Review Is Part of the Workflow
Unit ID: PAW-M08-U00 Estimated active time: 15-20 minutes
Review is not what happens if you have time
It is a step in the workflow with a name, an owner and a defined output โ the same as generation.
Treating it as optional is what produces the failures this module is about, and the reason it gets skipped is almost always that it was never scheduled.
Design the review when you design the prompt
The useful move is to decide, before generating anything, what you will check and how long it will take.
If the check would take longer than doing the task yourself, that is a finding: this task is not a good fit for AI assistance, and you have learned it before spending the time.
Review effort should match consequence
REVIEW LEVEL
Consequence if wrong Review
-------------------------- ----------------------------------------
trivial, easily corrected skim; check nothing was invented
mild, internal only check every figure and claim against source
affects a decision full rubric score + trace every claim
affects a person full review + a second human before it is sent
external or on the record full review + named approver + record kept
The right-hand column is the workflow. Most teams have only the top two rows in practice, and use them for everything.
Who reviews matters
The person who wrote the prompt is the worst reviewer of its output, for the same reason authors miss their own typos: they read what they meant.
For anything above the middle row, the reviewer should be someone who did not write the prompt.
Record that review happened
A one-line record โ who reviewed, when, what they checked โ is enough.
It is what lets you answer "was this checked?" months later, and it is what makes the review a real step rather than an intention.
Mini practice
Take three tasks you use AI for.
Place each on the review-level table.
For any that sit above the middle row without a named reviewer, decide now who that is.
