Short answer: You are not designing a medical site "for Claude". You design a coherent, crawlable and verifiable public system, then you monitor whether web search answers correctly reproduce that information. Anthropic documents web search and citations, but does not publish a formula by which a layout, markup, or term density produces source selection. Classic SEO, factuality, privacy and clinical review remain the foundation.

Precondition 1: separate truth types

In healthcare, the page must clearly distinguish clinical information from operational data. Indications, risks, contraindications and medical explanations have sources and clinical review. The address, the program, the services offered and the way of scheduling have operational owners.

If these two layers are mixed, any citation audit becomes difficult to interpret.

Precondition 2: entity registry

Keep provider, facility, service, location and organization as distinct entities. A doctor may work in several locations; a clinic may have several service lines.

The registry must include canonical name, URL owner, relationships, lifecycle status and last_verified.

Precondition 3: content ownership

For each material claim, define the source owner. An educational article can explain a concept, but should not become an owner for the program or the availability of the service.

Step 1: technical SEO before GEO

Check indexability, canonical, status codes, internal links and rendering. An accidental noindex or broken page is not fixed by rewriting for AI.

Step 2: profiles and author identity

Bylines, profiles and reviewers must reflect the actual editorial process. Don't add doctors as authors just to "enhance authority". For clinical content, the reviewer role may be distinct from the author.

Step 3: source provenance

Medical claims have verifiable sources and review dates. For volatile data, keep owner and update trigger.

Do not make up percentages, studies, results or patient experiences.

Step 4: information gain

Each article has a dominant issue and makes a real difference to the existing pages. Do not create series of pages that only change the wording of the query.

Link the explainer to the service page or source owner only when this is the logical next step. Google recommends crawlable HTML links and contextual anchor text.

Step 6: external critical profiles

Select directories, provider profiles and other relevant surfaces. Only fix sources you can legitimately manage.

A stale external source does not justify the correct first-party change.

Step 7: query set

Build public inquiries on services, providers, locations and general educational information. Do not use real patient data or simulate unnecessary sensitive cases.

Stores query, language, entity and timestamp.

Step 8: factual review

For each observation, classify claims correct',incomplete', wrong',unverifiable'. Only then does it analyze the source mix.

Step 9: source mix

Separate first-party, publications, directories, review platforms and other sources. A current independent source may legitimately be more appropriate than your own domain.

Step 10: severity and ownership

P0: medical claim or mistaken identity at risk. P1: service/provider/stale location. P2: inconsistent external profiles. P3: variations without decisional impact.

Each finding has an owner and a closing criterion.

Step 11: Recheck

After the fix, rerun the exact same questions and check for first-party truth first. A single favorable response does not demonstrate complete propagation.

Step 12: regression

Rebrand, relocation, provider lifecycle, guideline update or program change must trigger recheck on dependent surfaces.

Classic SEO you don't sacrifice

  • crawlable and indexable pages;
  • canonically coherent;
  • title and H1 oriented towards intent;
  • useful internal links;
  • original and helpful content;
  • reasonable page experience;
  • structured data only where it reflects the page;
  • ownership and update triggers.

What you don't implement

Don't create "Claude schema", don't repeat mechanical branding, fabricate FAQs and promise citations. Don't turn patient reviews into clinical evidence.

Acceptance criteria

The implementation passes the gate when:

  1. priority entities have owners;
  2. clinical and operational facts are separated;
  3. technical SEO is healthy;
  4. author/reviewer mapping is real;
  5. material claims have provenance;
  6. priority external profiles have status;
  7. the query set is versioned;
  8. findings have severity and owner;
  9. regression triggers are active;
  10. external citation outcome is reported separately.

Rollback and limitations

If a consolidation loses clinical or local context, revert to the previous version and redefine ownership. If an external platform cannot be updated, mark `external unresolved'.

Crawlability and first-party consistency do not guarantee inclusion, ranking or citation in an AI product.

How do you measure

Factual accuracy, first-party conflict rate, provider/location consistency, source mix, time-to-resolution and regression rate are useful metrics. Detectable referral can be tracked separately.

Claim ledger

  • FACT/EVIDENCE: Anthropic documents web search and source citations.
  • FACT/EVIDENCE: Google documents crawlable links and structured data for Organization/ProfilePage/Article.
  • PRACTITIONER GUIDANCE: healthcare implementation must prioritize factuality, safety, ownership and privacy.
  • INFERENCE: first-party consistency can reduce ambiguity in the interpretation of entities and claims.
  • NOT PROVEN: a layout or markup tactic that directly produces Claude citations.

Conclusion

Claude web citations does not require the abandonment of classic SEO. On the contrary, a robust medical website needs the same foundation: accessible pages, clear entities, verifiable claims and owners who can keep the truth public. Citations are a layer of observation over this foundation, not the reason to warp it.

Sources reviewed

Related insights