Engine policy
Anthropic's usage policy now names AI answers. Where is the line for AEO?
By Arnav Mukherjee, founder of TofuBofu · October 15, 2026
TL;DR
- From 12 November, Anthropic's rules name seeding AI answer sources with content that lies about its origin.
- The test is three words in the policy: origin, authorship, independence. Lie about one and you're over the line.
- Its September 2025 version never mentioned search engines or AI answers. This one names both.
- Google already calls manipulating its generative AI responses spam. Two vendors now draw the same line.
- Ask any AEO vendor who publishes, under whose name, and where the proof is.
I read Anthropic's new usage policy on the morning of 15 October with one question in mind: does anything we do for customers sit on the wrong side of it? That's not an abstract question for us. Our content engine drafts with Claude, so Anthropic's rules are our rules.
One bullet in it is aimed squarely at what our category works on. The previous version of the policy never mentioned the sources AI answers are built from. This one does. So this post is my reading of where the line now sits, what it leaves alone, and what it means for anyone buying AEO or GEO work.
What did Anthropic change on 8 October?
Anthropic published a new version of its Usage Policy on 8 October 2026. It takes effect on 12 November 2026. The summary post says most of the update clarifies existing rules, and that rules on deceptive activity, scattered across several sections, now sit in one: Do Not Engage in Deceptive Campaigns or Artificial Activity. The post says that section applies to deceptive activity "whether political or commercial".
The section lists six things you may not use Claude to do:
- Fake identities: Personas, accounts, outlets or organisations built to mislead people about where something came from. The policy's own examples include sockpuppets, astroturfing and fake reviews.
- Hidden sponsorship: Concealing who paid for content meant to influence decision-makers.
- Fake independence: Networks of outlets, sites or accounts that hide a common source.
- Seeding AI answer sources: Manipulating "the sources from which search engines or AI systems draw answers" with content that misrepresents its origin, authorship or independence.
- The tooling: Building or maintaining systems designed to run such campaigns.
- The service: Marketing, selling or raising money for any of the above.
The policy's own example of the fourth bullet is a network of sites posing as unaffiliated sources that corroborate the same claims. That's the oldest trick in search, aimed at the newest target.
I checked the version it replaces, effective 15 September 2025. It banned generating fake reviews, comments and media. It didn't mention search engines or AI answers once. So fake reviews aren't newly banned. What's new is that an engine vendor now names its own answer sources as something you can corrupt, and names selling that corruption as a service.
Who does the rule bind, and what does it not say?
It binds anyone who submits inputs to Anthropic's products. The policy names four groups: people using its apps, developers and businesses on its API, customers who reach Claude through a cloud provider or reseller, and the end users of products built on Claude. If an agency or a content tool runs on Claude, through any of those doors, this is its rulebook, and its customers' too.
Two things it doesn't say, and I'd hold any commentary to both:
- It's a usage rule, not a ranking rule. It governs what you may use Claude to do. It says nothing about how Claude picks which pages to cite, and nothing about penalising sites. Anyone telling you "Claude now demotes manipulative sites" is reading something that isn't there.
- It doesn't ban optimising for AI answers. The verbs are about deception: fake, conceal, pose, misrepresent. Writing a clear page that answers the questions your buyers ask is none of those.
What three words decide which side you're on?
Origin, authorship and independence. The bullet that names AI answer sources turns on whether content misrepresents one of them. That gives you a test you can run on any tactic in about ten seconds.
Run the common AEO tactics through it. This is my reading, not a ruling, and Anthropic decides what it enforces:
A comparison page on your own site, under your brand
Outside the rule
The origin is you, and it says so. Nobody is misled about who wrote it or how independent it is.
A detailed answer page drafted with AI, edited and published as your company's view
Outside the rule
The company stands behind it, and it is the company's view. Google's separate test still applies: it has to add value for readers.
A sponsored article on an industry site, labelled as sponsored
Outside the rule
The rule targets sponsorship that is misrepresented or concealed. A clear label is the opposite.
The same sponsored article with the label removed
Over the line
Concealed sponsorship of content meant to influence decision-makers is named in the policy.
Ten review blogs that look independent and all recommend you
Over the line
This is the policy's own example: sites posing as unaffiliated sources corroborating the same claims.
Reviews or forum posts written by people who never used the product
Over the line
Fake reviews and astroturfing are named, and fake reviews were already banned in the 2025 version.
See which AI engines name you today, before anyone sells you a shortcut
A free scan asks five AI engines your buyers' questions and shows each engine's answer in full.
Check your AI visibility freeHow does Google draw the same line?
