AI can draft a course in minutes: objectives, content, a quiz, the works. What comes back usually looks finished. That's the risk. Looking finished and being well designed are different things, and the gap between them is exactly where an instructional designer earns their keep.
This is a practical checklist for reviewing a course that an AI tool drafted, whether it came from a chatbot, an authoring tool's AI feature, or a course generator. Use it before anything goes to learners.
What AI drafts tend to get wrong
These patterns come up again and again in generated courses. None of them are hard to fix once you know to look.
Objectives that can't be measured
Generated objectives lean on "understand", "learn about", and "be aware of". They read smoothly and can't be assessed. Rewrite each one around an observable verb. (See Bloom's taxonomy verbs.)
Quizzes that test recall, not the skill
The easiest questions to generate are definitions and facts pulled from the source text. If the objective asks learners to apply or decide, those questions can't tell you whether they can. (See aligning objectives and assessments.)
Content in the order of the source document
Paste in a policy and you often get the policy back, reorganized into slides. That's the order the document was written in, not the order people learn in. Check for a hook, a connection to what learners already know, and practice before the assessment.
No real practice
Drafts tend to explain, then quiz. The middle step, where learners try the skill on a realistic case and get feedback, is usually thin or missing.
Confident facts that need checking
Generated text can state statistics, dates, and rules with complete confidence and no source. Every factual claim needs to go past your SME before it goes past a learner.
Generic examples
"A company", "an employee", "a customer". Learners transfer skills better from examples that look like their own work. Swap in your organization's real situations.
Accessibility left as an afterthought
Placeholders for images with no meaningful description, video suggestions with no transcript, and long unbroken screens. Check what the content plans for, not just what the final build passes.
The review checklist
- Every objective uses an observable verb at the level the job needs.
- Every assessment item maps to an objective, at the same level.
- Every objective is practiced, with feedback, before it's assessed.
- The course opens with a reason to care and a link to what learners already know.
- Content is sequenced for learning, not copied in source-document order.
- Every statistic, date, and rule has been verified by an SME.
- Examples come from your learners' real work.
- Text is chunked; narration doesn't just read the screen aloud.
- Every meaningful image has a real description; every video has captions and a transcript.
- A learner can tell where to start, what to do next, and how far along they are.
The part AI can't do
None of this is an argument against using AI to draft. It's fast, and a fast first draft is valuable. But the draft is the cheap part now. The expensive part, the part that decides whether a course changes anything, is judgment: is this the right problem, did we get the real expertise out of the SME, and is this design sound? That's the work that's getting more valuable, not less.
A faster first pass
Running this checklist by hand on every AI draft adds up. Pedagrade gives you the first pass in about a minute: paste the generated course and get a scored report against the learning science and accessibility, with a cited finding and a rewrite for every issue. It runs on private AI, so your draft is never sent to OpenAI or Anthropic. Then you do the part only you can: decide what to keep.
See what it finds in your course.
Paste a course or upload the file. In about a minute you get a scored report with every finding cited and the fix written for you. Your first audit is free.
Start your free audit