The most unsettling thing about AI-generated answers is that they move. Ask the same question two weeks apart and the recommendation can change — a competitor appears, your description shifts, a fact goes stale. If you check once and assume it holds, you are working from a snapshot of a moving target. Tracking, done consistently, turns that chaos into a signal you can act on.
Decide what to measure
Vanity metrics are tempting but useless here. Focus on measures that connect to whether a customer would actually find and choose you:
- Mention rate — across a set of realistic customer questions, how often does your business come up at all?
- Recommendation rate — when it comes up, how often are you actively recommended rather than listed as an afterthought?
- Description accuracy — is what the model says about you correct and current?
- Sentiment and framing — is the tone positive, neutral or subtly negative?
- Competitive share — against your named rivals, how often does the model put you forward instead?
Build a realistic question set
The quality of your tracking depends entirely on the questions you ask. Don’t ask “tell me about [my company].” Ask what a customer with a problem would ask: “what’s a good accountant for freelancers in Paris?” or “which CRM is best for a two-person sales team?” These category-level questions reveal whether you surface when it counts, not just when someone already knows your name.
Test across multiple models
ChatGPT, Gemini and Claude are trained differently and will not answer identically. A strong showing in one is no guarantee in another. Serious tracking covers several models so you understand your visibility across the tools your customers actually use, not just your favourite one.
Track on a schedule, not on a whim
Because answers drift, a single reading tells you little. What matters is the trend line. Running the same question set on a regular cadence — weekly or monthly — lets you see whether your visibility is improving, holding or slipping, and whether a change you made actually moved the needle.
A note on honesty
Be wary of any tool that claims to show your exact “rank” inside a consumer AI app, or that reports live positions the providers do not publish. Reliable tracking is transparent about its method: it runs real questions against real models and reports what those models say, clearly distinguishing measured results from estimates. Promptly Seen takes exactly this approach — genuine model responses, clearly labelled, tracked over time — because a number you can’t trust is worse than no number at all.