What is a prompt set?

A prompt set is the fixed list of questions you ask AI engines, on a schedule, to measure whether they recommend your product. It plays the role that a keyword list plays in SEO. Each prompt is a question a real buyer types, such as "what is the best social listening tool for a solo founder," and the set as a whole defines the category you are measuring. Change the set and the share of model number changes with it, so the set is written once and kept stable.

How to write a prompt set

Write the questions buyers ask before they know your name. Category questions ("what should I use for X"), comparison questions ("X vs Y for a small team"), and problem questions ("how do I find people complaining about my competitor") all belong. Brand questions ("is Oetzi good") do not, because they only measure buyers who already found you. Add constraints buyers actually use: price ceilings, team size, platform. Ten to thirty prompts is enough. Fewer misses the category, more costs money without changing the trend.

Why a prompt set has to stay fixed

Engine answers vary run to run, so measurement depends on repetition. If you rewrite prompts every week you cannot tell whether a change in the answer came from the engine or from you. Keep the set, ask it to each engine on the same cadence, and record the naming rate and the citations. That is LLM rank tracking. Add a prompt when you ship something new. Retire one only when the question stops being asked. Oetzi builds a first set from your site when you paste a URL and runs it through its share of model check at two credits per question.

Where prompt sets come from

The best source is the questions already in public. Recommendation threads on Reddit and Hacker News are prompt sets written by buyers, and the same phrasing works when asked to Perplexity or ChatGPT. Search console queries and support tickets are the next best source. Avoid inventing questions that flatter your positioning, because engines answer the question asked, not the one you wish buyers asked.

Questions

How many prompts should a prompt set have?

Ten to thirty for one product category. Below ten the set misses the ways buyers phrase the question. Above thirty the cost rises while the trend line barely changes. If you sell to two distinct segments, keep a separate set per segment rather than one large mixed set, so each score means one thing.

Should the prompt set include my brand name?

No. A prompt that names you measures whether the engine knows your brand, which it usually does once you ask. The useful measurement is whether you appear when the buyer does not name anyone. Keep brand prompts in a separate check if you want to catch factual errors about your product.

Should I ask the same prompts to every engine?

Yes. Identical wording across ChatGPT, Perplexity, Gemini, Grok, and Google AI is what lets you compare engines. Each retrieves from different sources and refreshes at different times, so the same prompt will surface different names, and that difference tells you where the gap in your coverage is.