Write the learning outcomes as things a student can demonstrate, not topics to cover. “Critically evaluate two competing treatment protocols using current evidence” gives the generator (and eventually your marking) something concrete to assess; “understand treatment protocols” doesn’t.
Be explicit about submission format up front. Whether you want a written report, a recorded presentation, a portfolio, or something else changes what “complete” instructions look like — deciding this before you generate, rather than editing it in afterward, produces a tighter, more coherent set of instructions and matching marking criteria.
Check the marking-criteria table against your own rubric conventions. It’s generated to be complete and specific, but if your department has a required rubric structure, a fixed number of criteria, or particular grade-band language, compare the two before issuing the assessment to students — this is also a good moment to reach for the Rubric Generator if you want that criteria table refined into a full four-level analytic rubric.
Don’t skip the academic integrity note, even for low-stakes tasks. It’s generated specifically for the task at hand rather than being generic boilerplate, so it’s worth actually reading rather than assuming it says what a standard note would say.
Use “adapt for a group” before the assessment goes out, not after a complaint comes in. If you already know a cohort has students who’ll need adjusted submission formats or timeframes, generate that version as part of your prep rather than reactively.