Start with the decision, not the keyword
An AI visibility prompt set is a collection of questions you repeatedly ask AI providers to understand how a brand appears in answers. It should represent real buying situations, not every phrase that contains a target keyword. A useful set helps you answer: which customer is asking, what are they trying to decide, what market and language are they in, and what evidence would make the answer useful?
Before writing prompts, define the product category, intended customer, main use cases, alternatives buyers compare, and the facts buyers need before choosing. If these basics are unclear on your own site, refine the positioning before measuring it.
Use four kinds of customer questions
Build a balanced set from different intents. For an example project-management product, a team might track:
- Category discovery: “What project management software works well for a 12-person product team?”
- Buying intent: “Which project management tools include workload planning for under $30 per user?”
- Comparison: “Compare [Product A] with [Product B] for a small remote team.”
- Brand understanding: “What does [Product A] do, and who is it designed for?”
These are example prompts, not a recommended universal list. Replace them with the words your customers use in sales calls, support conversations, search queries, and interviews. Include clear context when location, company size, platform, language, or constraints change the recommendation.
Keep prompts specific enough to interpret
Broad questions such as “best software” produce answers that are hard to connect to an audience or action. Add the decision context that matters: company size, workflow, budget, geography, must-have capability, or a known alternative. Avoid adding so many constraints that the prompt no longer resembles an actual customer question.
Keep a short rationale for each prompt. Record the audience and intent it represents, the market and language, and the page or product fact you expect to support a useful answer. This makes it easier to remove duplicates and understand gaps later.
Separate markets and providers
Language is not a substitute for market. A prompt in Italian may target Italy, Switzerland, or Italian-speaking buyers elsewhere. Configure the country or market explicitly where that distinction matters, then compare like with like. Likewise, review each supported AI provider separately before combining results; response formats, search behavior, and available citations can differ.
LLM Scan currently supports ChatGPT, Claude, Gemini, and Perplexity, subject to workspace plan and configuration. A provider failing to return an answer is a coverage issue, not evidence that a brand was absent. Keep failed requests out of the successful-answer denominator.
Start small, then expand based on evidence
Begin with a manageable set that covers your most important category, buying, comparison, and brand questions. Run a baseline, review the actual answers and sources, and remove prompts that are duplicates or do not represent a meaningful decision. Add prompts when customers, product focus, competitors, or target markets change.
Do not assume that a larger prompt count produces a better strategy. More prompts cost more to run and can make trends harder to interpret. A focused set with a clear audience and purpose is easier to maintain and to turn into useful work.
Turn results into a content decision
For each successful answer, note whether the brand is mentioned, which brands appear, what position the brand has in the answer, and which URLs are cited. If the answer cites a competitor comparison page, inspect what makes that source useful: concrete selection criteria, transparent pricing, implementation details, customer proof, or an answer to a specific objection.
Then make a page-level decision. Improve an existing page if it already serves the same question. Create a new page only when it fills a distinct buyer need and can include first-hand product facts or evidence. Avoid publishing near-duplicate pages just to target slightly different prompt wording.
Review the baseline honestly
Record the date, prompt set, market, provider, and successful-answer coverage alongside the visibility measures. Compare future runs against the same setup where possible, and annotate changes to prompts or product facts. A change in sampled AI answers is a signal to investigate, not proof that one content edit caused a citation or ranking change.
Use the AI visibility tracking guide to understand the metrics and LLM Scan documentation to configure domain tracking. The pricing page lists current prompt, provider, and run limits.