A documented collection of prompts used for repeatable visibility testing or competitor benchmark analysis.
A prompt set is a documented collection of questions or instructions used for a defined purpose. In AI visibility work, it can provide a consistent sample for observing how a brand is represented. In an application test, it can exercise expected tasks and known failure cases. The purpose should be clear before the questions are selected.
A useful set records the wording and relevant conditions, such as platform, date, conversation context and any market assumptions. Reusing the same prompts can improve comparability, but it does not guarantee identical responses. Models, retrieval results and product behavior can change, and repeated runs can vary.
The sample also limits the conclusion. A set of ten favorable questions about your own services does not measure visibility across all possible customer requests. Include realistic needs and document how the prompts were chosen. Keep excluded or failed runs visible in the method rather than quietly removing results that make the report less impressive.
Takeaway: Use a prompt set as a transparent sample, not a claim to measure the entire market. Keep the wording, conditions and inclusion rules stable enough to interpret comparisons.
Example
A training company tests questions about course suitability, prerequisites and scheduling. It records the same task conditions over time and distinguishes a brand mention from a citation or recommendation. When it adds a new course category, it versions the expanded set instead of comparing the new total directly with an unexplained old total.
What to check
Define the outcome categories and denominator before running the test. Preserve actual responses and source links where appropriate. If the prompt set changes, explain the change and avoid attributing every difference to a website edit.
For more detail, see OpenAI prompting and evaluation guidance.