Monitoring dashboards include a prompt manager. Open it with Manage prompts on the prompts view.

Adding a prompt

  1. Click Add prompt.
  2. Write the buyer question exactly as a customer would ask it — a full natural-language question, not a keyword. Good: "Which dental clinic in Munich is best for anxious patients?" Bad: "dental clinic munich".
  3. Assign a tag to group it with related questions (or create a new group).
  4. Save. The prompt is enabled by default and will be asked — on all 3 channels — starting with the next scan.

Editing and disabling

  • Edit rewrites the question or changes its tag. Answer history for the prompt is kept, but remember: a reworded question is effectively a new measurement — expect a step change in its results.
  • Disable (Off) keeps the prompt and its history but stops asking it. Re-enable any time.
  • Delete removes a custom prompt entirely. Prefer disabling if there is any chance you'll want the history.

A minimum number of prompts must stay enabled so scans remain statistically meaningful — the manager shows the limit and the count of enabled prompts, and notes that the next scan asks exactly your enabled set.

Demo dashboards

The public demo dashboard is read-only; prompt changes are disabled there. You need your own monitoring dashboard to manage prompts.

Practical guidance

  • Mirror real buyers. The best custom prompts come from sales calls and support tickets — questions actual customers asked before buying.
  • Keep a stable core. Leave most of the default 29 untouched so your long-term trend stays comparable; add custom prompts around the edges.
  • Prune, don't hoard. Every enabled prompt costs scan samples. Disable questions that stopped being relevant instead of letting the set grow stale.

Country (geographic simulation)

Each prompt can target a country (default: Global). When a country is set, scans instruct the AI to answer as a user located in that country, in the locale typical for that market, and web grounding is scoped to that country. The same question can be tracked in several countries — each combination is a separate prompt with its own results, and the dashboard lets you filter by country.

Honest scope: this is prompt-level geographic simulation. We steer the model with instructions and country-scoped search; we do not route requests through in-country proxies or change the model's actual IP location, and results can differ from what a real user physically in that country would see.

See also Prompt suggestions and CSV export.