A practical way to keep AI visibility measurement connected to the same buyer questions over time
A Frozen Question Set is Right Angle’s term for a version-controlled list of buyer questions and observation conditions chosen before measurement begins. It matters because a business cannot tell whether its standing improved if the questions, platforms, or rules keep changing with the results.
TL;DR
A Frozen Question Set holds the buyer questions reasonably constant so later observations can be compared against the same test.
The questions should reflect real moments of discovery, evaluation, comparison, and validation, not prompts designed to make the business appear.
The method measures observable signals such as appearance, representation, citations, and recommendation. It does not directly measure human trust or prove buyer preference.
The set can evolve, but changes should be dated, versioned, explained, and kept separate from comparisons made under the earlier version.
Recommendation Marketing Needs a Consistent Test
Recommendation Marketing is the work of making a business easier for buyers and AI systems to understand, verify, and recommend. That goal creates a measurement question: how do you know whether the business is gaining stronger standing in the questions buyers actually ask?
The answer cannot be one favorable screenshot. It also cannot be a collection of prompts selected after seeing which ones produced the best result. If the questions move whenever the answer is inconvenient, the business may be measuring its ability to choose favorable tests rather than any change in its market standing.
This matters because AI-assisted research is already part of consequential buying. Gartner surveyed 645 B2B buyers in August and September 2025 and found that 45% had used generative AI during a recent purchase, primarily to gather information about vendors and products. Those buyers used an average of seven information sources, and 69% preferred to validate AI-generated insights with a sales representative.
Source: Gartner, May 20, 2026
The point is not that AI has replaced the buyer or the seller. It is that AI can now influence which businesses enter consideration and how they are initially understood. Recommendation Marketing therefore needs a way to observe that evaluation without changing the test to fit the desired story.
What Right Angle Means by a Frozen Question Set
A Frozen Question Set has three parts.
The questions: a defined list of natural-language buyer questions selected before the first recorded run.
The conditions: the platforms, prompt wording, timing, context, and fields recorded each time.
The change rule: additions, removals, or revisions are dated and versioned rather than introduced silently.
The word frozen does not mean permanent. Buyers change. Markets change. Services change. New questions emerge. The freeze means that the current version is protected during its measurement period. When the set changes, the record shows what changed and prevents the new version from being treated as though it were identical to the old one.
This is not presented as an industry standard. It is Right Angle’s name for a practical control that protects Recommendation Marketing measurement from a moving test.
The Questions Must Come from the Buyer Decision
A question belongs in the set because it represents a meaningful buyer decision, not because it is likely to mention your company.
Pew Research Center’s analysis of Google activity from 900 U.S. adults found that 18% of observed searches in March 2025 produced an AI summary. Searches beginning with question words such as who, what, when, or why produced an AI summary 60% of the time. The study concerns Google behavior rather than B2B purchasing, but it reinforces a practical AEO principle: question-shaped discovery increasingly produces generated answers.
Keyword research still matters. It can reveal topics, demand, language, and search behavior. But a Frozen Question Set should go further by preserving complete buyer questions with enough context to expose the decision underneath the search.
A useful set should cover several moments in the journey:
- Problem recognition: What is causing this problem, and what are the realistic ways to solve it?
- Approach evaluation: What should a company like mine look for in a solution or partner?
- Provider consideration: Which businesses have credible experience with this kind of problem?
- Comparison: How do the leading approaches or providers differ for our situation?
- Validation: What evidence should I examine before trusting the recommendation?
Those questions connect the measurement to Recommendation Marketing. They test whether the business has earned a place in the buyer’s consideration, not merely whether its name can be produced by a prompt.
Why Freezing Makes Change More Interpretable
A reader recently asked how many repeated searches are needed before a pattern becomes meaningful. There is no universal number that works across every platform, question, market, and decision. Repetition helps only when the conditions are documented and sufficiently consistent to support comparison.
If ten different questions produce ten different answers, the business has ten observations. If the same buyer question is repeated under recorded conditions, the business can begin to see what holds, what changes, and which result deserves further investigation. That still does not create statistical certainty, but it produces a more interpretable record.
The frozen set also binds the measurer. It reduces the ability of the agency, platform, or business to remove an unfavorable question, add a flattering one, or redefine success after seeing the result. The method does not guarantee objectivity. It makes important choices visible enough to be challenged.
