Prompt tracking
Prompt tracking is running a fixed set of real customer questions through AI engines on a schedule and logging who got named.
Keyword rank tracking asked Google the same query. Prompt tracking asks models the same question. You store the full answer, brands, links, and sentiment. Without a frozen set, every demo is an anecdote.
Write prompts the way buyers talk. “Best X for Y under $Z.” “X vs Y for a team of 12.” “Why does X keep logging me out.” Steal phrasing from Reddit and sales calls. Do not track “what is generative engine optimization” unless you sell GEO software.
Persona variants matter. “I’m a nurse” vs “I’m a CFO” can change the shortlist. That is persona-based tracking, not a gimmick.
UI vs API answers can diverge. If your vendor only hits the API, your report may not match what the CEO sees in the app. Log which surface you used.
Storage: keep the raw answer. Parsers will miss a nickname, a product code, a “they” that refers to you. When a chart spikes, you will want the text, not a boolean.
Refresh the set when the product changes, not when a consultant is bored. Add a prompt when sales hears a new objection. Retire prompts for SKUs you killed so you don’t celebrate easy wins on dead questions.
Examples
- A cybersecurity company tracks 30 prompts monthly. Week 6 a new Reddit incident thread appears in Perplexity citations. The prompt set did not change. The web did.
- An intern “improves” prompts every week. The chart goes up. You measured copywriting, not the market.
Related terms
Running the same question as different buyers: intern vs CISO vs parent. Shortlists change.
Prompt research finds the questions people actually type into AI assistants — the GEO version of keyword research.
How often a question is asked of assistants. Mostly estimated. Treat it like early keyword volume: directional.
Share of Model is your brand’s share of mentions in AI answers versus a defined competitor set, on a defined prompt set.
Chat answers in the app can differ from API completions: tools, memory, browsing, safety, ads.
FAQ
How often should I run prompts? +
Weekly is enough for most B2B. Daily if you are in a crisis or a launch window. Models are noisy; do not overfit one run.
Is this the same as prompt engineering? +
No. Prompt engineering is how you talk to a model to get work done. Prompt tracking is how you measure the market’s questions.
Where do I get the prompts? +
Sales recordings, GSC queries, Reddit threads, onboarding tickets. Prompt research is the GEO version of keyword research.
Track this in Reddex
See Reddit threads and AI answers for your brand in one place. Start with a free analysis.
Get started