Manipulating AI Search Using Reddit Is Too Easy

Researchers at Cornell University demonstrated that just about 13 words of short text is enough to manipulate the output of AI agents like ChatGPT or Google AI search into spam or fraudulent content.

According to the research, Deep-research agents that retrieve information through real-time web scraping cite user-generated content (UGC) such as Reddit or Wikipedia in about half of all queries, and about 1/4 of citations originate from these UGC sites.

Key findings of the research are as follows:

• AEO (AI Engine Optimization): The practice of brands planting false content on Reddit, Quora, and similar sites to manipulate AI tools' answers is surging.
• Wide-ranging impact: A single manipulated Reddit comment alone can contaminate the generated answers across an entire series of related AI queries.
• Vulnerability: As data from UGC sites becomes a key factor determining the reliability of AI models, platforms' content moderation issues are directly translating into security problems for AI systems.