Good AI Studio sessions are usually simple: give Marve enough detail, check the output, then ask for specific changes.

Build on each session

When Marve gets something wrong, fix the specific assumption. If the correction should apply next time, save it in Context Manager.

Write a specific prompt
  |
  v
Review the result
  |
  v
Refine the prompt or audience
  |
  v
Add saved context
  |
  v
Use it next time

Prompt with specificity

A specific prompt gives Marve fewer assumptions to make.

Include:

• The marketing objective.
• The target population.
• Behavioral, transactional, or lifecycle signals.
• A timeframe.
• Required exclusions.

Example:

Generate an audience of Enterprise customers who have not logged in for 30 or
more days, opened a support ticket in the last 14 days, and have not expanded
seat count in the last 90 days. Exclude internal accounts and customers with an
active renewal opportunity.

Use Context Manager for standing rules

Put standing context in Context Manager. Keep one-time instructions in the prompt.

Review output carefully

Treat the first answer as a draft until you check the logic.

For audience suggestions, check:

• Segment size.
• Filters and exclusions.
• Date windows.
• Dataset joins.
• Field definitions.

For data exploration, check:

• Query timeframe.
• Cohort definitions.
• Aggregation logic.
• Sample size.
• Caveats Marve included.

Refine with targeted follow-ups

When the answer is close, change one thing at a time.

Keep context current

Update Context Manager when your team changes definitions, adds important datasets, creates new standard exclusions, or discovers a data quality issue. Stale context can be worse than missing context because it can steer Marve toward old assumptions.

You're all set! Return to AI Studio overview whenever you need the full section map.