E-E-A-T and AI: What Actually Gets You Cited

E-E-A-T stands for Experience, Expertise, Authoritativeness, Trustworthiness: four notions defined by Google in its Search Quality Rater Guidelines to describe credible content. Since AI engines became real acquisition channels, one idea has spread everywhere: the proof Google expects through E-E-A-T would be exactly what ChatGPT, Perplexity and Gemini use to choose the sources they cite.

This article used to say so too. Then we measured it. In our 317-page study, the signals usually associated with E-E-A-T (named author, trust pages, Organization schema, sameAs links, expert quotes) do not separate pages cited by AI from other pages on the same site. What remains useful is experience and expertise that show in the content itself. Here is the breakdown.

What exactly is E-E-A-T?

E-E-A-T is an evaluation framework for Google's human quality raters, not a score computed by an algorithm. Google formalized it in December 2022, when the first "E" (Experience) was added to the former E-A-T. The Search Quality Rater Guidelines, a 170+-page document updated regularly, describe how these raters judge page quality.

Google says it plainly on its page about helpful content: E-E-A-T is not itself a ranking factor. The same page explains that Google gives even more weight to strong E-E-A-T on YMYL topics, those that can affect people's health, safety or financial stability. And for its AI Overviews and AI Mode, Google states that no special optimization is needed beyond the usual best practices.

What our study measured

Our calibration study of September 30, 2026 covers 317 pages from 40 sites frequently cited in Beeleven's AI visibility audits. For each site, pages actually cited as sources by Perplexity, ChatGPT or Claude were compared with uncited pages from the same site, on the same topics. It is a correlation, not proof of cause and effect, and it covers large French websites.

The result: a named author, Organization schema, Contact, About or legal pages, social media links, sameAs links, expert quotes and an up-to-date copyright do not separate cited pages. What does: a short answer (one to three sentences, 15 to 70 words) right under the headings, headings phrased as questions, tables, a recent update date and more complete content.

One nuance matters: the study compares pages within the same site. It shows that these signals don't separate two pages, not that they are useless to a site. A brand with no address, no legal notice and no contact page loses its readers' trust, and some of these pages are mandatory. Adding them is simply not what gets one page cited rather than another.

What about the Princeton GEO study (Aggarwal et al., KDD 2024), often cited in defense of E-E-A-T? It shows, in its tests, that adding statistics, quotations and sources increases visibility in AI answers by 30 to 40%. On real pages, our study finds no effect from expert quotes. Numbers and sources remain good evidence for readers: that is what the Evidence & data criterion of the Detekia score measures.

The 4 pillars: what still helps you get cited

Experience — "Have you actually done what you are writing about?"

Google defines Experience as evidence that the author has first-hand experience with the topic. A product review written by someone who used the product. A medical guide authored by a practitioner. A technical tutorial written by someone who implemented the solution. In YMYL sectors such as health or law, this requirement is at its highest (see lawyers, doctors and AI: the YMYL rules), and it is precisely what mass-produced AI text lacks (see AI-generated vs human content).

This is the most useful pillar for getting cited, as long as it shows in the text: proprietary data, concrete cases, quantified results, an explained method. An article stating "well-structured pages get more citations" is less citable than one that states "across the [N] pages we compared, those with [feature] are cited in [X]% of cases, versus [Y]% for the others". The latter is extractable and verifiable. These elements overlap with two criteria of the Detekia score: Evidence & data (numbers, linked sources, attributions) and Content depth.

Expertise — "Are you qualified to write about this?"

For Google's raters, Expertise is documented: author pages with biographies, certifications, publications. On the AI side, our study sees no gap between pages with and without a named author, nor between sites with and without Organization schema or sameAs links.

The expertise that matters for citation shows in the content: precise vocabulary, an explained method, stated limits, cited sources, a complete answer to the question asked. An author bio remains useful to your readers, and it is expected in YMYL sectors. Person and Organization schemas describe who you are; they are useful, not decisive. For implementation, see our article on Schema.org for AI. This is especially true for service providers: see how consulting firms and IT services companies demonstrate their expertise to AI through case studies rather than an author page.

Authoritativeness — "Are you recognized by peers?"

Authoritativeness is about reputation: mentions in third-party sources, inbound links, presence in reference databases (Wikidata, Knowledge Graph). It is the pillar most in line with what we observe on the AI side, but it plays out mostly outside your site.

In our study, 79 to 89% of citations go to third-party sites (comparison sites, media, public sites), not to brand sites. And according to Ahrefs (May 2025, 75,000 brands), brand mentions across the web are the factor most correlated with visibility in AI Overviews (0.664), far ahead of backlinks (0.218). Being cited by others matters more than describing yourself. To go deeper: Backlinks and GEO: why domain authority matters for AI too.

