How AI Helps You Work with SMEs – Interviews and Reviews
Your subject matter expert (SME) has 45 minutes this week, if you're lucky. They know the content thoroughly, but they're busy, they talk fast, and their review comments arrive as "this is wrong" with no correction attached. You need their knowledge without exhausting their goodwill, and an AI SME workflow is one of the easiest ways to get both. How can AI help you get more from every hour with an SME?
AI helps most before and after the SME is in the room: it preps sharper interview questions, turns recordings into organized notes, spots gaps in what you captured, and makes review rounds faster by consolidating and clarifying feedback. The SME still supplies the expertise and the final word on accuracy, and you still run the conversation and make design decisions. What changes is the administrative work around them, which shrinks considerably. Used this way, AI SME collaboration respects the expert's time, and that respect is what keeps them willing to help on the next project.
Where Does AI SME Collaboration Save the Most Time?
The biggest savings come from the parts of SME work that involve organizing information rather than knowing it. Here's how the work usually divides.
| Task | What AI does well | What stays with you or the SME |
|---|---|---|
| Interview prep | Drafts question lists from existing documents | Choosing which questions matter for performance |
| Capturing the interview | Transcribes and summarizes recordings | Getting consent to record, asking follow-ups |
| Finding gaps | Compares notes to the outline and lists missing pieces | Deciding what's truly needed for the job |
| Review rounds | Groups comments, rewrites vague feedback as clear edits | SME confirms accuracy; you resolve conflicts |
| Accessibility checks | Drafts alt text, plain-language rewrites, transcripts | Checking every draft against the real content |
Notice the pattern: the tool handles volume and structure, while people handle judgment.
Before the Interview: Show Up Prepared
Nothing loses an SME faster than questions they've already answered in writing. Before you meet, give an AI assistant such as ChatGPT, Claude or Microsoft Copilot the existing materials, including the old course, the procedure, the product manual and last year's slides, and ask it to do three things.
- Summarize what the documents already cover, so you don't ask about it.
- List the decisions a worker has to make on the job that the documents don't explain.
- Draft 10 to 15 interview questions focused on those decisions, mistakes and edge cases.
Then edit aggressively, removing the trivia questions and keeping the ones about what people get wrong and why. Cathy Moore's action mapping approach is an excellent filter here: if a question doesn't help you understand what people need to do, it can wait.
One more preparation step worth trying is to ask the assistant to role-play a skeptical new hire and list the questions that person would ask. It's a quick way to find the obvious gaps that experts no longer see.
During the Interview: Stay Present
Ask permission to record, explain why, and tell the SME how the recording will be stored. Many people agree once they hear it means fewer follow-up emails.
With a recording or meeting transcript running (many teams use the built-in transcription in Microsoft Teams or Zoom), you can stop scribbling and actually listen, which is where the valuable material comes from. Ask "What does a beginner usually get wrong here?" and "Can you walk me through the last time this went badly?" Then follow the answer wherever it leads.
Use your company's approved tools for this, because SME interviews often include internal processes, customer details or safety incidents that shouldn't go into a personal account on a public chatbot.
After the Interview: Turn Talk Into Structure
This is the step that used to consume an entire afternoon. Feed the transcript to your assistant and ask for:
- A clean summary organized by task or decision point.
- A list of every number, rule or warning the SME mentioned, with the timestamp.
- Open questions and contradictions to confirm.
- A draft outline that maps content to job tasks.
The timestamped list is the most useful item on that menu. It lets you return to the exact moment in the recording and confirm the detail, which protects you from both your own memory and any summary errors. Send the SME a short follow-up containing only the open questions, because three focused questions get answered and a 12-page transcript doesn't.
Can AI Make SME Review Rounds Less Painful?
Yes, and this is where many teams notice the difference first. Review rounds stall for three reasons: feedback is scattered, comments are vague, and reviewers disagree with each other. AI helps with all three.
Consolidate the Comments
Export comments from your review tool or collect the emails, then ask the assistant to group them by screen or section and remove duplicates, so you get one organized list instead of four inboxes.
Clarify Vague Feedback
"This doesn't sound right" is difficult to act on. Ask the assistant to suggest what the reviewer probably meant and draft a specific question back: "Did you mean the pressure limit should be lower, or that the order of steps is wrong?" You confirm with the SME; the tool just speeds up the back-and-forth.
Surface Conflicts Early
When two reviewers give opposite instructions, have the assistant flag them side by side, then take the conflict to the person who owns the content. Don't let a tool pick a winner.
Scenario-based courses get the most out of this, because SMEs know exactly how real conversations and decisions go wrong. Once you have their stories, AI branching scenarios can turn those stories into a first draft of choices and consequences for the SME to check.
Should You Tell Your SMEs You're Using AI?
Yes, and be upfront about it. Tell them which parts of the process use AI (prep, transcription, summaries, comment cleanup) and which parts never will (deciding what's accurate). Many experts are relieved rather than worried, because it means fewer meetings and fewer repeat questions.
Being open also improves quality. An SME who knows a summary was machine-drafted reads it more carefully, while an SME who thinks you wrote every word may skim it and assume you got it right. Add a one-line note to each review package, something like "Sections marked with a star were drafted with AI from your interview. Please check them closely." It takes seconds and it sets the right expectation.
Some SMEs will want to try the tools themselves, and you should encourage it. An engineer who drafts their own rough explanation with an assistant and sends it to you often provides better raw material than a rushed email would.
What Should Stay Out of the AI Workflow?
A few categories of information don't belong in a chatbot, even a good one:
- Confidential details your organization hasn't cleared for the tool you're using.
- Personal information about employees, patients or customers mentioned in an interview.
- The final call on whether a procedure, number or warning is correct.
- Decisions about what to leave out of the course. That's a design judgment, and it's yours.
One question I always add to SME interviews, with or without AI, is who struggles with the current training and why. Ask about workers who use screen readers, who are color blind, who work in loud places or who are reading in a second language. Experts often know exactly where the old course failed people, and they rarely get asked. AI can then help you draft the plain-language versions, transcripts and descriptions those learners need, with you checking each one.
Make Every SME Minute Count With AI
AI won't replace the SME, and it shouldn't try. Its job is to make every minute with the expert count through tighter preparation, cleaner notes, fewer follow-ups and review rounds that don't drag on for weeks.
On your next project, use AI for interview preparation and transcript cleanup only, and track how many follow-up emails you send compared with your last course. If the number drops and your SME notices the difference, you'll have the evidence for using it across the rest of your process.