People at work
Who has time to check the AI draft?
New guidance from the American Psychological Association calls for meaningful review. Give employees the source, time to compare it and authority to reject a draft.

If AI produces a customer message and a busy employee must approve it, the review needs a place in the workload. The manager must give that employee the source material, enough time to compare it and authority to stop a wrong message.
The American Psychological Association’s guidance dated 16 November links meaningful human oversight with job design. It draws on established research about attention, responsibility and working with automation, alongside newer evidence about AI. Its findings about AI-specific verification work remain emerging. This is guidance for decisions, not a compliance standard or proof that any product improves work. In the accompanying release, APA executive lead psychologist Dennis Stolle calls for “making sure human oversight is meaningful”.
Make rejection possible before setting the deadline
For a Caribbean retailer, a wrong delivery promise can leave a staff member handling a complaint they did not create. Decide who can delay sending a message, where they find the current policy, and who resolves a disagreement. A manager who requires approval should also decide which other work can wait.
Compare a message with a short source
Here is a hypothetical review exercise. Source note: “Orders confirmed by noon leave on the next working day. Delivery date depends on the courier. Customers may cancel before dispatch.” Draft A: “Your order will arrive tomorrow. You cannot cancel.” Draft B: “If confirmed by noon, your order leaves on the next working day. The courier determines the delivery date. You may cancel before dispatch.”
The reviewer should reject Draft A: it invents an arrival guarantee and reverses the cancellation rule. Draft B preserves the supplied conditions, though the employee must still confirm that this policy applies to the customer’s order. The expected output is a selected draft with a reason tied to the source. No live customer details are needed.
Try the exercise with the people who will review messages. Ask them to point to the words that support each promise. A correct answer in this small exercise does not establish competence across all customer cases; it shows whether these instructions and this source are usable.
Check whether management has made room for review
The Executive AI Readiness Score gives a business leader a structured self-assessment of their preparation for AI. Bring the AI-use policy, the name of the accountable manager and the current review arrangements. Start the questionnaire, choose answers that those records support, and inspect the result and section recommendations.
The result includes a score, a leadership category and section recommendations. Its questions include whether the leader understands the tools and has established accountability. These are rule-based interpretations of your answers, not a qualification, a comparison with other executives or a prediction of business results.
Use one weak area to start a conversation with staff. For the message-review problem, write down who can reject a draft and reserve time for comparing it with the policy. Then ask the reviewer to complete the exercise using only the materials supplied. Missing sources or an unresolved approval dispute are reasons to hold the message.
Sources and further reading
- American Psychological Association, 16 November 2026: Deploying AI into workplaces
- APA, 16 November 2026: Practical guidance on AI at work
- Executive AI Readiness Score
Sources checked 19 November 2026.