How to choose prompts for an AI visibility report
Choose AI visibility prompts from the questions buyers ask before selecting a product. A useful set covers category discovery, product comparisons, and practical buying constraints. Its scope determines what the report can tell you, so write down the business decision before writing the questions.
Start with the decision you need to make
“Do models recommend our product for small kitchens?” is a workable reporting question. “How visible are we in AI?” needs more scope: visible to which buyer, asking about which need?
Gather wording from sales questions, customer research, support tickets, and site-search data you are authorized to use. Search Console can help identify search phrasing. None of these sources gives a complete census of questions asked inside AI products, but each can ground your sample in an actual customer concern.
Remove personal details and confidential customer information. Rewrite the concern as a standalone buyer question that a model can answer without access to the original conversation.
Cover distinct needs without filling a quota
For a reusable bottle brand, a starter set might include the following. These are example prompts, not measured high-volume queries.
| Purpose | Example question |
|---|---|
| Category discovery | Which reusable bottles work well for commuting? |
| Product requirement | Which reusable bottles fit in a small daypack? |
| Care | Which reusable bottles are easiest to clean? |
| Price constraint | Which reusable bottles cost less than $40? |
| Comparison | How do Brand A and Brand B compare for commuting? |
| Purchase | Where can I buy replacement lids for Brand A? |
Keep each question focused enough that you can recognize a useful answer. Change the use case or requirement when that difference matters to the business. Rephrasing the same question many times can overweight one concern in the aggregate.
Braign supports up to 100 questions and 10 selected models per report. Those are scope limits, not targets. Start with a set you can review closely and price it in the calculator before checkout.
Separate branded questions from unbranded discovery
“Which reusable bottle should I buy?” asks the model to choose brands. “Is Brand A a good reusable bottle?” supplies the name and asks for an assessment. Both can be useful, but they measure different situations.
Label those groups and inspect their answers separately. A question containing your brand can make a mention much more likely. Combining many branded questions with a few unbranded ones can make the aggregate presence rate hard to interpret as category discovery.
Avoid building every prompt around a preferred claim, such as “Why is Brand A the best option?” That wording tells the model what conclusion you want. A question such as “What are the limitations of Brand A for commuting?” gives you a clearer task for reviewing answer treatment.
Record the model and search conditions
Run the same saved questions across the selected models. Record model IDs and the search settings used for the run. Models can differ in capabilities and outputs, so read their results separately before relying on a combined percentage.
Braign calls the selected models through an API. The results should be described that way. A consumer chat application can add its own retrieval, system instructions, account context, or personalization; sharing a model family name does not make the two experiences identical.
Adding a country or budget to a question makes that context part of the prompt. It does not establish that you sampled users located in that country. Keep those distinctions in the report handoff.
Keep a stable set for comparisons over time
After the first run, inspect ambiguous questions and check whether their answers address the intended purchase decision. If a question needs changing, document the change and treat the revised wording as a new baseline.
Braign's report settings can organize questions into topics and identify owned websites. Those settings do not rewrite the questions sent to models or change the purchased scope. Saved topics help you read the results; they are not instructions to the answering model.
For recurring reviews, keep the core question set stable. Investigate new needs in a separately documented scope so you can still explain changes in the original set. The guide to comparing runs covers matching and incomplete results.
See the evidence in a report
Explore the public sample, or choose your brand, buyer questions, and models to price your own report.