Skip to the content
searchboar.

AI search and visibility

Prompt set: meaning and practical example

A prompt set is a defined collection of questions selected for studying AI answers. It determines the topics, situations and audience represented by observations. A deliberately assembled business checklist is not automatically representative of every real user question.

Also known as: tracked question set · AI question sample

At a glance

  • Selected questions define the scope of the result.
  • Interpret branded and unbranded questions separately.
  • Changing the sample can change aggregate rates without any underlying performance change.
On this page
A bounded research sample is selected from different customer questions.
The composition of a prompt set determines which questions AI-visibility conclusions actually describe.

Why does the sample affect the result?

Questions containing your brand name examine something different from general provider-selection questions. The former checks the known description of a brand; the latter examines a limited form of discovery. Neither replaces the other.

Prompt tracking therefore starts with question design. A large set can remain biased if it repeatedly rephrases one situation. A smaller, well-organized set may reveal more useful distinctions without supporting unjustified market-wide conclusions.

How should questions be selected?

Start with the actual offer and customer decisions. Separate definitions, comparisons, specific requirements and local provider searches. Search intent helps name the distinction; business relevance determines whether to retain a question.

Keyword research can inspire questions, but Google keyword volume is not the frequency of a full sentence asked in AI. AI-suggested prompts remain ideas until evidence connects them with real customer needs.

Example: four situations for a door restorer

Imagine this deliberately designed checklist. It is neither measured demand distribution nor a universal sampling standard.

GroupExample question
UnderstandingWhat can be restored on an old wooden door?
ComparisonWhen should a door be restored rather than replaced?
ConditionsWhat determines whether work can happen on-site?
Brand checkWhat services does Example Door Workshop offer?

Like branded search, the last question supplies the company name. Mixing it with the others can inflate an overall mention rate as an apparent measure of spontaneous discovery.

What are coverage and bias?

Coverage describes which important situations are included. Bias arises when selection systematically favors an outcome: only large cities, one product family or removing difficult topics after seeing results.

More questions alone do not fix this. Small prompt edits can change the task. Label intentional variants rather than treating five near-identical formulations as evidence of five independent needs.

How do comparisons remain interpretable?

Preserve a baseline list and identify additions separately. Removing questions about a discontinued service may be reasonable, but document the change beside old and new results. Compare the unchanged subset where useful.

This keeps the mention-rate denominator understandable. Adding easier questions can raise the total even if every old answer remains unchanged. A good set makes the observational boundary explicit: which questions are covered and which are not.

Sources

  1. Ahrefs: tracking custom prompts
  2. Semrush: Prompt Tracking functionality and metrics
  3. Benjamin Tannenbaum: observational study of AI brand presence, preprint