Choose your next AI practice from your run notes
Knowing the feature names tells you little about how you check the work. Your last task's notes give you a better starting point.
What you'll make
one next practice exercise linked to an observed gap in your own task notes.
What you'll need
ChatGPT and a short record of a task you tried: its brief, result, check and correction. Use your own notes or the fictional example; allow 15 minutes. Keep other people’s performance and private information out of the exercise.
Example result.
| Observed gap | Practice check |
|---|---|
| Date status changed | Compare proposed and agreed |
| Source check unrecorded | Quote the supporting sentence |
| Repair unchecked | Recheck the changed sentence |
1. Find a practice in your notes
Open ChatGPT on the web, choose Chat › New chat. Paste a brief note with four parts: what you asked for, what came back, what you checked and what you changed. “Looked fine” is an observation about appearance; it isn’t a recorded source comparison.
Read [my run notes]. Separate behaviours evidenced in the notes from behaviours that weren't recorded. Choose one useful next AI practice, give its input and output, and specify the source check that will show whether I did it. Don't rate my competence, infer my intentions or judge anyone else. Missing evidence means not recorded, not failed.[my run notes] means the four-part record from step 1.
Why this works: a specific task exposes a specific skill to practise. A broad confidence score usually hides it.
What you should see: one practice exercise tied to your notes, with missing records kept distinct from observed mistakes.
2. Check the suggested gap
Find the passage behind the suggested exercise. If the answer says a check failed, your notes should record the error. If they contain no record of that check, the answer should say not recorded.
Why this works: the request makes the difference between missing notes and a failed check explicit.
What you should see: one exercise you can try and one check you can describe before starting.
3. Reduce it to one short session
Turn the suggested exercise into a 15-minute session using fictional material. Include preparation, one AI request, source comparison and rechecking a correction. State what result I should record. If that won't fit, shrink the input; don't omit the check.Why this works: the session practises the whole skill, including checking the repaired result.
What you should see: a small task with a visible finish, rather than an open-ended training programme.
Check the result
- Is the exercise linked to a passage in your notes?
- Are missing records described as missing records?
- Does the session include a second check after repair?
The usual failure: the assistant labels you “poor at reviewing” because your note didn’t describe a source comparison.
The source comparison was not recorded. That does not establish that I failed to do it. Remove the personal rating. Keep the next practice as a way to make the check visible, and distinguish observed errors from missing notes throughout.Worked example
Fictional example.
Bookshop event update: asked AI to summarise the organiser's note.
Draft said the reading would happen on Saturday. Source called Saturday proposed.
Corrected the word to proposed. Did not record a final source comparison.A weak response says “You are a beginner reviewer.” The repair gives a useful exercise instead: summarise a new fictional note containing one proposed and one confirmed event, compare both status words with their source sentences, then recheck any correction.
The wrong status is evidenced. The missing final check is a gap in the record. Neither supplies a general judgement about the person running the task.
Go deeper
Keep evidence of the changed behaviour
Repeat the exercise on another short input and record the decisive comparison. If you now record the final source check, that is an observable improvement in this practice. It isn’t evidence for a productivity claim or an employment decision.