Claude’s new policy: check your AI search placements

If an agency uses Claude to help get your business mentioned in AI answers, ask where it places content and who controls those sites or accounts before approving its next campaign. A higher mention count cannot answer that question.

Anthropic’s newsroom dates its policy announcement to October 8. The published replacement policy, effective November 12, explicitly prohibits using its services to seed search-engine or AI-answer sources with content that misrepresents its origin, authorship or independence. Its example is a network of apparently unaffiliated sites corroborating the same claims.

That creates a concrete owner decision: approve identifiable, accountable placements; hold work whose apparent independence the vendor cannot substantiate.

Evidence: The forthcoming policy contains that explicit search-and-AI provision. Editorial inference: Buyers of Claude-assisted visibility services should inspect the origins of placements before commissioning more. This is a procurement recommendation, not evidence that any particular agency has broken a rule.

The previous policy, effective September 15, 2025, already prohibited fake reviews, deceptive comments and falsely attributed content. November 12 should not be treated as permission to continue those practices until then.

Who needs to act? Owners commissioning Claude-assisted placement campaigns, including work delivered through an agency or software integration. The replacement policy covers app, API, authorized reseller and integrated-product users. Businesses whose workflows never use Claude are outside this particular provider-policy change.

Our reading: ordinary factual drafting for an openly identified business does not trigger this specific deceptive-seeding provision by itself. It still needs normal accuracy and approval checks. Nothing here establishes that a disclosed placement will improve rankings or earn an AI citation.

For checking the quality of a monitoring report, use our separate AI visibility tool buying checklist. This decision concerns what the supplier publishes.

Before the next approval, request one row for every proposed placement. This is The Memo’s suggested purchasing record, not an Anthropic form:

Field	Evidence to request
Destination	Exact website, account or proposed publication
Control	Publisher, account owner and any relationship to the agency
Payment	Who funds the placement and what the audience will see
Attribution	Named author or organization and the basis for that identity
Claims	Supporting records for recommendations, results or customer experiences
AI involvement	Whether Claude contributes to drafting, automation or distribution
Correction	Person responsible and a route to amend or withdraw the material

Ask for a preview showing the surrounding page, byline and disclosure. A paragraph pasted into an email cannot show how the final placement represents its source. Keep the completed register with the approved brief so a replacement account manager can understand the arrangement.

Consider a hypothetical agency offering ten “independent recommendations.” Its register reveals that it controls all ten sites. Ownership alone does not settle every policy question. But presenting those sites as unrelated corroboration is exactly the representation the owner should challenge. Request a revised proposal that accurately identifies the relationships, then review it again before authorization.

Use three decisions. Approve when the evidence supports the represented identity and relationships, and the claims are checked. Hold when ownership, sponsorship or authorship remains unanswered. Reject the proposed method when its value depends on invented customers, concealed common control or fabricated independence. These are editorial purchasing thresholds, not a platform compliance certification.

If questionable placements are already live, pause additional placements, preserve the register and request corrections through the responsible publisher. Resume only after checking the corrected presentation. Add the register to your existing AI content policy rather than commissioning another dashboard.

Based on documentation retrieved October 9, 2026, and editorial analysis. This was not a hands-on product test or an audit of any vendor.