Prompt and Prompt Set
The question asked of the model (prompt), and the list of questions used to measure visibility, defined once and frozen (prompt set). Measurements are only comparable if the set stays fixed.
A prompt is the input given to a language model; in the context of business visibility, it's the question a customer asks: "I want glass fitted on my balcony in Karşıyaka, who would you recommend?" A prompt set is the list of 15–30 questions chosen for measurement.
Why a set
A single question is an anecdote; engines give different answers to the same question. A set covers variety across service, area and customer type, and because the same questions are asked at every measurement, results can be compared month to month.
How to build one
- Questions are written in the customer's own words; everyday language, not industry jargon.
- The business's own name never appears in the questions; "is X any good" doesn't measure visibility.
- Service × area × need combinations are spread evenly.
- Once you've defined the set, freeze it. If it has to change, start a new measurement series; don't mix it with the old one.
At protan.ai
For every measurement, a draft prompt set is generated from the site's content, reviewed by an operator and frozen. On the monitoring subscription, the frozen set is asked in exactly the same form every month; the report shows which state each engine produced for each prompt.
Related term: Share of voice