What is a prompt?
A prompt is the text you give an AI model to get a response: a question, an instruction or a task. "What is the best accounting software for freelancers?" is a prompt, and so is "Summarize this contract in five bullet points."
In AI visibility, the word has a narrower meaning. Here a prompt is the unit of measurement: a fixed question that a prompt tracking tool sends to AI models on a schedule, recording which brands, sources and sentiments appear in each answer. The quality of your data depends directly on how well those prompts reflect what real buyers ask.
Tracked prompts fall into two main groups. Category prompts are unbranded questions a potential customer asks before they know you, such as "best payroll software for small businesses". Branded prompts include your name, such as "Is [brand] reliable?" or "[brand] vs [competitor]".
Why prompts matter for AI visibility
Category prompts measure discovery: whether AI models put you on the shortlist when buyers do not have a brand in mind yet. Branded prompts measure reputation: what models say when someone asks about you directly, and whether those facts are correct. You need both, but they answer different questions.
A prompt set that leans too heavily on branded questions flatters you. A model almost always mentions a brand that is named in the question, so your visibility and AI share of voice look far better than they are. If a tool reports close to 100 percent share of voice, your prompt set is probably too branded.
How to build a good prompt set
Collect real questions from sales calls, support tickets, site search and Search Console queries.
Rewrite them the way a customer who has never heard of you would ask an AI assistant, in full sentences.
Mix types: category, comparison and problem-based prompts, plus a smaller share of branded ones.
Add the details buyers use, such as audience, budget, location or language.
Keep the core set stable, so trends reflect changes in the answers and not in your questions.
Because of AI answer volatility, one check proves little. Prompt tracking in Trackbase runs your prompt set every 24 hours across 11 AI models and turns the answers into visibility, sentiment and share of voice data.
Examples of prompts
Category prompt for a Dutch bicycle webshop: "Which e-bike is best for a 20-kilometer daily commute under 2,500 euros?"
Comparison prompt for a B2B payroll software vendor: "Compare payroll tools for companies with 50 employees in the Netherlands."
Problem prompt for a project management tool: "How can a small agency stop missing client deadlines?"
Branded prompt for a meal kit service: "Is [brand] worth it for a family of four?"
Frequently asked questions
What is the difference between a category prompt and a branded prompt?
A category prompt is an unbranded question about a need or product type, such as "best CRM for small agencies". A branded prompt names a specific company, such as "Is [brand] good for agencies?" Category prompts show whether AI models recommend you to new buyers. Branded prompts show how models describe you when someone asks about you directly.
How many prompts should I track?
Enough to cover your main products, audiences and buying questions, but few enough to review properly. A practical start is a focused set of category and comparison prompts that you expand once you see where the gaps are. A stable set matters more than a large one, because you need consistent prompts to compare results over time.
Why do I get different answers to the same prompt?
AI models generate each answer anew, so the wording, the recommended brands and the sources can change between runs, even minutes apart. Answers also differ per model, per country and sometimes per user, because logged-in accounts can be personalized. That is why AI visibility tools run the same prompts repeatedly and report trends instead of relying on a single answer.