Trustworthiness — "Can I trust you?"

Trustworthiness is the central pillar of the E-E-A-T framework. It encompasses transparency (legal notices, privacy policy, contact information), technical security (HTTPS, absence of deceptive content), and factual reliability (cited sources, dated data, published corrections).

On the AI side, two things need separating. Factual reliability matters: external sources cited in the text (the method is detailed in our guide to sourcing your content to get cited by AI), dated data, a recent update date, since freshness is one of the things that separate cited pages. Trust pages and HTTPS, on the other hand, are a foundation your readers expect, sometimes a legal requirement, but they do not separate cited pages in our study.

Why the convergence idea doesn't hold

The usual argument is technical. AI engines that cite sources (ChatGPT search, Perplexity, Gemini and Google's AI Overviews) rely on search engines: ChatGPT, for instance, on Bing, but also on its own index fed by OAI-SearchBot and on other providers (Peec AI analysis). Those engines would use E-E-A-T to rank results, and AI would inherit it.

Two facts contradict this reasoning. Google itself says E-E-A-T is not a ranking factor in itself, and that its AI features require no special optimization. And on real pages, the E-E-A-T signals you can add to a site (author, schema, trust pages) do not separate cited pages.

What truly converges is simpler: Google and AI engines both look for useful, complete and verifiable content. The proof you put in the text serves both. Tags and trust pages serve your readers. The real effect still needs checking: an AI visibility audit of your brand shows whether ChatGPT, Gemini, Claude and Perplexity cite you, and who they cite instead of you. To understand what the Detekia score measures, see the complete GEO score guide.

Checklist: 10 actions, ranked by what they deliver

The first five are about content: they match what separates cited pages in our study or the criteria of the Detekia score. The next five build your readers' trust: worth doing, without expecting citations from them.

1. A short answer under every heading. One to three sentences (15 to 70 words) answering the heading's question, ideally with the heading itself phrased as a question.
2. Proprietary data and concrete examples. Numbers from your own practice, real cases, an explained method: Experience and Expertise, made visible.
3. External sources cited in the body text. Links to studies, reports, reference publications, with clear attributions ("according to…").
4. Complete content. Cover the question fully, with a table whenever you compare options or figures.
5. A visible, genuine update date. In the HTML and in metadata (dateModified in JSON-LD), changed when the content really changes.
6. A named author and an author page. Full name, bio, verifiable professional links. Expected by readers, especially in YMYL sectors.
7. Person and Organization schemas. name, jobTitle, worksFor, logo, founder. Useful to describe who you are, not decisive for citation.
8. Complete trust pages. Legal notices, privacy policy, terms of service, contact page with physical address. Some are mandatory.
9. Strict HTTPS. Valid certificate, HSTS: the technical foundation expected of any site.
10. Reviews and press mentions. On your site (customer reviews count in the External signals criterion of the Detekia score), and above all on third-party sites, where most citations go.

Illustrative example: before and after on a service page

Before

This example is fictional: it shows the changes to make, not the results of a real client. Consider a typical service page from a web agency:

• Author: "The team" (no individual name)
• No publication or update date
• A long introduction before the first useful piece of information
• Vague headings ("Our approach", "Our commitments")
• No numbers, no examples, no external sources

For an AI engine looking for an answer, nothing here is extractable or verifiable.

After

• Question headings ("How long does a website redesign take?"), each followed by a two-sentence answer
• A table comparing the options offered
• Data from real projects and a step-by-step method
• 3 external sources cited (industry studies, sector reports)
• A visible update date
• For readers: "Sarah Chen, CTO, 12 years of web development experience", with a link to her author page, and complete legal pages

The page now gives search engines and AI engines extractable answers and verifiable elements. No ranking or citation is guaranteed. To check the effect, ask AI engines the same questions before and after the changes.

Conclusion: expertise shows in the text

E-E-A-T remains a good lens for writing credible content. But it is not a ranking factor in itself, and the signals associated with it on a site do not get one page cited rather than another in our study.

What makes the difference is experience and expertise visible in the content: short answers under question headings, proprietary data, examples, method, sources, tables, recent updates. And a reputation built outside your site, where AI engines find most of their citations.

Three actions to start this week:

1. Add a short answer under every heading of your 5 most-visited pieces of content
2. Add your own numbers to those pages, and at least one external source per key section
3. Measure your starting point with a free GEO score: answer structure, depth, evidence and freshness, among others