Global companies are increasingly employing sophisticated “black-hat” tactics to manipulate artificial intelligence (AI) assistant recommendations, embedding hidden instructions in digital content to influence outputs for commercial gain.
Microsoft reported more than 50 distinct poisoning attempts from 31 companies across 14 industries within a 60-day period leading up to February 2026, targeting major AI assistants such as ChatGPT, Copilot, Claude, Google Gemini, and Perplexity.
This new form of AI recommendation poisoning involves creating content that appears to be advice for human readers but functions as training data designed to influence AI models, according to Nicholas Thompson, CEO of The Atlantic.
Unlike traditional search engine optimization (SEO) spam, AI manipulation can occur within the AI’s internal memory, retrieval processes, source selection, or reasoning, making it significantly harder for end-users to detect, Microsoft stated.
The “manipulation surface” for AI extends beyond a single website, encompassing review profiles, comparison pages, online forums, product documentation, and third-party commentary, according to Purna Virji, a principal consultant at Peec AI.
Companies like Shopify are reportedly creating listicles and other content specifically designed to ground AI models in a particular way, aiming to influence what those models remember and recommend, Thompson said.
These tactics mirror the early days of SEO spam but possess greater stealth and potential impact on AI assistant outputs, according to industry observers.
Microsoft indicated that the rapid rise of these sophisticated manipulation techniques poses a new challenge for the integrity of AI-driven information.
Source: Search Engine Journal
