Prompt Injection
Updated · By the Everything-PR Editorial Team. Reviewed by a human editor before publication.
Prompt injection is an attack in which instructions hidden in content an AI system reads, such as a web page, email, document or user message, cause the model to ignore its intended behavior and follow the attacker's instructions instead. It ranks first in OWASP's list of risks for large language model applications.
There are two forms. Direct prompt injection is typed into the chat by the user. Indirect prompt injection is planted in material the engine retrieves, for example hidden text on a web page, so the instruction arrives without the user ever seeing it. Indirect injection is the variant that matters most to communications teams, because it can steer what an engine says about a company or its competitors.
Platforms treat this as a security problem, not a marketing tactic, and brands that try to game answers this way risk reputational damage if it is discovered. The defensive work is monitoring: tracking what engines say about the brand through citation monitoring, watching for planted content, and correcting false claims at the source. It is closely related to model poisoning and can produce an AI-generated crisis.