A prompt is the question a user types into a language model. Prompt tracking means running the questions that matter to a brand against engines on a schedule and recording the resulting visibility.
A good prompt set has three groups: category questions ("best X tool"), problem questions ("how do I solve X") and brand questions ("is X reliable"). Tracking only brand questions hides your real standing inside the category.
Keeping the prompt set fixed and versioned matters too. When the set changes, measurements stop being historically comparable, so changes should be recorded with dates.
Prompt tracking is the equivalent of rank tracking
In classic SEO you track keyword positions. In generative search there is no position; you track questions instead. You fix a question set, ask it across engines on a schedule, and record whether the brand appears in the answer.
One difference matters. A classic query returns largely the same result each time, while a generative engine regenerates the answer on every request. So the meaningful figure is not a single measurement but a rate across repeated ones.
Building a good question set
Questions should come from real user language — Search Console queries, questions the sales team hears, support tickets. Keep three groups apart: questions containing your brand name (awareness), category questions (competition) and problem questions ("how do I do X").
Once the set exists, do not change it. Adding questions breaks comparability with history; if you need more, track them as a separate set.
Frequently asked
How many prompts should I track?
At least 30-40 to keep volatility manageable. In smaller sets a single changed answer distorts the picture.
How often should I measure?
Weekly or monthly, but <em>regularly</em>. Measurements taken at irregular intervals do not produce a trend. What matters is not the frequency but that the interval stays constant.
Manual or tooled measurement?
Either works. Manual is free but requires discipline to hold conditions constant: same questions, same language, and the same stated location for location-dependent ones. Tooling buys repeatability.
