Short answer: if a publisher is not cited by Claude in an answer, the problem is not automatically "lack of GEO authority". Before conclusions, check that the article is accessible, that the topic owner is clear, that the information is current, that there are no duplicates or conflicting versions, and that the monitored question requires the type of evidence that the page provides. Anthropic documents web search and citations, but does not publish a formula by which publishers can control source selection.

Failure mode 1: the article is not the real owner of the subject

A publisher can have breaking news, explainer, guide and analysis on the same topic. If they all try to answer the same question, the information system becomes ambiguous.

Define the role of each page. The news explains the event, the explainer provides stable context, and the guide solves a practical task.

Failure mode 2: the item is stale

A page can have backlinks and traffic, but the data be outdated. It does not assume that historical reputation compensates for old information.

Check material claims, sources and date of last actual revision.

Failure mode 3: canonicalization is unclear

AMP, print versions, syndicated copies, pagination or historical URLs can create multiple versions. Check canonical, redirects and indexability before any AI analysis.

Failure mode 4: the article has sources, but no clear provenance

A text can say "according to experts" without indicating who and what evidence supports the claim. For technical, financial or scientific topics, source provenance matters editorially independent of AI.

Failure mode 5: third-party citation is more appropriate

If the question calls for primary documentation, a vendor or institution may be the more direct source than a secondary article. The fact that the publisher is not cited is not necessarily an error.

The audit must take into account the type of claim.

Public content can be hidden by paywall, consent manager or JavaScript. It checks what information remains accessible and whether the page can be identified without fragile interactions.

Don't confuse the delivery issue with the editorial issue.

Failure mode 7: author and profile are inconsistent

For editorial articles, the byline, profile and Article markup must match. Author identity does not guarantee citation, but inconsistency is real finding.

Failure mode 8: internal linking does not clarify the owner

If the new news is only related to each other, the stable explainer can become near-orphan. Google recommends crawlable links and contextual anchor text.

The graph should reflect editorial roles.

Failure mode 9: syndicated copies become more visible

Republished content may appear on other domains. Check syndication agreements, canonical handling and if the publisher's version remains clear.

It does not automatically assume that an external copy is "stolen" or that the engine ignored the original source without evidence.

Failure mode 10: the query set is too commercial

A publisher can monitor questions asking for recommendations or current availability, even though its content is explanatory. Intent mismatch produces false conclusions about the citation.

Failure mode 11: a single answer becomes a trend

Generative responses may vary. Keep the same query set and repeat observations.

It separates brand mention, owned citation, third-party citation and factual accuracy.

Failure mode 12: a proprietary tool becomes the source of truth

An "AI visibility score" can be useful internally, but requires the formula and the denominator. It wasn't presenting it as an Anthropic metric.

Reproducible decision tree

  1. Is the URL accessible and indexable as intended?
  2. Is the canonical clear?
  3. Is the page the real owner of the question?
  4. Are material claims current?
  5. Are the sources identifiable?
  6. Are the byline and author profile consistent?
  7. Don't paywall/consent hide essential information?
  8. Do internal links clarify the owner?
  9. Doesn't syndication create ambiguous versions?
  10. Does the query ask for the type of evidence provided by the article?
  11. Does the observation repeat itself over time?
  12. Can the conclusion be formulated without an invented score?

If the first five answers are negative, the problem is not "Claude citations".

How do you build the baseline

Select 20-50 queries by content type and save date, material response, sources, publisher URL and factual accuracy. In parallel, it keeps the editorial owner and review status of the page.

Do not mix all the sections of the site into one percentage.

Severity

P0: wrong or conflicting factual information. P1: wrong owner/canonical, paywall blocking public page completely. P2: author mismatch, stale links, syndication ambiguity. P3: editorial opportunities and layout.

What measurements after remediation

It measures factual accuracy, owner conflict rate, source mix, citation observations and referral separately. A correct answer that cites another source may be an acceptable result.

An example

A publisher has three articles about the same AI product: release news, an explainer, and a review. The query "how does the product work?" cite the vendor's documentation, not the publisher. This can be completely normal. The query "what issues did the independent reviewers find?" may require another type of source.

Correct diagnosis starts from query ownership, not from wanting the domain to appear in any response.

Stop criterion

Move audit to monitoring when editorial owners are clear, critical pages are current, delivery is robust, and new observations no longer produce material findings. Don't keep rewriting just to track the variation of a citation.

Claim ledger

  • FACT/EVIDENCE: Anthropic documents web search and citations to sources.
  • FACT/EVIDENCE: Google documents canonicalization, Article markup and crawlable links.
  • PRACTITIONER GUIDANCE: publisher diagnostics must separate owner, provenance, delivery and source mix.
  • NOT PROVEN: a universal Claude authority score or publisher selection formula.

Conclusion

For publishers, the lack of a citation is a very weak symptom. Wrong owner, stale article, ambiguous canonical or intent mismatch are more concrete and fixable problems. Diagnose these layers before you invent a story about authority.

Sources reviewed

Related insights