Question sets and measurement types
How questions are generated, the four measurement types, buyer personas and target market, and expanding and accepting candidates.
On this page
A question set is the group of user questions PeakGEO uses to test AI answers. It is not a keyword list or a system prompt template. It is the kind of question a real user would ask an AI assistant, with a task, a scenario, constraints and alternatives in mind, one intent per question. Every metric is built on the question set, so where a question comes from and which type it belongs to decide what the numbers mean.
In the console, you maintain the question set under the Questions group, on the Prompts page.
How questions are generated
When you start a new brand diagnosis, PeakGEO writes questions from the brand profile, target market, buyer persona and competitor list you confirmed:
- Derive topics: diagnosis topics are planned first. You can select, rename, add or remove them.
- Generate per topic: each topic is generated independently in the background. If one topic fails, you only retry that topic; finished topics are kept.
- Check automatically: every question goes through an independent automatic check of whether it fits the industry, role and scenario, and whether it reads naturally.
- Deduplicate: AI-generated questions of the same type that say nearly the same thing are deselected automatically. Questions you write yourself are kept.
Question count targets
A formal question set aims for 24–40 questions: at least 10 organic discovery and at least 10 organic recommendation questions, usually 2–4 brand verification questions, and at least 6 questions for each confirmed organic topic. These are coverage targets; questions are never added just to hit a number.
If the final set lacks organic discovery or organic recommendation questions, you'll be told this is a limited-scope diagnosis and the result will be marked as partially complete.
The four measurement types
Every question has a measurement type, which decides whether it counts toward the visibility metrics.
| Measurement type | Console label | What the question looks like | Counts toward organic mention / risk mention rate | Counts toward organic recommendation rate |
|---|---|---|---|---|
| Organic discovery | Discovery question | Brand-free; learning about a category, solution or approach | Yes | Denominator only |
| Organic recommendation | Recommendation question | Brand-free; asking the AI to recommend or choose | Yes | Denominator; numerator when the answer explicitly recommends your brand |
| Brand verification | Brand verification question | Names your brand or product to see how AI describes and judges it | No | No |
| Custom | Custom question | A question you create by hand in the console | No | No |
A few notes:
- Why don't branded questions count? When the question already names the brand, the answer is bound to mention it, which says nothing about whether users would discover the brand on their own. Brand verification questions are for checking whether AI describes the brand accurately.
- Automatic reclassification: an organic question that contains identity terms such as the brand name, website or a competitor is treated as a brand verification question, never as an organic visibility sample.
- Recommendation rate denominator: the numerator only comes from organic recommendation questions, while the denominator is all valid organic answers (discovery plus recommendation). See Metric definitions for the full definitions.
Buyer personas and target market
- Buyer persona: in the Confirm brand step, pick one preset or write your own. Questions are written only from the persona you choose; if you choose none, PeakGEO does not guess who the user is. A persona is a point of view on the product, not a verified customer profile.
- Target market: sets the language and region of the questions. Each question carries its region, which is recorded with every answer.
Comparisons across runs only pair answers with the same question, model, region and sampling conditions. Results after a change of persona or market belong to a different sampling scope and should not be read as an improvement or a decline. See Comparing runs and confidence intervals.
Automatic check and question status
Each question on the Prompts page shows one of these statuses:
| Status | Meaning |
|---|---|
| Checking | Saved and being checked automatically |
| Formal | Passed the check and counts as a formal sample; may include Check notes |
| Invalid | Did not meet the bar and is excluded from measurement |
| Check failed | The check itself did not complete; click Retry check |
Only questions that pass the check are used in new tests. Editing a question sends it back through the check. There is no manual review step.
Managing the question set
On the Prompts page you can:
- Switch between Theme view and List view, browse questions by theme, or search question text.
- Use New Prompt to add your own question, or edit and delete existing ones. Deleting a question also removes its answer records, trends, evidence snippets and references from related actions.
- Add to core retest set: checked questions are preselected whenever a retest is started. This only saves configuration; it never starts sampling.
- View audit results to jump to Tracking and compare each model's answers.
New or edited questions update their results after the next diagnosis or retest.
Candidate pool: expand and accept
Beyond the questions generated when the brand was created, you can add questions drawn from search demand.
- In Demand candidates, click Expand. PeakGEO pulls search-engine suggestions for your category, products and competitors, filters out navigational phrases, and rewrites the rest as natural user questions.
- New candidates are deduplicated against formal questions and earlier candidates, and each shows the suggestion it came from. Up to 40 open candidates are kept.
- Accept or Dismiss each candidate. Candidates are only suggestions; nothing enters the formal question set until you accept it.
- Accepted questions go through the same automatic check as questions you write, and only count toward measurement once they pass.
Candidates are also classified as organic discovery or brand verification depending on whether they contain brand identity terms.