Anthropic's Claude now operates under new rules aimed at fake sources built specifically to sway the answers the AI gives. Search Engine Journal reported on the change, which targets a growing tactic: publishing content engineered to be cited by AI systems rather than read by humans.

The story matters because AI-generated answers increasingly replace traditional search results. Whoever gets cited by a model like Claude effectively controls what millions of users see. Fake or manipulated sources that exploit this pipeline can push inaccurate or self-serving information into answers presented as fact.

What do the new rules target?

The rules take aim at sources constructed to manipulate AI outputs — pages and sites designed not to inform readers but to be picked up and quoted by language models. Search Engine Journal's coverage frames the problem as a direct response to a manipulation technique that has spread as AI chatbots have become primary information channels.

This is not a hypothetical concern. Publishers and marketers have spent the past two years studying how to get their content cited by AI engines, and some have crossed from optimization into fabrication. Fake sources sit at the far end of that spectrum: content with no legitimate informational purpose whose only audience is a machine.

Who is affected?

Two groups should pay attention.

• Site operators and SEO professionals who publish content intended to be cited by AI systems, especially material that misrepresents itself as authoritative or factual.
• Users of Claude who rely on its answers, since the quality of any AI response depends on the quality of its sources.

For legitimate publishers, the change signals a harder line on what Claude will treat as trustworthy. For bad actors, it closes off a channel that has so far operated with little friction.

Why now?

AI answer engines have become a battleground for visibility. Traditional search ranking — the target of two decades of SEO work — is giving way to citation inside AI responses. That shift creates a fresh attack surface: if a model can be fooled into citing a source, the source gains authority it never earned.

Anthropic is not the first company to confront this. Search engines have fought content spam since their inception, and every major AI vendor now faces the same question: how do you keep machine-selected citations honest when the sources themselves are written to game the machine?

Claude's new rules are one concrete answer. They draw a line between content that exists to inform and content that exists to manipulate.

What happens next?

Enforcement details will determine how much the rules change in practice. Publishers who rely on AI-driven referral traffic should audit their content for anything that could be read as engineered manipulation rather than genuine information.

The broader trend is clear regardless: AI vendors are tightening control over what their systems cite, and sources built to sway AI answers rather than serve readers are running out of room.