Short answer: In an enterprise site, the absence of a Claude citation does not prove a GEO problem. Anthropic documents the web search and citations, but does not publish a source selection formula. In practice, enterprise bottlenecks frequently arise from fragmented ownership, duplicate pages, conflicting documentation, misaligned localizations, technical access, and the lack of a reproducible query set. Diagnosis must start with observable facts, not an "AI visibility" score.
Cause 1: business units publish different versions of the same truth
Marketing, docs, support and regional teams may describe the same product in incompatible ways. If first-party sources contradict each other, you can't interpret the absence of a citation as a flaw in the AI platform.
Define the actual owner and role of each surface.
Cause 2: domains and subdomains have unclear ownership
Enterprise sites can use www, docs, support, developers and regional domains. If canonicalization and internal linking do not clarify the relationship, source attribution may vary.
Maintain a registry of public areas and their role.
Cause 3: Locales are not equivalent
The EN page can be current, and the RO or DE version can keep old claims. Don't consider two URLs equivalent just because they have the same structure.
The audit must compare the material content, not just the hreflang.
Cause 4: content gated too soon
Some essential information is hidden in PDFs, authenticated portals or forms. If the public response cannot be easily verified, the problem is informational access.
Do not open private data to crawlers. Create a suitable public surface for public facts.
Cause 5: JavaScript hides critical content
Google can process JavaScript, but rendering is a separate step. Other crawlers may have different capabilities. If the home page has little content and everything depends on a widget, the robustness decreases.
Cause 6: structured data does not reflect the page
Organization, Article or Product markup may declare entities, authors or properties that do not appear or are no longer correct. Markup is not a parallel layer of truth.
Cause 7: Intent duplicates between marketing and docs
Two pages can answer the same question, one promotional and one technical. If both claim full ownership, systems may see competing sources.
Cause 8: author identity is inconsistent
On editorial content, byline, author profile and Article markup must match. A broken profile doesn't explain all citation problems, but it's a real conflict that needs to be fixed.
Cause 9: external sources are more current
A marketplace, partner or publication may have current information before the website is updated. It does not assume that the first-party must be cited if the first-party is stale.
Cause 10: the query set is changed retrospectively
If the team changes the questions until the brand appears, you no longer have a benchmark. Version prompts, language and period.
Cause 11: a screenshot becomes a trend
Generative responses may vary. Repeat the observations and separate brand mention',owned citation', third-party citation' andfactual accuracy'.
Cause 12: Enterprise change log is missing
Rebrand, launch, pricing change, migration and PR can influence outputs. Without logging, any changes can be misattributed to "GEO".
Reproducible decision tree
- Does the question have a clear first-party owner?
- Do the first-party sources contradict each other?
- Is the language and region correct?
- Is the relevant URL public and canonical?
- Is critical content robustly available?
- Does the structured data reflect the page?
- Do marketing, docs and support have distinct roles?
- Is author identity consistent where it applies?
- Are the priority external profiles current?
- Is the query set versioned?
- Do you have multiple observations, not just one screenshot?
- Does the change log explain major interventions?
If the first six answers are negative, the problem is first-party architecture, not the Claude citation.
How you build the audit
Export URL, business unit, region/language, type, canonical intent, owner, status, canonical, structured-data role, and last-reviewed. For queries, save prompt, date, sources and factual accuracy.
Keep raw evidence and don't compress everything into a composite score.
How do you prioritize
P0: critical wrong or contradictory information. P1: owner/canonical conflict for important pages. P2: localization, author identity or external profile drift. P3: cosmetic variations.
Stop criterion
Close finding when first-party truth is consistent, owners are clear, and repeated observations no longer show the identified problem. The absence of a first-party citation in a run does not automatically reopen the case.
How do you handle knowledge hubs and enterprise PDFs
Many organizations keep information in PDFs, knowledge bases or portals that do not have the same update cycle as the website. It does not assume that the newest version is also the most easily accessible to the public. For each material claim, note the source owner, version and publication date.
If an HTML page summarizes a PDF, it checks not to turn a limited condition into a general claim. In the audit, it classifies source current',source stale', summary mismatch' andaccess restricted'. These statuses are more useful than a generic citation score.
How do you separate the distribution problem from the truth problem
If Claude uses a third-party source that correctly reproduces the information, the lack of own domain in the citations is a different issue than factual accuracy. It first closes the errors of truth and only then decides whether source ownership deserves a separate intervention.
Claim ledger
- FACT/EVIDENCE: Anthropic documents web search and citations.
- FACT/EVIDENCE: Google documents crawling, rendering, canonicalization and structured data for eligible types.
- PRACTITIONER GUIDANCE: enterprise diagnostics must separate ownership, technical access and source observations.
- NOT PROVEN: a universal Claude visibility score or a public publisher selection formula.
Conclusion
In enterprise, "Claude doesn't quote us" is too vague to be diagnostic. It starts with ownership, consistency and technical access, then measures source behavior on a fixed query set. Only in this way can you separate the real problem from a convenient explanation.
Sources reviewed
- Anthropic Help Center, Using web search: https://support.anthropic.com/en/articles/10684626-using-web-search
- Google Search Central, JavaScript SEO basics: https://developers.google.com/search/docs/crawling-indexing/javascript/javascript-seo-basics
- Google Search Central, canonicalization: https://developers.google.com/search/docs/crawling-indexing/consolidate-duplicate-urls
