Choosing and organizing prompts
How to build a prompt set that reflects real buyer questions, and how topics and intent labels shape your reporting.
Updated
A prompt is a question Signal asks every enabled answer engine on each run. Your metrics are only as meaningful as the prompts behind them, so it is worth spending time on this step.
Write prompts the way customers ask
Use the language customers use in sales calls, support tickets, and search queries. Prefer complete questions ("What is the best accounting software for a 20-person nonprofit?") over keyword fragments ("nonprofit accounting software"). Answer engines respond to intent and context; specific prompts produce answers that are easier to act on.
Cover four intents
| Intent | Example | What it tells you |
|---|---|---|
| Discovery | Best project management tools for agencies | Whether you are on the shortlist at all |
| Comparison | Acme vs Contoso for client reporting | How you are positioned against a named rival |
| Evaluation | Is Acme worth it for small teams? | How engines describe your strengths and drawbacks |
| Brand | What does Acme do? | Whether engines describe you accurately |
Brand prompts almost always mention you, so they inflate mention rate. Signal reports metrics by intent so you can separate "are we discovered" from "are we described correctly."
Group prompts by topic
Topics are free-text labels such as "Pricing", "Integrations", or "Nonprofits". Use them to group prompts that map to a product line, audience, or content cluster. Topic filters apply to every chart and table.
Keep the set stable
Changing prompts changes the denominator. When you add or remove prompts, trend lines reflect a different question set. Prefer pausing a prompt over deleting it; paused prompts keep their history but are skipped on new runs.
How many prompts do you need?
Each prompt is asked once per engine per run. With 25 prompts and 4 engines, one run produces 100 answers, which supports a mention-rate estimate with a 95% interval of roughly ±10 percentage points. Precision improves as answers accumulate across runs. See Metrics, denominators, and confidence intervals.