Anthropic's updated usage policy, effective Nov. 12, prohibits using Claude to seed the sources that search engines and AI systems draw answers from with content that misrepresents its origin, authorship, or independence. The company published the changes Oct. 8, and violations can trigger suspension or termination of access for any user, from API developers to consumers of apps built on Claude.

The new language sits in a section titled "Do Not Engage in Deceptive Campaigns or Artificial Activity." It bans:

• "Manipulate the sources from which search engines or AI systems draw answers by seeding them with content that misrepresents its origin, authorship, or independence (e.g., networks of sites posing as unaffiliated sources corroborating the same claims)"
• Fake personas, outlets, and reviews
• Spreading content through "fake or ostensibly independent outlets, websites, or accounts" to hide their shared source
• Building tools for deceptive campaigns and selling such activities as a service

Why did Anthropic act now?

The company ties the new section to findings in its September threat intelligence report. Anthropic identified around 70 websites that appeared to be independent local news outlets, all linked to a France-based ad agency. The operation used Claude to generate articles in a consistent format, each carrying three to four internal links, for automated publishing. The report stated the articles were "specifically designed to boost their site's authority rankings on search engines."

The network published at least 8,913 articles across around 20 languages. Most drew little engagement from real audiences, and the report does not say whether the sites improved rankings or surfaced in AI-generated answers. Anthropic removed the Claude account and banned the organization behind the activity.

What changed in the policy structure?

Anthropic says the previous policy already barred fake-account networks and fake news sites, but those rules were scattered across sections on elections, fraud, privacy, and disinformation. The prior version, effective Sept. 15, 2025, did not reference search engines or AI responses at all. The new consolidated section covers any deceptive activity, political or commercial.

The October post does not mention search engines by name, even though the new clause explicitly targets manipulation of the sources AI systems and search engines cite.

What happened to automated publishing?

The previous policy listed "Media or professional journalistic content" as a high-risk use case, covering use of Anthropic's products "to automatically generate content and publish it for external consumption." The new version lists 11 high-risk areas requiring review and disclosure, including legal, medical, and financial advice and decisions about credit, housing, and employment. Publishing is no longer among them.

Anthropic says its requirements for high-risk recommendations have not changed and that the section was rewritten for clarity. The post does not explain why the publishing category disappeared.

Two rules survive intact: users may not submit "AI-assisted work without proper permission or attribution," and consumer chatbots must disclose that users are talking to AI. The new text drops the explicit ban on presenting results as "human-generated" but keeps the prohibition on making people believe they are chatting with a human.

How does this compare with Google's approach?

Google's spam policies already describe "attempting to manipulate generative AI responses in Google Search" as spam, a clarification added in a May 15 documentation update. A June spam update enforced that policy, and a Cornell Tech paper showed why enforcing it at the source is difficult. The enforcement models differ: Google penalizes websites, potentially demoting or removing them from results, while Anthropic enforces against Claude users and can limit or revoke access.

Anthropic updates its usage policy annually and says it will keep revising it as Claude's capabilities and risks evolve. Marketers and agencies have until Nov. 12 to audit whether Claude-generated content runs across networks of sites presented as independent, and to compare review and disclosure procedures against the new 11-item high-risk list.