Build a prompt set
A prompt set is the list of questions AppearIn runs against ChatGPT, Gemini, Perplexity, on a schedule. Each run is a dated snapshot. The set defines what you can see: your visibility is only measured for the questions you ask. Aim for five to ten prompts that sound like real customer questions, grouped by intent, and you have a foundation every later metric rests on.
Before you start
Build your prompt set inside a project, after you've confirmed the brand profile and competitors. The set you write here is what Share of Voice and your visibility metrics are calculated over.
Think in customer intents, not features
A prompt is a question a real person might type into an assistant. The most useful sets cover the different reasons someone would ask about your category. Three intents matter most:
| Intent | What the customer wants | Example prompt |
|---|---|---|
| Informational | To understand the category or solve a problem | "how do startups keep track of projects?" |
| Commercial | To compare options before buying | "best project management tools for startups" |
| Navigational | To learn about a specific brand | "is Acme good for small teams?" |
Commercial prompts are where most visibility battles are won or lost, since that's where the engine recommends specific tools. But a set built only of "best X" questions misses the informational searches that introduce your brand earlier in the journey. Spread your prompts across the intents your buyers actually move through.
Write five to ten real questions
List how customers actually ask
Write each prompt the way a buyer would phrase it to an assistant, not the way you describe yourself. "Best project management tools for startups" is a real question. "Acme is the leading AI-powered workspace" is your tagline, and an engine will never be asked it.
Cover the intents that matter
Make sure your set includes at least one informational prompt, a few commercial ones, and a navigational one. Five to ten total is enough to start. You can grow the set later without losing the history on existing prompts.
Accept AppearIn's suggestions
AppearIn suggests prompts from your brand profile, your category, features, and competitors. Accept the ones that sound like questions your customers ask, edit the ones that are close, and skip the rest. Suggestions are a fast start, not a finished set.
Group by intent or segment
Group prompts so you can read results by slice, for example commercial prompts separately from informational, or "startup" prompts separately from "enterprise". Grouping makes it obvious where you're strong and where a competitor owns the answer.
Run and schedule
Use the onboarding seed run for your first snapshot, then keep recurring runs on so the same prompts refresh together. Because the data is probabilistic, a single run tells you little. The trend across many runs is what you read.
Worked example
Suppose Acme makes project management software for startups. A solid starting set of seven prompts might be:
- "how do small startups keep projects organized?" (informational)
- "best project management tools for startups" (commercial)
- "affordable project management software for a small team" (commercial)
- "Beta vs Acme for a 10-person startup" (commercial)
- "project management tools with a free plan" (commercial)
- "is Acme good for early-stage teams?" (navigational)
- "what should I look for in a startup PM tool?" (informational)
Run this against the connected engines and you might find Acme is mentioned in the "best project management tools for startups" answer on ChatGPT and Perplexity, but Beta takes the top slot on Gemini. That difference is exactly what the set is for: it tells you which engine, and which intent, to work on.
What good looks like
A healthy prompt set has:
- Five to ten prompts, each phrased as a question a customer would actually ask.
- A mix of informational, commercial, and navigational intents.
- No near-duplicates, each prompt covers distinct ground.
- Grouping by intent or segment, so results are readable at a glance.
- Weekly runs turned on, turning single answers into a trend.
Common mistakes
Describing yourself instead of asking like a customer.An engine is asked questions, not pitched taglines. "Best X for Y" beats "Acme is the leading" every time.
Too few prompts.One or two questions give you a sliver of coverage. If you don't ask the question, you can't see whether you appear in its answer.
Near-duplicates."best PM tools for startups" and "top project management tools for startups" measure almost the same thing. Spend that slot on a different intent or segment instead.