LLM mention rate is the foundational metric of answer engine optimization. For a given buyer question and a given model, it is the share of answer runs that include your brand. Track it per model, per question, and over time, and you have an objective measurement of whether AI is recommending you.
Why mention rate, not single-sample checks
An LLM answer is a draw from a distribution, not a fact. Asking ChatGPT once whether it recommends a product can return very different answers across runs, even with temperature set low. Mention rate handles that by running the same question dozens of times and reporting the percentage that named you, with a confidence interval that shrinks as samples grow.
How many samples are enough
Twenty-five to fifty samples per model per question is enough to detect changes of about ten percentage points in mention rate. If you are looking for finer movement, say from 40 to 45 percent, you want a hundred samples or more. The cost is small per question because token volume per sample is low.
Common mistakes
The two most common mistakes are sampling too rarely (running once a month, then arguing about whether a 5 percent change is real) and not separating models (averaging ChatGPT and Gemini together hides the platform where you are strong or weak). Always report mention rate per model and per question, and re-run on a regular cadence so trends are visible.