A Responsible AI Checklist for Everyday Work

Start with the result, not the tool

Pick one task where you can judge whether the result is useful. Drafting an outline from public notes is easier to review than asking a tool to run your whole business. Write a short brief: what needs doing, who it is for, what the output should contain and what a good result looks like.

Set a stopping point. Ask for a draft or recommendation before allowing publication, payments or changes to an important system. A confident response is not evidence that the task was done correctly.

Check what information you are sharing

Use a small sample of public or invented information when testing. Remove names, contact details and confidential material that the task does not need. Check your organization’s rules and the service’s current data controls before using work information.

Permissions should match the task. A tool that only needs to read a document should not automatically receive permission to delete files or send messages. Review connected accounts and revoke access you no longer need.

Run this checklist before you rely on the result

• Purpose: can I explain who benefits and what useful result I expect?
• Input: have I removed private or irrelevant information?
• Scope: does the tool know what it may do and when it must stop?
• Evidence: can I check important claims against the original material?
• Accuracy: have I checked names, dates, calculations, quotations and links?
• Perspective: have I looked for missing context or unfair assumptions?
• Authority: will a responsible person review consequential changes before they happen?
• Learning: have I recorded what worked, what failed and what should change next time?

Example: turning public notes into an event invitation

Give the tool the event’s confirmed title, audience, date, venue, price and booking link. Ask for a short draft using only those details. If something is missing, ask it to mark the gap instead of inventing an answer.

Compare the draft with the source notes. Open the booking link, confirm the time zone and check that the price includes what you intend. A person responsible for the event should approve the final invitation. The AI helps with wording; the approval still belongs to you.

Use a review loop instead of expanding too quickly

After the task, note how much correction it needed and whether it helped the intended audience. Repeat the same task with one improvement before adding more tools or giving broader access.

The ADD VALUE Framework provides a structure for that review: notice the situation, define the outcome, devise a plan, validate assumptions, act, learn, understand and evolve. You can use the free worksheet to plan the next test.