Anthropic bans using Claude to seed AI answer sources with fake sites

Anthropic published its 2026 Usage Policy update on 8 October 2026, and from 12 November 2026 the policy prohibits using Claude to "manipulate the sources from which search engines or AI systems draw answers by seeding them with content that misrepresents its origin, authorship, or independence." The policy's example is a network of sites posing as unaffiliated sources that corroborate the same claims. The rule binds people and companies that use Claude, not search engines, and it targets deception about who is behind content rather than content production for AI visibility as such.

What Anthropic announced on 8 October 2026

In its announcement of the 2026 Usage Policy update, Anthropic says most of the changes clarify existing rules and add examples showing how they apply to Claude's newer capabilities. The updated policy takes effect on 12 November. The post covers deceptive campaigns, elections, weapons, surveillance, high-risk use cases, hardware that takes physical actions, abusive behavior toward its models and the Supported Regions Policy.

For marketers, the change that matters is a new section titled "Do Not Engage in Deceptive Campaigns or Artificial Activity." Anthropic writes that it has seen state media outlets, government propaganda offices and commercial firms using Claude to run networks of fake accounts and fabricated news sites. Its policy already prohibited that activity, but the restrictions were scattered across the elections, fraud, privacy and disinformation sections. The new section consolidates them and applies to deceptive activity of any kind, "whether political or commercial."

The announcement itself does not mention search engines. The search clause sits in the policy text, published on Anthropic's Usage Policy page with an effective date of 12 November 2026. Search Engine Journal compared that text with the previous version, which took effect on 15 September 2025, and reports that the earlier policy did not reference search engines or AI responses.

Who the Usage Policy applies to

Anthropic's Usage Policy applies to anyone who submits inputs to its products or services. The policy lists individuals using its apps such as Claude.ai and Claude Code, developers and businesses using its API and developer platforms, customers reaching Claude through cloud providers and authorized resellers, and the end users of products that integrate Claude. A marketing tool built on Claude therefore puts its own users under the same rules, even if they never visit an Anthropic website.

Enforcement is account-level. The policy says Anthropic's Safeguards Team runs detection and monitoring, and if it suspects a violation it "may warn you or throttle, limit, suspend, or terminate your access" to Anthropic's products and services.

What the deceptive campaigns section prohibits

Under "Do Not Engage in Deceptive Campaigns or Artificial Activity," the policy bars using Anthropic's products to do six things:

• Create or operate fake personas, accounts, media outlets or organizations, or impersonate real ones, to mislead people about origin or attribution, how widely a view is held, or whether they are dealing with a human. The policy's examples are sockpuppets, astroturfing, fake reviews and bots that claim to be human.
• Misrepresent or intentionally conceal the sponsorship of content meant to influence public opinion, officials or other decision-makers.
• Distribute content through networks of fake or "ostensibly independent outlets, websites, or accounts" to hide that it comes from a common source.
• Manipulate the sources search engines or AI systems draw answers from by seeding them with content that misrepresents its origin, authorship or independence.
• Build, improve, maintain or contribute to tools, systems or infrastructure designed for deceptive campaigns, including astroturfing, sockpuppet account creation, coordinated inauthentic behavior or scaled impersonation.
• Market, sell or solicit funding for any of the above as a product or service.

The last line is the one vendors should read twice. Selling these tactics as a service is listed as a violation in its own right, separate from running them.

Dates and figures behind the update

The figures behind the update all come from Anthropic's own announcement, policy text and September 2026 threat intelligence report.

The September case behind the new section

Anthropic ties the new deceptive-campaigns section to its September 2026 threat intelligence report. One case in that report describes what Anthropic calls a commercial "influence-as-a-service" operation. Anthropic says it identified and removed an account that used Claude to mass-produce and rewrite political content, distributing fabricated news stories across approximately 70 fabricated news websites. The sites were amplified by 70 matching X accounts and more than 250 inauthentic commenting accounts. They were set up to look like independent local newsrooms, and Anthropic traced the operation to a France-based digital advertising agency.

The mechanics read like a content-marketing pipeline pointed at the wrong goal. According to the report, the network used Claude to build a standardized content pipeline: every prompt demanded a fixed JSON output structure, formatted HTML, exact character limits and three to four internal links per article, which let the actors generate and publish at scale. Anthropic writes that the articles "were specifically designed to boost their site's authority rankings on search engines." The articles carried the names of journalists who, Anthropic found, did not exist.

The infrastructure gave the network away. The report says the domains were registered from France within a ten-week window in mid-2025 and hosted on shared infrastructure behind a single deployment. That is how investigators connected roughly 70 sites that looked independent to one operator account.

The outcome was modest. The network published at least 8,913 articles in about 20 languages, but Anthropic says most of the content generated little observable engagement from real audiences, and it rated the activity Category Two on the Brookings Institution's Breakout Scale. Anthropic banned the account and the organization associated with it. The report does not say whether the sites gained search rankings or appeared in AI-generated answers, a gap Search Engine Journal also points out.

Where legitimate GEO ends and manipulation begins

The clause does not ban using Claude to write content, to structure pages for AI answers, or to publish at volume. Its trigger is misrepresentation: content that is false about its origin, its authorship or its independence. The policy's own example, sites posing as unaffiliated sources corroborating the same claims, describes a tactic aimed at how answer engines treat agreement across sources as a signal.