That principle is consistent with broader AI evaluation practice. NIST’s Generative AI Profile calls for robust testing, evaluation, validation, and verification processes to be applied iteratively and documented. Right Angle’s method is narrower and market-facing, but the shared discipline is straightforward: define the test, preserve the conditions, and document what changes.
How to Build a Modest Set
You do not need a platform to begin. You need a question source, a written protocol, and the willingness to keep the test even when the answer is disappointing.
- Collect real language. Review sales conversations, intake forms, proposal questions, search data, customer interviews, and emails.
- Map the decision. Include questions from problem recognition through validation rather than clustering everything around provider selection.
- Include questions you may not win. A set built only around your strengths cannot diagnose where your standing is weak.
- Define the conditions. Record the platform, exact wording, date, relevant context, and whether a fresh or continuing session is used.
- Define what you will capture. At minimum: whether the business appears, how it is described, whether it is recommended, which alternatives appear, and which sources are cited.
- Date and freeze Version 1. If a question changes, preserve the old version and explain why the new one is different.
For a self-directed exercise, begin with a set small enough to repeat consistently. Right Angle recommends roughly six to ten questions as a practical starting range, not as a scientific threshold. The strength of the exercise comes from relevance, consistency, and documentation, not from reaching a magic number.
What the Set Measures and What It Does Not
A Frozen Question Set can help you observe whether your business appears, whether the description is accurate, whether credible sources support the answer, which alternatives are introduced, and whether the system moves from mentioning the business to recommending it.
It does not directly measure buyer trust. Trust exists in the human decision. It does not prove that AI caused a sale, that every buyer received the same answer, or that a change in output resulted from your marketing. Those conclusions require additional evidence.
This distinction protects the Recommendation Marketing category. The goal is not to manufacture a score that claims to measure trust. The goal is to build the clarity, credibility, and evidence that make a business more worthy of recommendation and to observe relevant signals honestly enough to learn whether the work may be taking hold.
Personal Orientation and Governed Measurement
A modest personal set is useful for orientation. It can reveal inaccurate descriptions, weak evidence, missing answers, unstable appearances, and unfamiliar alternatives. It helps you decide where to investigate and what to improve.
A governed measurement program must go further. It needs documented selection criteria, controlled versions, defined observation conditions, repeated collection, quality review, and transparent limits on what the results support. The implementation belongs inside the program; the principle should remain visible to the client.
Right Angle is developing its own question set under this discipline. The set will be frozen before its baseline is recorded. When results are ready to publish, the method and limitations should be clear enough that an unfavorable finding could survive alongside a favorable one.
Build Version 1 This Week
Start with one real buyer conversation. Write down the questions that appeared before the buyer was ready to speak with you. Then organize them across the decision, choose a modest set, define what you will record, date it, and stop changing it long enough to learn from it.
If you would like to share your set, send me the questions and how you chose them. A direct message is appropriate if the questions reveal competitive strategy. I am more interested in the reasoning behind the set than in whether Right Angle appears in the first result.
If you want a structured view of how your business appears across the questions that matter, Right Angle combines Answer Engine Optimization with Sightline to examine visibility, representation, citations, and competitive presence.
Coming next: what a responsible baseline can tell you, what it cannot, and why the before picture matters even when it is uncomfortable.
Frequently Asked Questions
What is a Frozen Question Set?
A Frozen Question Set is Right Angle’s term for a version-controlled list of buyer questions and observation conditions selected before AI visibility measurement begins. The active version remains stable during its measurement period, and later changes are dated, explained, and separated from earlier comparisons.
How many questions should the set contain?
There is no universal number. Right Angle recommends six to ten questions as a manageable starting range for a self-directed exercise, not as a scientific threshold. The set should be large enough to span meaningful buyer decisions and small enough to repeat consistently.
Can a Frozen Question Set change?
Yes. Freezing prevents silent changes; it does not require permanence. When a question is added, retired, or rewritten, create a new dated version and avoid presenting results across versions as a direct like-for-like comparison without qualification.
Should the set replace keyword research?
No. Keyword research remains useful for understanding topics, demand, and search language. The question set adds complete natural-language buyer questions and decision context so the measurement reflects how people may use AI-assisted research and evaluation.
Does a Frozen Question Set measure buyer trust?
No. It measures observable answer behavior such as appearance, representation, recommendation, citations, and alternatives under defined conditions. Buyer trust is a human outcome that requires additional evidence.
