Guaranteeing AI Visibility Is Impossible: Here's What Brands Should Measure Instead

The author, who works with AI brand visibility, received three commercial offers promising to place brands in the top of ChatGPT answers, even with a specific position within 30 days. He explains that such guarantees are physically impossible because AI models generate text probabilistically, word by word, and even with temperature set to 0 and a fixed seed, slight variations in server computations and batching lead to different outputs. Moreover, modern models are ensembles of smaller models, adding another layer of unpredictability. Additionally, AI answers depend on both the model's training data and live web search, both of which change independently and cannot be controlled by a brand. Instead, the author suggests measuring Share of Model, the percentage of controlled prompts where a model mentions a brand without a direct hint, by running a sample of 15-20 real prompts before and after content optimization. He references a 2024 study by Aggarwal et al. that showed a 30-115% increase in citation frequency from proper optimization. He also warns about red flags in contractors, such as promising specific positions or unified growth percentages across different platforms, and advises asking for reproducible measurements rather than position guarantees.