Choose situations, not keywords
Good AI visibility prompts sound like the questions customers ask when they are choosing, comparing, or trying to solve a specific problem. Start with those situations. Write one clear question for each, keep the wording natural, and only include details that would change a useful answer.
A prompt is not a keyword with a question mark added. “Bristol startup accountant” may help you label a topic, but “Which accountants in Bristol work with early-stage software companies?” gives an assistant a recognisable task, audience, and location.
There is no single perfect wording. The aim is a small, reviewed set that covers the real reasons someone might look for a business like yours. Keep that set stable when you compare reports, because changing the questions also changes what you measured.
What is the best accountant?
Which accountants in Bristol help early-stage software companies with monthly accounts?
Use five question types to find useful prompts
These five question types stop a set from becoming ten versions of the same search. They are prompts for your review, not a quota. Leave out a type when it does not fit how customers choose your business.
Which accountancy firms in Bristol work with early-stage software companies?
Who can help a UK software startup prepare monthly management accounts?
Which startup accountants work with businesses in Bristol and the South West?
What should a growing startup compare when choosing an online accountant versus a local firm?
What should a first-time founder look for in an accountant for a software business?
Illustrative examples: these questions describe a fictional accountancy business. They are examples of structure, not a recommended question set for every firm.
Build a focused question set
- Collect the situations customers bring you.
Look at enquiry emails, sales notes, support conversations, and the language on your important service pages. Write down the problem, audience, location, and decision behind each enquiry.
- Turn each situation into a complete question.
Use the words a customer would recognise. Keep technical terms only when the customer would use them or the distinction changes the answer.
- Change one meaningful detail at a time.
If two questions change the audience, location, and service together, you will not know which difference explains the answers you see.
- Remove duplicates and weak variations.
“Best accountant for startups” and “top startup accountant” ask essentially the same thing. Keep the version that sounds most like a real customer and use the spare space for another buying situation.
- Keep branded questions separate.
“What does Northstar Accounting offer?” checks whether an assistant can describe a known business. It does not test whether that business appears when someone is still discovering options.
Start small enough that you can read every answer. A balanced set you understand is more useful than a long list nobody reviews. Mentsight is designed to draft questions from your site and let you edit, remove, or add to them before any report runs.
Review every question before you measure
Read the set aloud as if a customer had sent each question to you. Then use this checklist before keeping it:
It has a clear job. You know whether the person is discovering, comparing, or solving a particular problem.
It sounds natural. It reads like a complete question, not a string of search terms.
Its details matter. The service, audience, or location would genuinely change which answer is useful.
It is distinct. Another question in the set does not already cover the same decision.
You would act on the answer. Whether you appear or not, reading the response and its sources could inform a sensible next check.
Once you approve the set, save the exact wording with each report. Our measurement method explains why Mentsight keeps the questions, full answers, assistant, date, and returned citations together.
What the questions cannot prove
A well-designed question set makes a report easier to interpret. It does not turn the report into a permanent market ranking. AI answers can change between assistants and between runs of the same question.
The wording can also shape the result. A location, audience, or service detail may be necessary, but it narrows what you are observing. Record the exact prompt and resist generalising one answer beyond that scope.
Finally, appearing in an answer does not establish why you appeared. Read the full response, inspect any sources it returned, and treat a pattern as a reason to investigate rather than proof of cause.