Listen: Why we don't do prompt tracking.
Instead of infinite prompt variations, this measurement framework tracks canonical core entities across AI models to score brand visibility over time.
Transcript
Why try to track every single variation of a search prompt? People ask for things in infinite ways, and without their personal history or location, trying to trace every freestyle prompt is a losing battle.
Instead of chasing endless variations, we track the core entity. For example, rather than monitoring a hundred different ways to ask about running shoes, we monitor the phrase "running shoes" itself.
We feed these core entities into models from Google, OpenAI, and Anthropic. We use one fixed, unvarying prompt that asks them to recommend brands. Because the wording never changes, any shift in the results reveals an actual change in the model itself or in the web results it retrieved.
We run these tests with web search turned on, to see what the models find online, and with web search turned off, to see what they already know from their training data. Because responses can vary from run to run, we repeat these probes up to one hundred times to get a reliable, consistent score.
This isn't meant to perfectly mimic real-world chat sessions. It is a controlled measurement framework. It lets us isolate variables and see exactly how search engine optimization, or SEO, experiments affect what these models recommend and cite.