Prove It · Module 4: Stakes and trail
Low, medium, high
Every task with AI output in it sits in one of three bands, and you can usually tell which in a few seconds.
Low: nobody else relies on it, or it will be reviewed before use, or it is easily reversible. A first draft you will rewrite. A brainstorm. A summary for your own notes. One check: skim the specifics for anything that would embarrass you.
Medium: others will read it and may act on it, but a wrong detail is recoverable. An internal briefing. A team update. A draft for a colleague to review. Checks one, two and five: specifics, sources, second opinion. Five minutes.
High: it goes to a client, a regulator, a board, the public, or into a decision that is hard to undo. Anything with a figure that will be quoted. Anything in a regulated field. All five checks, sources-first where possible, and a trail. As long as it takes.
| Band | Checking |
|---|---|
| Low | Skim the specifics. Fix anything embarrassing. |
| Medium | Checks 1, 2 and 5. About five minutes. |
| High | All five checks. Work from sources where possible. Keep a trail. |
In practice. A communications lead sorted her week: social copy (low), internal newsletter (medium), a statement to a journalist (high). She spent thirty seconds, five minutes and forty minutes checking respectively. The forty minutes found a wrong date that would have been in print.
At your desk. List the last five things you used AI for. Band each one. Write down how much checking you actually did. Note where the effort did not match the band.
Write it down.
- My five tasks and their bands:
- Where I under-checked:
- Where I over-checked: