E-E-A-T and YMYL for clinical websites: what the framework actually asks for

Two acronyms govern whether medical content ranks, and both are widely misunderstood in ways that produce expensive, useless work.

YMYL — Your Money or Your Life. A category of topic, not a quality standard. Content that could affect a person's health, safety, financial stability or wellbeing. All clinical content is YMYL by definition.

E-E-A-T — Experience, Expertise, Authoritativeness, Trustworthiness. The framework Google's human quality raters use to assess pages, particularly YMYL ones.

The most consequential misunderstanding first: E-E-A-T is not a ranking factor. There is no E-E-A-T score. It is a description of what Google is trying to measure, published so that raters assess consistently and so that publishers understand the target. The algorithms attempt to approximate it through signals they can actually compute.

That distinction is not pedantry. It determines what you should build, because you cannot optimise for a rubric — you can only supply the computable signals that stand in for it.

Why YMYL classification changes the arithmetic

For an ordinary query, a mediocre result costs the user a few wasted minutes. For a health query, a wrong result can cost considerably more. Google's own guidance directs raters to apply the strictest standards to YMYL content and to assign the lowest ratings to YMYL pages lacking evident expertise.

Two practical consequences.

The floor is higher. Content that would rank adequately in an unregulated category can fail entirely here. "Good enough" writing about project management ranks; good enough writing about chemotherapy does not.

The ceiling is reachable. Because the bar excludes so much, a clinical practice that does the work properly faces a genuinely thin field. This is the reverse of most categories, where effort produces diminishing returns against entrenched competition.

The same asymmetry operates in answer engines, more sharply. A system deciding whether to name an individual doctor in a health answer has direct exposure if it is wrong. Its caution is your opportunity, because most of your competitors have supplied nothing that would let a cautious system proceed.

The four components, and what each actually requires

Experience — the newest and most misread

Added to the framework in December 2022. It asks whether the author has first-hand experience of the subject.

For clinical content this is the easiest component to satisfy and the one most often thrown away. A surgeon who has performed two thousand of a procedure has experience no content operation can manufacture. The failure is not absence — it is that the experience is nowhere stated in a form anything can read.

What to supply: the clinician's own account. Case selection they use. What they have observed across their series. Where their practice differs from the textbook and why. First-person clinical writing, which reads entirely differently from the third-person marketing prose that occupies most practice sites.

A ghostwritten article about a procedure demonstrates no experience. The same clinician spending forty minutes dictating what they actually do, transcribed and edited, demonstrates all of it.

Expertise — credentials, in machine-readable form

Formal qualification to speak on the subject. For medicine this is unusually well-defined and unusually verifiable: registration with a medical council, specialist qualification, institutional appointment.

What to supply: a Physician entity carrying the registration number as a structured identifier naming the issuing council, medicalSpecialty as a schema.org enumeration, affiliation, and hasCredential. Then author and reviewedBy on every piece of clinical content referencing that entity by @id.

The gap here is almost never the credential. It is that the credential exists as text under a photograph.

Authoritativeness — what others say

Reputation beyond your own domain. Whether the wider world treats this person as a source.

What to supply: sameAs links to independently maintained records — PubMed author queries, ORCID, medical council registers, the institution's own staff listing, society office-holdings. Then, over time, citation by others: press, guidelines, other clinicians' reference lists.

This is the slowest component. It accrues at the speed of the outside world and cannot be accelerated with budget, which is precisely why it is the durable moat.

Trustworthiness — the one Google calls most important

Google's own guidance identifies trust as the most important member of the family, with the other three supporting it.

For a clinical site, trust is mostly operational rather than editorial:

• Transparent identification of who is behind the site, and their credentials
• Contactability — a real address, a real phone number, a named person for complaints
• Accurate, current information with visible review dates
• No deceptive patterns, manufactured urgency, or claims you cannot substantiate
• Clear disclosure of commercial relationships and conflicts of interest
• Correct handling of personal data, with a privacy notice a patient can actually read

The conflict-of-interest point is underused. A practice that publishes its financial structure — what it earns from, what it does not — is supplying a trust signal almost nobody else offers. Where a service earns from interpretation and never from treatment, saying so plainly is both true and structurally persuasive.

