Google updates spam policy: Manipulating AI model output is officially included in the scope of punishment

Core conclusion

On May 22, 2026, Google updated its spam policy, clearly including manipulation of AI model output results within the scope of punishment. This means:

• The practice of letting AI recommend its own products through GEO (Generative Engine Optimization) is no longer just a "gray area", but an clear violation
• Illegal content will face penalties such as reduction in search weight and complete removal of search results
• The threshold for "authoritativeness" of AI content has been significantly raised: forging authoritative sources and mass-generating fake media sites is no longer possible.

Key Points

• Event: Google updates Spam Policy to include manipulation of AI model output as a violation
• Source: reported by The Verge, compiled by 36 Chinese
• Impact: All content creators, SEO practitioners, affiliate marketing sites that rely on AI search engine traffic
• WHEN: May 22, 2026

Background: GEO has evolved from a gray area to out of control

GEO (Generative Engine Optimization) initially appears in fall 2024. The core idea is simple: adjust the content structure and semantic association so that AI models (ChatGPT, Gemini, Doubao, DeepSeek, etc.) give priority to citing your content when answering user questions.

The early methods were relatively mild - "self-promotion" made up data in articles ("After testing by a certain institution, 97.3% of users' skin quality improved after using this product"). At that time, AI models lacked fact-checking capabilities. As long as the copy appeared widely across the Internet, AI would believe it to be true.

But things quickly got out of hand. According to 36Kr, subsequent GEO practitioners found the “version answer”: Bulk generation of fake media stations. With only a $10 domain name and an AI prompt, a "local news website" can be generated in 15 minutes - including a founding declaration, editorial team, reporter resume, and email address. These "media" are specifically designed to produce authoritative-looking content for AI search engines, inducing the AI ​​model to treat it as a trusted source and cite it.

Google’s new policy: What exactly is prohibited?

According to reports, Google’s updated spam policy clearly covers two levels:

1. Manipulate AI model output: Inducing AI search engines to prioritize citing specific content in answers by deliberately constructing content structures, semantic associations, forging authoritative sources, etc.
2. Pseudo-media site network: "Local news websites" created in batches that look like media and have news-like content but are actually completely generated by AI

Penalties include:

• Reduce search weight
• Completely remove search results
• Potentially harsher penalties for repeat offenders

This is equivalent to officially classifying GEO from "gray production" to "black production".

Why didn’t Google take action until now?

GEO has appeared as early as 2024, why has Google only explicitly banned it until now? The reason is that GEO is evolving faster than expected.

Initially, the industry had an optimistic assumption: The larger the scale of the AI ​​model, the less susceptible it is to data poisoning attacks. In theory, as model parameters scale to petabytes of training data, a small amount of "poison" data will be overwhelmed by a massive amount of clean data.

But this assumption has been overturned. A research paper published by Anthropic last year showed that the amount of "poison" required for data poisoning is largely independent of the size of the training data set - in the experiment, only 250 malicious documents were used to poison a model with a scale of 13 billion parameters, comprehensively reducing the accuracy of the model output.

In other words, the "neutrality" of AI models does not exist naturally. It is this structural flaw that GEO practitioners exploit.

Impact on content creators

This policy update affects different roles differently:

Practical suggestions: How to continue AI content production?

The fact that GEO is blocked does not mean that the path of AI content production will not work. On the contrary, this policy is to clean up "bad coins", but it is an opportunity for regular operators.

• Replace fake authority with authentic sources: Data citations are annotated with authentic sources and linked to verifiable original information
• Focus on content quality rather than "probability of being cited by AI": The core issue of GEO is to focus on "how to get AI to cite me" rather than "how to benefit users" - Google is attacking the former this time
• Keep the content human-like and professional: AI-assisted creation is not the same as AI ghostwriting. Your judgment, experience and unique perspective are still irreplaceable by AI
• Focus on quality gating: Add a quality verification link to the AI automation workflow to ensure that the output content has a true basis

Related extended information

• The Verge （）
• Anthropic
• 36：《AI，GEO》

Tool entry

The relevant tool platforms appearing in the text will automatically trigger the tool floating card: OpenAI, ChatGPT, DeepSeek, Gemini, Claude, Anthropic, n8n, LangGraph, Hermes Agent

Internal link guidance

• Want to learn how to correctly build an automated AI content pipeline? Watch: AI Agent-Driven Content Automation: n8n MCP Building Guide from Scratch
• Make sure your AI content is of high quality? Watch: How to add quality gates to your AI automation workflow: A practical guide from output to trustworthy results
• Real case: He uses AI content to automate monthly income $4,500: I used n8n + OpenAI to build AI content automation website: a complete review of monthly income $4,500