# The AI task audit: assist, automate or leave it alone?

Original 5HO practice material. Fictional example. Reviewed 5 October 2026.

## Useful result

You'll finish with a decision record for assistance, automation or no AI use.

Some work benefits from a draft. Some work follows a fixed rule. Some work needs a person to make the call. Sorting those first prevents a very efficient solution to the wrong problem.

## Set up

Allow 25 minutes. List five recurring tasks or use the sample. Keep confidential task details out of an unapproved account.

## Practice input

Tasks: summarise internal notes; copy approved reference numbers into a table; decide a refund exception; draft a public event reminder; remove duplicate files from a shared drive.

## Do the task

1. For each task identify whether the rules are fixed and how a wrong result would be found.
2. Classify it as human-led, AI-assisted or a candidate for tested automation.
3. Choose one low-consequence task for a manual trial before connecting tools.

## Prompt

Classify these tasks and explain the reason for each. Include required review, access needed and a first safe test. Treat irreversible file removal and refund exceptions separately from drafting.

## Refine

Offer a non-AI alternative for the reference-number task. Do not assume every repetitive activity needs a language model.

## Expected checks

A simple rule-based transfer may suit reference numbers. Refund exceptions remain human-led. Duplicate-file detection can propose a list without deleting anything.

## Save

Save your task map and the one experiment you will actually run. Leave the other tasks as decisions, not an automation backlog.

## Sources

- Writing clear prompts: https://help.openai.com/en/articles/10032626-prompt-engineering-best-practices-for-chatgpt