How the framework maps to computable signals

Since E-E-A-T is not directly measurable, the practical question is which signals plausibly stand in for it. Our working map:

Note that almost every row resolves to structured data or a verifiable external link. That is not a coincidence. Structured data is the only channel through which a page can make an unambiguous, checkable claim about a person's qualifications.

Which is why we keep returning to the same unglamorous point: the entity work is not a technical nicety alongside the content strategy. In a YMYL category it is the content strategy's delivery mechanism.

What this means for answer engines specifically

The AEO layer sits on top and asks something slightly different. Classical ranking asks is this page good enough to show. An answer engine asks is this claim safe enough to repeat, and is this entity safe enough to name.

The additional requirements:

Extractability. A model citing you lifts a passage. That passage must survive removal from its context. One question per section, answered completely, with the answer in the first sentence.

Unambiguous factual sentences. "Lu-177 PSMA therapy is indicated for metastatic castration-resistant prostate cancer after progression on androgen receptor pathway inhibition" is liftable. "We offer world-class personalised treatment options" is not — there is nothing in it to extract.

Resolvable entities. Covered above, and it is the gate. A model cannot attribute a clinical claim to a doctor it cannot resolve.

Corroboration the system can check. citation[] to indexed literature converts your claim into a claim about a claim that exists in a database somebody else maintains.

The failure modes we see most

Medical review theatre. A "Medically reviewed by Dr X" line in the HTML, with no reviewedBy markup, no review date, and — on inspection — no actual review process. This fails on every count including the ethical one, and in Australia it edges toward misrepresentation.

Author entities that do not resolve. A byline linking to a profile page with no structured data. The name is there; the entity is not.

Content volume as a proxy for authority. Forty unbylined posts build topical breadth on a domain and attach expertise to nobody. In a category where the citable unit is a person, domain-level topical authority is the wrong asset.

Credentials that only exist on your own site. Self-assertion without corroboration. One sameAs link to a council register does more than a page of qualifications in prose.

Stale clinical content with no review date. A page about a treatment protocol written in 2021, unreviewed, with no date visible. This is a trust problem before it is a ranking problem, and it is the most common finding on established hospital sites.

What to do first

If you have limited capacity, in order:

1. Build one complete Physician entity per clinician. Two hours each, once. Registration identifier, specialty enumeration, affiliation, knowsAbout, sameAs.
2. Add reviewedBy and lastReviewed to every page making a clinical claim. Only after the review has actually happened.
3. Publish three pieces the clinician genuinely authored. First person, case selection, what they actually do.
4. Add citation[] to the pages carrying the strongest clinical claims. Verify every identifier resolves.
5. Fix the site-level trust surface. Named contact, real address, complaints route, privacy notice, conflict disclosure.

Steps one and two are a week. Step three is a quarter. Steps four and five are an afternoon each and are the ones most often postponed indefinitely.

The honest summary

E-E-A-T is not a thing you can optimise. It is a description of what a careful assessor would look for, and the useful response is to actually be the thing rather than to signal it.

For clinical practices this is unusually tractable, because the underlying substance already exists. Registered clinicians with real qualifications, genuine experience and verifiable institutional affiliations are not manufacturing credibility — they are failing to publish it in a form anything can read.

Most of the work is translation, not creation. Which is fortunate, because in a YMYL category the alternative — manufacturing the appearance of expertise — does not survive contact with systems designed specifically to detect it.

Sources

• Google Search Quality Rater Guidelines — the current edition, for YMYL definition and the E-E-A-T framework. Read the source rather than summaries of it.
• Google Search Central, Creating helpful, reliable, people-first content — the public-facing statement of the same principles.
• SparkToro / Similarweb zero-click study, June 2026 — 68.01% of US Google searches ended without a click in Jan–Apr 2026.
• Bain & Company / Dynata — 83% zero-click rate where an AI Overview is present.
• schema.org vocabulary for Physician, MedicalWebPage, MedicalAudience, reviewedBy, citation.

Verify each against the primary source before republishing. Guidelines are revised without announcement.