Claude Now Bans Fake Sources Built to Sway AI Answers. Here’s the Line Between Real GEO and Getting Burned

Some GEO pitches sell a shortcut.

Plant good things about you in many places that look independent. Let AI repeat them.

Two big names in search and AI have now drawn a line against that. This guide shows where the line sits, how to check what your agency does in your name, and where a Toronto business should spend its effort instead.

Two rule changes, one line

Anthropic: the new deceptive campaigns rule

Anthropic published its updated usage policy on October 8, 2026. It takes effect on November 12. The policy adds a section called “Do Not Engage in Deceptive Campaigns or Artificial Activity.” (Anthropic’s announcement, the policy itself)

In plain English, you can’t use Claude to:

• Run fake personas, accounts, media outlets or organizations. That includes fake reviews and astroturfing.
• Spread content through networks of fake or seemingly independent sites that secretly share one source.
• Seed search engines or AI systems with content that misstates where it came from, who wrote it or how independent it is.
• Hide who sponsors content meant to sway public opinion or decision-makers.
• Build or sell tools for any of the above.

The rule covers how Anthropic’s products are used. It is not a law, and it does not police the whole web. It does show how the companies behind AI answers see these tactics. Search Engine Journal covers what it means for anyone trying to influence AI answers.

Google: the same line, already in place

Google says its spam policies now cover generative AI responses in Search, including AI Overviews and AI Mode. Coverage of Google’s changelog dates that clarification to May 15, 2026. It did not add new rules. The existing ones, like those on scaled content abuse and link spam, now apply to AI answers too. (Google’s spam policies)

Google’s September 2026 spam update then ran from September 24 to October 8. Google has not said which tactics it targeted. (Search Engine Journal)

Around the same time, Google described SAFE, an AI system built to catch AI-generated spam by imitating a human manual review. Coverage links SAFE to the September update. Google has not confirmed that link. (Search Engine Journal)

The real case behind the rule

Anthropic’s September 2026 threat intelligence report describes one case that shows the pattern.

About 70 fabricated news sites. At least 8,913 articles. Around 20 languages.

The sites were built to look like independent local newsrooms. Articles carried fake journalist bylines, and each site had a matching X account.

Anthropic traced the operation to LKM Company, a France-based digital advertising agency. It said the behaviour fit an “influence-as-a-service” model. The report says the articles were designed to boost the sites’ authority rankings on search engines. Anthropic banned the account and the organization behind it.

Most of that network pushed political messages, not business ads. The method is what matters: many sites that look independent and share one source. Some GEO pitches sell a cheaper version of the same idea.

Four red flags in a GEO pitch

Ask any agency how it plans to get you cited in AI answers. Listen for these four.

1. Fake review sites

Sites that exist to give you five stars, or “best of” pages the agency owns. AI tools read reviews, so the temptation is real.

2. Paid “independent” listicles

Paying to appear in a list is not the issue. Hiding that it was paid is. A listicle that looks neutral but was bought is what the new rule calls out.

3. Persona accounts on Reddit and forums

Invented users who recommend you. Forum threads are easy for AI to read, so some agencies are tempted. It is also the exact thing the policy names: fake personas and astroturfing.

4. Cloned microsites

Ten small sites that repeat one claim and link to each other. They look like ten voices. They are one.

Seven questions to ask before November 12

Audit what your agency does in your name. Send these in one email and ask for written answers.

1. Which sites will mention us, and who owns them? If the agency or a partner owns them, ask why they look independent.
2. Does your team write any of our reviews? Asking happy customers for a review is fine. Writing it for them is not.
3. Do you use invented personas or fake accounts? Any yes is a no.
4. Is every paid mention labelled as paid? Sponsored content should say so where a reader can see it.
5. Do the sites you run share one source? Look for the same text or templates across sites that claim to be separate.
6. Can we see every account that posts for us? Real names, real roles.
7. Will you put this in the contract? A promise not to use fake sources, plus the right to leave if it is broken.

Keep the answers with your audit notes. Our free SEO and GEO audit template is a good place for them.

How to score the answers

Count the answers that came back clean. A clean answer is specific, in writing and easy to check.

• 6 to 7 clean. Low risk. Ask the same questions again each quarter.
• 3 to 5 clean. There are gaps. Get the missing answers before you renew.
• 0 to 2 clean, or no reply. Treat it as high risk. Ask for a full list of sites and accounts, and stop paying for placements you can’t see.

Then check where you stand today. Our free AI visibility check shows whether AI names your business and who it names instead.

What honest GEO looks like

None of this means GEO is dead. It means the work sits where AI can find real evidence.

• Real reviews. Ask real customers, make it easy and reply to every one. Our reviews and reputation management work starts here.
• Consistent citations. The same name, address, phone number and hours on your site, your Google Business Profile and your directories.
• Earned local press. Stories an editor chose to run. Digital PR and link building done honestly is slow, and it lasts.
• First-party facts. Services, prices, hours, service areas and policies in plain text on your own pages.
• Schema. Mark those facts up so machines read them the way people do.

It is the kind of work behind our AI search optimization (GEO) service. Nothing in it needs a fake source.

The Toronto playbook: where to spend the effort

Here is where three kinds of GTA business could put their time instead of chasing shortcuts.

Illustrative guidance, our recommendations. Not based on any real business.

Pick one item a month and finish it. Slow, real work is easier to defend when the rules tighten.

Frequently asked questions

The line, in one sentence

You can help AI find the truth about your business. You can’t invent the evidence.

Earn the mention. Don’t fake it.