A workable test, which is an interpretation and not Anthropic's wording, is whether a tactic would still work if every reader knew who wrote the content and who paid for it. Tactics that pass that test sit on the legitimate side:

• Clear, well-sourced pages on your own site, under your own brand.
• Content structured with direct answers, tables and FAQs so AI systems can quote it accurately.
• Original research or data published under your name and offered to real publications that decide independently whether to cover it.
• Guest articles with your byline and affiliation visible.

Tactics that depend on hiding the common source match the policy text directly: networks of sites that look independent but share an owner and repeat the same claims, invented bylines, fake reviews, and the same message pushed through ostensibly independent outlets. The policy's disinformation section separately bars creating fake personas or accounts to falsely attribute content or mislead audiences about its origin.

How the rule lines up with Google's spam policies

Google's published spam policies for web search already cover neighbouring ground. Google defines spam to include "attempting to manipulate generative AI responses in Google Search," and a 15 May 2026 entry in Google's Search Central documentation updates log clarified that the spam policies apply to generative AI responses in Search. Its scaled content abuse policy targets generating many pages mainly to manipulate search rankings rather than to help users, however those pages are made. Its site reputation policy covers third-party content published on a host site mainly to benefit from that host's established ranking signals. Its link spam rules cover links created mainly to manipulate rankings.

The two systems differ in who they punish. As Search Engine Journal puts it, Google enforces its rules on websites, which can rank lower or drop out of results, while Anthropic enforces its policy against Claude users by limiting or revoking access. Anthropic's rule changes nothing about how Google, ChatGPT or any other system ranks or cites a page. A site network built with a different AI model is outside Anthropic's reach, though search engines' own spam rules apply whatever tool wrote the pages.

Other changes in the update that touch marketing

The new policy's High-risk Use Case Requirements list 11 areas, including legal, medical, finance, credit, insurance, housing and employment. In those areas, a qualified person must review an AI recommendation before it reaches someone who might rely on it, and the affected individual must be told that AI was used. Anthropic says these requirements have not changed and that the section was rewritten to state more directly which recommendations are covered.

Search Engine Journal reports that 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. Publishing is not on the new list, and Anthropic's post does not explain why. That removal does not loosen the deceptive-campaigns rules: automated publishing that hides its source is still covered by the new section.

Two further changes are relevant to political and civic advertisers. The elections section is renamed "Do Not Undermine Democratic Processes" and focuses on deceiving voters or disrupting elections, and Anthropic removed its blanket prohibition on personalized vote and campaign targeting, saying deceptive targeting and misuse of voters' personal data remain prohibited under other sections.

What Anthropic did not say

• The 8 October announcement does not mention search engines or AI answers; that language appears only in the policy text.
• Anthropic does not describe how it will detect source seeding specifically, beyond its general detection and monitoring.
• The threat report does not say whether the 70-site network gained rankings or appeared in AI answers.
• Nothing in the documents suggests other AI companies or search engines are adopting the same wording.

What this means for Thai marketers

If any content work for your brand runs through Claude, directly or inside a third-party tool, the new rules apply from 12 November 2026. Search Engine Journal suggests checking by that date whether any content created with Claude, in-house or by a vendor, is published across networks of sites presented as independent of each other. Ask vendors plainly how they build AI visibility, and whether any sites in their plan are owned by them or by you without saying so.

The stronger long-term position does not need hidden networks at all. Generative engine optimization built on your own brand, verifiable sources and pages that answer questions directly is compatible with Anthropic's rule and with Google's spam policies. If you want a structured plan for AI SEO that stays on the right side of both, that is the place to start.

FAQ: Anthropic's rule on seeding AI answer sources

Does Anthropic's policy ban using Claude for SEO or GEO content?

No, the policy bans content that misrepresents its origin, authorship or independence, not SEO or GEO work in general. Writing and optimizing pages under your own name is not what the clause describes. Creating sites that pose as unaffiliated sources to corroborate the same claims is.

When does the new rule take effect?

The updated Usage Policy takes effect on 12 November 2026. Anthropic published it on 8 October 2026, and the previous version had been in force since 15 September 2025, according to Search Engine Journal.

Does this change how Google or ChatGPT rank my site?

No, it is Anthropic's usage policy and it binds Claude users, not search engines or other AI systems. Google applies its own spam policies to websites, whichever tool produced the content.

What happens if someone breaks the rule?

Anthropic says it may warn, throttle, limit, suspend or terminate access to its products and services. In the September 2026 case, it banned both the account and the organization behind the 70-site network.

Is running several brand websites a violation?

Not by itself, based on the policy text, which targets sites that misrepresent their independence or hide a common source. Openly branded sites that do not pose as unaffiliated outlets are not what the example describes, though Anthropic has not addressed that case directly.

Anthropic's update draws a line that many AI-visibility tactics will have to sit on one side of: content can be produced at scale, but it cannot lie about who is behind it. If you want to review your current GEO or content approach against that line, Relevant Audience can help you map what to keep and what to change before 12 November.