Claude's new rules ban faking the sources AI cites

The clause sits in a new section of the policy called Do Not Engage in Deceptive Campaigns or Artificial Activity. Anthropic's announcement says the policy already prohibited fake-account networks and fabricated news sites, but the restrictions were scattered across several sections. The new section gathers them, applies to deceptive activity "whether political or commercial", and for the first time names search engines and AI systems.

Alongside the clause on search engines and AI systems, the section bars using Claude to create or operate fake personas, accounts, media outlets or organisations, with sockpuppets, astroturfing and fake reviews among its examples. It also bars concealing who sponsors content meant to influence decision-makers, spreading content through ostensibly independent sites or accounts to hide a common source, building tools for such campaigns, and selling any of it as a product or service.

Fake reviews were already banned. The previous version of the policy, in force from 15 September 2025, prohibited generating "fake reviews, comments, or media". It did not mention search engines or AI systems.

The policy applies to anyone who uses Anthropic's products, from people in the Claude app to businesses building on its API. If Anthropic suspects a breach, the policy says it may warn the user or throttle, limit, suspend or terminate their access.

The announcement says Anthropic has seen state media outlets, government propaganda offices and commercial firms using Claude to run networks of fake accounts and fabricated news sites. Its September threat report describes one case: roughly 70 news sites that appeared to be independent local newsrooms, traced to LKM Company, a France-based digital advertising agency.

The network used Claude to write articles and to rewrite real journalists' stories, each with three to four internal links, and the report says the articles "were specifically designed to boost their site's authority rankings on search engines". It published at least 8,913 articles in about 20 languages. Anthropic says it disrupted the operation before it could build an authentic audience, and that most of the content drew little observable engagement. The report does not say whether any of the sites reached AI answers.

That network sold political coverage to whoever was paying. The method it used, many sites that look unconnected carrying the same output, is the one the new clause names.

Google's spam policies define spam in Search to include "attempting to manipulate generative AI responses in Google Search", and treat using generative AI tools to produce many pages without adding value for users as scaled content abuse. Google acts against the websites. Anthropic's rule acts against the people using Claude to write the content, whichever search engine or assistant it is aimed at.

AI answers about a brand lean on pages the brand does not run, which is what makes faking those pages tempting. When we searched the names of 50 well-known UK brands on Google UK three times each on 6 October, 78 of the 150 searches showed an AI Overview. Google cited at least one website the brand does not run in 74 of those 78, and the brand's own site in 71. Two showed no sources we could capture.

Wikipedia, cited in 31 overviews, and Trustpilot, in 27, were the outside sources Google used most. The brands, sectors and every cited domain are in our reading of UK brand searches.

As written, the rule turns on misrepresentation: content that pretends to come from someone else, or to be independent of you when it is not. That would cover a "best of" list on a site that looks neutral but is run by you or your agency, reviews written in customers' names, or the same claims placed across several sites that hide a common owner. Pages you write with Claude and publish under your own name say where they came from.

Ask any agency or GEO supplier that produces content for you which AI tools it uses, where it publishes on your behalf, and whether any of those sites present themselves as independent of you. Selling such work as a service is itself on the banned list. Anthropic's rule binds only Claude users, but Google's spam policies apply to the pages whichever tool wrote them.

The sources that count can be earned openly: reviews from real customers, accurate comparison-site listings, coverage that names you, and corrections proposed on a Wikipedia entry's talk page. To see which pages AI reads about your category today, the free AI visibility check asks buyer questions about your business on Claude with live web search, with no login, and shows who it names instead of you and the sources it cites. For how Google picks the pages behind its summaries, see our Google AI Overviews rank tracker page.