Before completing a complicated prompt, an AI can be instructed to double check the format with the user.

This serves the users and the AI company. Users can check for misalignments with the AI upfront, and only spend time waiting on a response that is likely to be accurate. The AI on the other hand doesn't waste processor time on responses the user will regenerate anyway.

This pattern is commonly used with the autofill and workflows patterns, as these are complicated and time-expensive actions that its best to get correct upfront. It is also found in inline prompts to avoid overwriting the user's own content until the user has confirmed.

This is not a new UX pattern. Advanced tools like Zapier will test a workflow with sample data from a single record before turning on the workflow for additional uses. What makes it application to AI distinct is the way it keeps the user in charge of the AI, whether the AI is operating as an assistant or an agent.

Details and variations

Before overwriting user data or committing significant user time, double check the AI's understanding of the user's intent with a sample response

  • Show the result for the user to review
  • Consider showing multiple versions for the user to select from
  • Allow the user to revert or cancel the change without impacting their data
  • When the user allows, the AI may overwrite the existing data with new data from the prompt



Keep the user in charge

When working on advanced prompts, AI can spend considerable time and effort completing a task. Giving the user the ability to double check the AI's work keeps them in the driver's seat and only commits their time, energy, and in some cases money when they know the AI understands their intent.

Potential risks

Don't waste user time

There's a fine line between helpful and hinderance. Don't impose a pause in the user momentum only to save processing time for the AI company.


Notion shows 5 sample responses when a prompt is applied to a table column before it generates a response for each row
Only when the user confirms that the sample prompts look good, Notion will update all records with the prompt
Salesforce Copilot creates sample responses in the chat that can be sent to a user interacting with support
When prompting Github Copilot, responses are shown inline and must be confirmed before they replace what the user has already written
Code uses sample responses to ensure it understands the users' intent goes a step further and creates multiple versions of re-writes for the user to choose from, confirming before overwriting
No items found.