Every AI visibility measurement rests on one decision: which questions you ask. Choose questions nobody asks, and your numbers describe an imaginary market. Change them every month, and you cannot compare one period to the next.
This article sets out a method to build a question set that reflects real demand, stays stable enough to measure, and leads to specific actions.
Why the question set matters more than the tool
AI assistants answer the question they are given. Small changes in wording, context or constraints can change which brands appear. That has two consequences:
- Your question set defines your market. Share of voice and mention rate only mean something relative to the questions you track.
- Stability is part of the method. Answers already vary between runs. If the questions also change, you can no longer tell market movement from measurement change.
So the goal is a set that is representative, organized and stable.
Step 1: Start from real demand
Good questions come from evidence of what people actually ask, not from what you would like to rank for. Useful sources you already have:
- Search Console queries. Long, question-shaped queries in the Performance report are a direct view of how people phrase needs. Google notes that traffic from its AI features is included in the Performance report under the Web search type.
- Sales and support conversations. The questions prospects ask before buying, and the objections they raise.
- On-site search. What visitors type into your own search box.
- Reviews and public discussions. How customers describe the problem in their own words, including in forums and community threads.
- Competitor positioning. The use cases and audiences competitors claim, which buyers will often use as comparison criteria.
Write questions the way a person would ask an assistant: full sentences, with context. “Best accounting software” is a keyword. “What accounting software works best for a freelance designer in the UK who needs to handle VAT?” is a question.
Step 2: Cover the main intents
Group questions by what the person is trying to do. A balanced set usually includes:
| Intent | Example question | What it reveals |
|---|---|---|
| Category discovery | “What are the best tools to schedule social media posts for a small agency?” | Whether you are in the default candidate list |
| Constrained choice | “Which social scheduling tool is affordable for a team of 3 and supports LinkedIn?” | Visibility on specific needs and niches |
| Comparison | “Tool A vs Tool B for agencies: which is better?” | How you are positioned against named rivals |
| Problem solving | “How can I approve client posts before they are published?” | Whether your features are linked to the problem |
| Validation | “Is Tool A reliable for agencies?” | How assistants describe your reputation |
| Local or market | “Best social scheduling tool for agencies in Germany” | Market-specific visibility |
Keep branded questions (those that name your brand) in their own group. They tell you how you are described, but they inflate visibility if mixed into the unbranded set.
Step 3: Organize by topic
Inside each intent, tag each question with a topic: a product line, a use case, an audience or a feature. Topic groups are what make results actionable. “Low visibility overall” is hard to act on. “Low visibility on approval workflow questions” points to a page, a message and a set of sources.
A simple tagging scheme:
- Topic: the product area or use case
- Intent: one of the groups above
- Market: country and language
- Branded: yes or no
- Priority: how close the question is to revenue
Step 4: Localize properly
If you work in several countries or languages, do not just translate your questions. People in different markets phrase needs differently, face different regulations, and consider different competitors.
For each market:
- write questions in the local language, as a local buyer would ask them,
- adjust constraints (currency, regulation, local platforms),
- review the competitor list, since local players often matter more than global ones.
Step 5: Keep the set stable, and change it deliberately
A question set is a measurement instrument. Treat changes the way you would treat changes to an analytics setup.
- Keep a stable core. Most questions should stay unchanged across periods.
- Never silently reword. A reworded question is a new question. Retire the old one and add the new one, with dates.
- Add new questions in labeled batches. That way you know when a change in totals comes from new questions rather than market movement.
- Review on a schedule. For example each quarter, check whether the set still reflects demand and your offer.
Step 6: Turn answers into actions
Measurement is only useful if it leads to work. For each topic group, read the answers and ask four questions.
1. Are we mentioned? If not, look at which brands are named and which sources are cited.
2. Are we described accurately? Wrong pricing, outdated features or the wrong audience point to inconsistent information on your site or on third-party pages.
3. Which sources are cited? These are the pages that shape the answer: your own pages, comparison articles, review platforms, directories, forums.
4. What would change the answer? Usually one of a short list of actions:
- create or improve a page that directly answers the question,
- clarify positioning on existing pages (who it is for, what it does, where),
- earn presence on the third-party sources that are cited,
- correct inaccurate public information or respond to reviews,
- fix technical issues that prevent pages from being crawled or indexed.
An example
A language school tracks questions in France and Spain. In Spain, it rarely appears on “online Spanish courses for business” questions, while it is visible on general ones. The cited sources for that group are two comparison guides and a public forum thread, and none mentions the school’s business courses. Its own business course page is thin and mostly made of images.
Actions: rewrite the business course page with clear, readable facts in Spanish, contact the authors of the two guides, and join the relevant discussions openly as the school. The group is then measured over the following periods to confirm whether the change holds.
Checklist
- Questions sourced from real demand (Search Console, sales, support, reviews)
- Written as natural questions with context
- All main intents covered, branded questions kept separate
- Every question tagged by topic, intent, market and priority
- Questions written natively for each market
- A stable core, with dated additions and retirements
- Each weak topic group linked to at least one concrete action
How SearchFirefly helps
SearchFirefly asks your tracked questions in each market’s language across ChatGPT, Gemini, Perplexity, Google AI Overviews and Google AI Mode, and shows for every question the answer, the brands named and the sources cited. Alongside your Search Console performance and tracked keywords, that turns your question set into prioritized recommendations, each with the exact fix and a check after release.