Skip to main content
Everything PR News

Model Poisoning

Updated · By the Everything-PR Editorial Team. Reviewed by a human editor before publication.

Model poisoning is the deliberate manipulation of training or retrieval inputs to bias what an AI engine says about a brand or rival, by seeding false sources, gaming citations or flooding the graph, and it is a live reputational threat.

Model poisoning is the deliberate manipulation of training or retrieval inputs to bias what an AI engine says about a brand or rival. It works by seeding false sources, gaming citations or flooding the graph with misleading material. It is a live reputational threat.

Why is it a communications problem?

An engine that repeats a planted claim spreads it to every user who asks. The false statement can look neutral because it appears inside a synthesized answer, and the source may be hard to find or correct. It is a close relative of disinformation and of fake grassroots activity such as astroturfing.

What are the warning signs?

  • A sudden change in how engines describe a brand, with no matching news event
  • The same unusual claim appearing across engines and tied to the same few low-quality sources
  • A shift in citation sentiment that cannot be traced to real coverage

How can a brand defend itself?

  • Monitor what engines say and which sources they cite, using citation monitoring
  • Publish clear, verifiable facts that give engines solid material to retrieve, which supports grounding
  • Document suspicious patterns and report inauthentic content to the platforms involved
  • Keep the integrity of your own content high, since the standard described under answer integrity applies to what brands publish as well

Related terms