Google got there first, from the other side. Its spam policies are ranking rules, not usage rules, and they now define spam to include "attempting to manipulate generative AI responses in Google Search". The page was last updated on 28 August 2026.
- Scaled content abuse: Google defines it as many pages generated "for the primary purpose of manipulating search rankings and not helping users", however they're made. Its examples include using generative AI to make many pages without adding value, and running several sites to hide the scale.
- Fan-out pages: Google's guide to its generative AI features warns against writing separate content for every variation of a query, including fan-out queries. Doing that primarily to manipulate rankings or generative AI responses, it says, violates the scaled content abuse policy.
Put the two side by side and they agree. Anthropic's test is honesty about who made something. Google's test is whether it helps a reader or only exists to be picked up. A tactic that fails either one is a tactic you're renting until someone notices. My read is that the other engine vendors will write something similar. Neither text says so, and I'll update this post if I'm wrong.
What three questions should you ask an AEO vendor now?
The policy turns a vague worry into three concrete questions. Put them to any agency or tool before you sign:
- Will anything go out that isn't under my name? Your site, your byline, or a placement that says who paid. Anything else needs an explanation.
- Will you create accounts, personas, sites or reviews that look independent of me? If the answer is yes, the vendor is now selling something Anthropic's policy says can't be marketed using its products.
- Can you show me the page you published and the AI answer that cites it? Honest work leaves a trail you can open. Our guide to evaluating an AEO agency has the full list.
Here's how we answer them, since we'd expect the same scrutiny. Articles from our content engine publish to the customer's own site, under the customer's brand. We cut LinkedIn, Reddit and X seeding out of the product on 19 August, because content meant to be seen rather than cited was never the job. We don't write reviews, for anyone. And when one of our team works inside a customer's account, they work as themselves, so every change is attributable to a named person.
The honest version of AEO was always the only one worth paying for: make your own pages the clearest answer to your buyers' questions, then earn the third-party mentions. On 12 November that stops being our opinion and becomes a vendor's written rule.
Frequently asked questions
Does Anthropic's new usage policy ban AEO or GEO?
No. It bans seeding the sources that search engines and AI systems answer from with content that misrepresents its origin, authorship or independence, and it bans building or selling tools for that. Publishing accurate content under your own name, on your own site, is outside that wording. That is our reading of the text, not legal advice.
When does Anthropic's updated usage policy take effect?
On 12 November 2026. Anthropic published the new version and a summary of the changes on 8 October 2026. The version it replaces took effect on 15 September 2025.
Who does Anthropic's usage policy apply to?
Anyone who submits inputs to Anthropic's products or services. The policy names people using its apps, developers and businesses using its API, customers reaching Claude through cloud providers and resellers, and the end users of products or services that integrate Claude. It governs what you may use Claude to do. It says nothing about how Claude chooses which sources to cite.
Were fake reviews already banned before this update?
Yes. The version effective 15 September 2025 already prohibited generating deceptive or misleading digital content such as fake reviews, comments or media. What is new in the 2026 version is a single section on deceptive campaigns that names the sources search engines and AI systems draw answers from, plus explicit bans on building tools for such campaigns and on selling them as a service.
Does Google have a similar rule for AI answers?
Yes. Google's spam policies define spam to include attempting to manipulate generative AI responses in Google Search. Its guide to generative AI search says that writing separate pages for every fan-out query, primarily to manipulate rankings or generative AI responses, violates its scaled content abuse policy.
Can I use AI to write content meant to get cited by AI engines?
Neither vendor bans that. Google asks that AI-assisted work meet its Search Essentials and spam policies, and its scaled content abuse policy targets many pages made without adding value for users, however they are made. Anthropic's rule targets content that lies about who made it or how independent it is. Content that is yours, published under your name, reviewed and genuinely useful sits outside both.
What should I ask an AEO or GEO agency after this change?
Three things. Will anything be published that is not under your own name, on a site you own or a placement that says who paid for it? Will they create accounts, personas, sites or reviews that look independent of you? And when they say they get you cited, can they show you the page they published and the answer that cites it? A no, a no and a yes is the answer you want.
Sources and further reading
- Anthropic, Usage Policy, effective 12 November 2026: the section "Do Not Engage in Deceptive Campaigns or Artificial Activity". Read 15 October 2026
- Anthropic, 2026 Usage Policy update, 8 October 2026: Anthropic's own summary of what changed and why
- Anthropic, Usage Policy, previous version effective 15 September 2025: checked for any mention of search engines or AI answers. There is none
- Google Search Central, Spam policies for Google web search: the definition of spam and of scaled content abuse. Page last updated 28 August 2026
- Google Search Central, guide to generative AI features in Google Search: the warning on fan-out pages. Page last updated 10 July 2026