Managing prompts

Prompt Lab is where you manage the questions Maya asks AI assistants and review every answer. This article covers creating and organizing prompts; reading the results is covered in Understanding Prompt Lab metrics.

What makes a good prompt

Write prompts the way a customer would actually ask, in their language:

• Discovery — "What are the best romantic hotels in Cappadocia?"
• Comparison — "Hotel A vs Hotel B, which is better?"
• Brand — "What do reviews say about Hotel A?"
• Problem / support — "How do I cancel my reservation at Hotel A?"
• Buying intent — "I want to book a boutique hotel in Cappadocia, where do I start?"

A balanced set covers the whole funnel. Discovery prompts show whether you enter the conversation at all; comparison prompts show how you stack up once you're in it; brand prompts show how assistants describe you to someone who already knows you.

Adding prompts

One at a time

Use New Prompt, type the question, and set:

• Country — the market the question simulates. Assistants give different answers per market.
• Language — the language of the question and expected answer.
• Intent — how the prompt is categorized in metrics and filters.

Bulk upload

Use Bulk Upload to import a CSV of prompts. Include country and language columns — rows without them fall back to the defaults (US / English), which is usually not what you want for non-US brands. Review the mapping screen before confirming.

Suggestions

The Weekly Suggestions tab proposes new prompts based on your existing answers, the competitors that appear in them, and (if connected) your Search Console queries. Accept the ones that match how your customers actually talk.

How runs work

• Each active prompt runs on a daily schedule, staggered through the day. The list shows the last run and the next scheduled one.
• Every run queries each enabled AI provider and stores the full answer, your mention, your position, competitors, citations, and sentiment.
• Historical runs are kept, so trend lines reflect real history rather than snapshots.

Organizing

• Labels — tag prompts (e.g. by product line or campaign) and filter the list by label.
• Filters — narrow by country, language, intent, or free-text search.
• Export — download the current view as a spreadsheet for reporting.

Editing and pausing

Editing a prompt's text creates a fresh measurement series — position history from the old wording stays with the old runs. Prefer adding a new prompt over rewording one that has useful history. Pausing a prompt stops future runs but keeps its history visible.