Your Reusable Discernment Checklist
Combine the course into one practical checklist covering product, process, and performance, then tailor it to the kinds of AI-assisted work you actually do.
Why a Checklist
Pilots and surgeons use checklists not because they lack skill but because skilled people skip steps when they’re busy, confident, or interrupted. Reviewing AI output has the same problem: by the tenth draft of the day, “it looks fine” starts to feel like a check.
A good discernment checklist is short, scaled to stakes, and tailored to the work you do. Here’s a general version to start from.
The Discernment Checklist
0. Set the level (Module 1)
- What are the stakes, checkability, and reach? Which rung of the review ladder does this need?
1. Product: facts and numbers (Module 2)
- Have I extracted the load-bearing claims: names, dates, figures, quotes, citations?
- Is each one confirmed against an independent source, not the AI?
- Do citations exist and say what’s claimed?
- Do the numbers pass a sanity check for magnitude, units, and totals?
2. Product: fairness and completeness (Module 2)
- Did it cover everything I asked for?
- Who is affected but missing? Whose context is treated as the default?
- Would the swap test change anything important?
3. Process: reasoning (Module 3)
- Do I know what it assumed about my situation, and are those assumptions true?
- Have I seen the strongest counter-argument?
- Can I explain the reasoning in my own words?
4. Process: analysis and technical output (Module 3)
- Did it answer my actual question, with my definitions?
- Were all the inputs used correctly?
- Have I recomputed or tested at least one result myself?
5. Performance: the collaboration
- Was the AI just agreeing with me? Did I ask it for weaknesses?
- Did it flag gaps and uncertainty, or fill them silently?
6. Ready to ship (Module 4)
- Is everything still unverified either removed or clearly flagged?
- Does my reviewer know where AI was involved?
- Would I put my name on this?
You won’t run every line every time. At the glance level you might use only items 0 and 6. At the verify level you run all of it.
Tailor It to Your Work
The general checklist is a starting point. Your version should reflect the errors that actually show up in your work. Use the loop:
Examples of tailored items:
- A grant writer: “Every funder requirement in the call matches the current call document, not last year’s.”
- A teacher: “Reading level checked; examples reflect my actual students.”
- An HR coordinator: “Policy references match our current handbook, not generic employment law.”
- A small-business owner: “Prices, hours, and product names checked against our live website.”
Common mistake: building a 40-item checklist that covers everything, then never using it.
If a checklist takes longer than the task, it gets skipped. Keep your everyday version to about one screen, with a short “glance” version for low-stakes work. Put the items that catch your most common errors at the top.
- Copy the general checklist into a document.
- Think of the two kinds of AI-assisted work you do most often. For each, add at least two tailored items based on errors you’ve seen or the risks you identified in this course.
- Cut or merge items until the everyday version fits on one screen. Mark three items as your glance version.
- Test it. Ask an AI assistant to produce a realistic piece of your typical work, such as an email to a client, a lesson summary, or a project update, using real-sounding but non-sensitive details. Run your checklist on it.
- Note which items found something and which didn’t. Adjust one item based on what you learned.
1. Why do experienced people still benefit from a discernment checklist?
Checklists protect against predictable lapses, not lack of skill. They keep the important steps from being skipped.
2. Which is the best example of a tailored checklist item?
Tailored items are specific to the errors that actually happen in your work, which makes them quick to apply and likely to catch something.
3. A checklist item hasn't found a problem in six months of use, on low-stakes work. What should you consider?
Part of the improvement loop is trimming checks that never fire, so the checklist stays usable.
Up Next
Time to put it all together in the course challenge.