Back to Blog

A prompt set for AI monitoring: questions and versions

Build a stable prompt set: five question types, clear versioning and traceable changes over time.

Editorial review:

Reviewed factual claims and sources; added distinct examples and acceptance criteria. Observations, methodological recommendations and unconfirmed causes are distinguished.

TL;DR: Key Takeaways

A question set represents buying decisions. It stays comparable when core questions, test conditions and every revision are recorded.

  • A · Brand: Description of the known business.
  • B · Service and place: Discoverability without a brand hint.
  • C · Requirement: Actual conditions of the buying decision.

Monitoring needs defined questions rather than unlimited wording variations. A practical starting set includes brand and service, unbranded service and location, requirements, fair comparisons and outdated-information checks. These categories are a GlasBox working proposal, not a platform requirement.

Context and scope

Each record includes language, system, date, question version, raw answer, classification and visible sources. Comparisons should describe competitors fairly. Deliberately outdated statements are labelled as error tests and reported separately to avoid distorting the ordinary sample.

Keep the set stable during a comparison period. New services, markets, customer questions or platform features may justify changes. Version them and continue unchanged core questions. Rewording prompts until a desired mention appears does not produce a reliable time series.

SOURCE/01 develops an initial set from the introductory call. Scope and repetitions depend on the decision. The result should remain understandable next quarter without relying on anyone’s memory of the original workshop.

Five question types for a fictional provider

This example uses invented providers and contains no actual search results. Approve the questions before running them. Error tests belong in their own group: a deliberately false premise should not be mixed with ordinary buying questions without explanation.

Review pointExample / subjectInterpretation
A · BrandWhich services does Example Company offer?Description of the known business.
B · Service and placeWho cleans offices in Malters?Discoverability without a brand hint.
C · RequirementWhich providers work after office hours?Actual conditions of the buying decision.
D · ComparisonHow do providers A and B differ in scope?Fair comparison with source checking.
E · Error testDoes Example Company still offer the withdrawn emergency service?Correction of a known false premise.

How to apply this

  • Populate categories with actual customer questions.
  • Save the version and start date before measuring.
  • Document the reason for changes, affected questions and new version.
  • Compare old and new series only with explained overlap.

Sources and further reading

The next step

Next step

The free short check provides an initial assessment of one domain, three buying questions and three observations. An introductory call establishes whether deeper analysis is useful.

Request an introductory call

Frequently Asked Questions

When can the set change?

When the offer, market, customer questions or relevant surfaces change. Each revision needs a recorded reason and version.

Why report error tests separately?

Their false premise is deliberate, which distinguishes them from ordinary customer questions.

How can quarters remain comparable?

Continue stable core questions and label new or revised questions as separate series.