could never individually rate each page on the open web." A rater's judgement of your page does not move your page.

The same guidelines define authoritativeness per topic, not per site. Raters are asked to "consider the extent to which the content creator or the website is known as a go-to source for the topic." That is a judgement about a subject, not a site-wide number.

What this means for you:

• There is no numeric quantity called E-E-A-T inside Google to raise.
• Any audit that hands you an "E-E-A-T score of 62" built that number itself.
• Google's statements are assertions, not independent measurements. They are still the best available evidence for how Google's own surfaces work.

For the full definition and the history of the myth, see the E-E-A-T glossary entry and our digest of Google's helpful content guidance and E-E-A-T.

Google's AI features documentation (last updated 10 December 2025) sets one bar. To appear as a supporting link in AI Overviews or AI Mode, "a page must be indexed and eligible to be shown in Google Search with a snippet." It adds: "There are no additional technical requirements."

The same page says there are no "special optimizations necessary" for these surfaces. It also warns that "indexing and serving isn't guaranteed." Eligibility gets you into the pool. It does not get you picked.

It helps to split an AI answer into stages:

• Access: can the crawler fetch and index the page. You control this.
• Retrieval: does the page enter the model's context for a given query. The operator controls this, partly.
• Generation: which retrieved pages get used and cited. The model controls this.
• Parametric: what the model knows with no web access, frozen at a training cutoff. Nobody changes this quickly.

E-E-A-T is not a checkable gate at any of these stages. As of October 2026, AI Overviews often do not trigger at all, and other engines such as ChatGPT and Perplexity do not use Google's concept. OpenAI's public description of its ranking is that it uses "multiple factors." Our digest of Google's AI features documentation covers the eligibility rules in detail.

The score is a product feature, not a Google metric. Tools gather signals they can see, such as author bylines, an About page, review counts and backlinks. They weight those signals and print a number.

The weights are the vendor's guess. Nothing outside Google can check them against how Google ranks, because Google publishes no such score.

Domain authority follows the same pattern. Domain Authority is a Moz metric, and no search or AI engine computes it. Google publishes no site-level authority score. (Google does describe a news-scoped "topic authority" system, so it is not accurate to say Google has no notion of authority at all.)

The practical test is simple. Ask what ground truth the score is checked against. If the answer is "our own model," treat it as an opinion. The same logic applies to tools that report a position inside AI answers, which we cover in why AI has no rank to check.

OtterlyAI ran a public test in 2026 that bears directly on E-E-A-T signals. It launched two new domains about capybaras, each with roughly 1,000 AI-generated posts published at once.

One site showed an author box under every post, presenting a fictional wildlife biologist with more than 15 years of field experience. The other had no author. Otterly reports the site with the fictional expert lasted about 14 days in Google's index. The site with no author lasted about 90. Otterly reports zero manual actions on either.

This fits Google's written policy. Google's spam policies define scaled content abuse as "when many pages are generated for the primary purpose of manipulating search rankings and not helping users." Google's AI content guidance says the tool is not the problem.

Be careful with the conclusion:

• It is one site per arm, so the author box cannot be separated from other differences between the domains.
• It does not prove author boxes hurt.
• It does show that a fabricated credential did not protect thin, scaled content.

The lesson for E-E-A-T is narrow and useful. Signals that describe expertise are worth nothing if the expertise is not real, and faking them adds risk.

Some E-E-A-T practices are worth doing for readers. One has a positive signal in the evidence. Others are not levers at all.

What you cannot control: whether an AI Overview triggers for a query, how engines rewrite the user's question, which retrieved sources the model cites, and how well known your brand already is. Brand size is the biggest confounder in almost every study of AI citations. Accept that, and spend effort where you have a lever.

Work from the gates you control toward the ones you can only influence.

1. Confirm access. Check that key pages are indexed and not blocked by noindex, nosnippet or max-snippet:0. Check robots.txt per bot: OAI-SearchBot controls ChatGPT search, while GPTBot controls training only. Then check that your firewall or CDN does not turn crawlers away. The AI Crawler Checker requests a page as each AI crawler and shows which ones get blocked.
2. Check rendering. Most non-Google AI crawlers do not execute JavaScript, according to Vercel's crawler data. Make sure prices, product names and key claims are in the HTML before scripts run.
3. State your facts. Put your price, your numbers and a one-line definition of what you do on pages that can be fetched. Fix any page that says it differently.
4. Make authorship real. Name the people who wrote or reviewed the content, and say what they actually know. Cite your sources. Do this because readers trust it, not to raise a score.
5. Earn mentions elsewhere. When engines cite other sites for your topic, being named on those sites is the route into the answer. Treat it as an association, not a guarantee.
6. Measure each engine separately. The free checker on this site shows whether AI engines name you for up to three buyer questions. It is a snapshot, so recheck after changes. For more on options, see free AI visibility checkers.

These are the errors we see most often in E-E-A-T advice for AI search.

• Buying an E-E-A-T audit for the score. The number has no ground truth behind it. If the audit lists concrete problems (missing authors, uncited claims, conflicting prices), the list is useful and the score is not.
• Inventing authors or credentials. A fictional expert is deception. Otterly's fictional-biologist site was the first one Google removed.
• Adding schema to "signal E-E-A-T." Google says structured data isn't required for generative AI search. Ahrefs' matched test of 1,885 pages found AI Overview citations fell 4.6% after schema was added. Existing markup does no harm, but it is not a lever. See whether schema markup helps AI citations.
• Rewriting for an "AI-friendly" style. Google's generative AI guide says you don't need to write in a specific way for AI search, and that there is no ideal page length. Our digest of Google's generative AI optimization guide and the note on ideal content length cover this.
• Assuming Google's concept applies everywhere. ChatGPT, Perplexity and Claude do not document using E-E-A-T. Each engine has its own index and retrieval rules.
• Overcorrecting. "E-E-A-T isn't real, so delete author bios" is not supported either. Real authorship still helps the people reading your page.