Describe the task, not just the topic. A rubric for “a written case analysis requiring evidence-based recommendations, submitted as a 1500-word report” will generate far more usable criteria than one built from “written assessment” alone — the generator can only make criteria specific if the task itself is specific.
Read every cell before you issue the rubric, especially the boundary between “Satisfactory” and “Needs Improvement.” That boundary is where most student disputes over grades actually happen, so it’s worth the extra minute to confirm the distinction is one you’d be comfortable defending in a grade appeal.
Check the weights against your own judgment of what matters most. The generator produces weights that sum to 100%, but “correct by arithmetic” isn’t the same as “correct for this assessment” — if one criterion matters more to you than the generated weight reflects, adjust it before publishing the rubric to students.
Share the rubric with students before the deadline, not after. A rubric is most useful as a guide while students are still working, not as an explanation after the fact — issue it alongside the assessment brief, not as a surprise at marking time.
Use per-criterion regeneration to fix, not to fish. If a single row needs work, regenerate just that row with a clearer sense of what you want changed — regenerating the same criterion repeatedly hoping for a better result at random is a less reliable path to a good rubric than writing a more specific task description up front.