Skip to content
Try for free

Prompt monitoring: building your prompt set

In short

Prompt monitoring means regularly running a set of relevant prompts against AI systems and evaluating how often you're mentioned/cited. The foundation is the right prompt set: questions in your audience's real language, sorted by funnel stage and topic. You don't need a perfect set to start — five good prompts with ten runs each already reveal your biggest blind spots.

Why the prompt set decides everything

Prompt monitoring is the core discipline of AI visibility measurement — but it's only as good as the prompts you run. Measure questions no one asks and you measure noise. The goal is to mirror your audience's real questions: in their language, with their intent, across the whole buying process.

Where to find the right questions

Pull your audience's language from real sources rather than from gut feeling: internal searches on your website, support tickets, sales-call transcripts, frequent customer questions. Mind the difference between jargon and colloquial phrasing — both yield different AI answers. Cover different funnel stages: informational questions ("How does X work?"), comparative ("X vs. Y", "best tool for Z") and purchase-related ("provider for X in Germany").

Key principles when building

Three things to know. First: you don't have to hit the exact wording — small rephrasings yield very similar answers; topic and intent matter. Second: run each prompt many times (around 100×), because AI answers fluctuate; a single query is worthless. Third: prioritise commercially relevant prompts that trigger a live web search — they change often and are worth monitoring. You don't have to start perfect: five prompts, ten runs, look at the sources — and you'll immediately see where you're missing.

Key takeaways

  • The prompt set is the foundation; wrong prompts measure noise.
  • Use the audience's real language (internal search, tickets, sales calls).
  • Cover all funnel stages and run each prompt many times.
  • Start small: 5 prompts × 10 runs reveal the biggest gaps.

Sources

  • Knowhow_GEO_Landwehr.md (Landwehr/Peec AI podcast) — prompt-monitoring mechanics, audience language, non-determinism, low-effort start (5 prompts × 10 runs).
  • Alpar et al.: Generative Engine Optimization, Rheinwerk 2026 (prompt/intent logic).

Frequently asked questions

How many prompts do I need for prompt monitoring?

Five good prompts to start. For robust monitoring the set grows by topic and funnel — what matters is quality (your audience's real questions) over quantity.

Do I have to adjust my prompt set constantly?

No. Small rephrasings barely change the result. Revise the set when your offering, audience or their language changes.

How do I know a prompt is relevant for monitoring?

If your audience actually asks it (or similar) and it fits your offering commercially or informationally. Sources for this: internal search, support tickets, sales calls.

About the author

Christoph Schempershofe, founder of VISIBILIS. Since his studies he has combined marketing with technology: from websites and brand building through search engine marketing to AI visibility (GEO). As a lecturer at FOM and IU he teaches marketing, online and search engine marketing and content management systems.

LinkedIn