AEO and GEO for health content: how to get cited by answer engines
Answer engine optimisation and generative engine optimisation work differently from classic SEO. A concrete playbook for lab and health content, with the structures that get extracted.
Search is splitting into two surfaces: the ten blue links, and the answer. The second surface is generated, cites a handful of sources, and ignores everything else. For health content — where users ask direct questions and want a direct answer — that shift is not marginal.
What changes
Classic SEO optimises for ranking a URL. AEO (answer engine optimisation) optimises for being extracted into an answer. GEO (generative engine optimisation) optimises for being cited as a source inside a generated response.
Three practical consequences:
- Structure beats prose. A model extracting an answer needs a sentence that stands alone, near a heading that names the question.
- Specificity earns citations. “Around 24–48 hours” is not citable. “Median turnaround is 31 hours across 34 CLIA-certified labs” is.
- Authority signals are audited. Answer engines weight verifiable facts, dates, named standards and consistent claims across pages.
The structures that get extracted
The answer-first paragraph. Directly under an H1 or H2 question, put a two-to-three sentence answer that makes sense in isolation — no pronouns pointing backwards, no “as we discussed above”.
<h2>What is LOINC 58410-2?</h2>
<p><strong>Quick answer:</strong> LOINC 58410-2 identifies Hemoglobin
measured as mass concentration in blood, reported quantitatively.
It is the code most laboratory information systems use for a standard
CBC haemoglobin result.</p>Definition blocks. Question as heading, answer as first sentence, elaboration after. This is the single highest-yield change on most health sites.
Key facts tables. Machine-readable facts with units and dates: number of assays, countries served, turnaround, compliance status. Answer engines quote tables more readily than paragraphs.
Numbered procedures. “How do I order a lab test by API” should be answered with a numbered list of steps, each one sentence. Lists survive extraction; narrative does not.
Schema that actually helps
Structured data is not decoration. FAQPage, HowTo, Product with offers, Organization with sameAs, and Article with dateModified all give an extraction system unambiguous facts. Keep them consistent with the visible page — mismatches are a trust signal against you.
Two rules:
- Only mark up content that appears on the page.
- Keep
dateModifiedhonest. Stale dates on health content suppress citation more than missing markup does.
E-E-A-T in practice, not as a slogan
For health content, experience and authority are load-bearing. Concretely:
- Name the reviewer. “Reviewed by a licensed clinical scientist” with a link to credentials beats an anonymous page.
- Cite primary sources. LOINC terms, CLIA program pages, CDC references — link them.
- Show the update log. A visible “last reviewed” date with what changed.
- Keep claims bounded. Screening versus diagnosis is not a stylistic choice; it is the difference between citable and disallowed.
The pages that earn citations
Not every page. In practice five page types do most of the work:
| Page type | Why it gets cited |
|---|---|
| Definition pages | One entity, one clear answer |
| Comparison pages | “X vs Y” questions are asked verbatim |
| Specification pages | Prices, turnaround, limits — concrete and checkable |
| FAQ collections | Mirrors the question distribution almost exactly |
| Compliance/regulatory explainers | High-trust queries with thin competition |
A lab test detail page is a specification page: LOINC code, specimen, turnaround, price, what it measures, and the API call to order it. That page can answer a dozen distinct queries because each fact is isolated and labelled.
Measurement
Rank tracking does not capture this. Track instead:
- Whether your domain appears in AI-generated answers for a fixed query set (sample monthly, manually)
- Referral traffic from assistant surfaces, segmented
- Citation of your facts in third-party content
- Branded search lift after publishing definition pages
What not to do
Do not stuff pages with question variations, do not generate hundreds of near-duplicate city or symptom pages, and do not publish medical claims a clinician would not sign. Answer engines are, if anything, more sensitive to low-quality health content than classic search — the downside is a site-wide trust penalty, not a single lost ranking.
The winning strategy is unglamorous: fewer pages, each answering one question precisely, with facts someone can verify.