30 interview questions to assess a candidate’s AI skills.

How well they work with the AI itself. The same for every role.

✓Listen for ✕Watch out for

1. 1

   When did AI last get something wrong in your work? How did you notice?

   ✓A specific moment, and the check that caught it.

   ✕“It’s usually right.”

2. 2

   Share your screen and show me a prompt you used this week. Why did you write it that way?

   ✓Context, constraints, an example, and what done looks like.

   ✕One line, and hoping.

3. 3

   When do you choose not to use AI?

   ✓Clear limits: sensitive data, judgment calls, things faster by hand.

   ✕“Never”, or “always”.

4. 4

   What do you check before AI-made work goes to someone else?

   ✓Concrete checks: sources, tests run, numbers spot-checked.

   ✕“I read it over.”

5. 5

   Which AI tools do you use for which jobs, and why?

   ✓Chat, agents and AI inside other tools, with a reason for each.

   ✕One tool for everything, no reason.

6. 6

   Tell me about a task you broke into steps for the AI instead of asking once.

   ✓Splits the work and checks each step before moving on.

   ✕Everything in one giant prompt.

7. 7

   What does an AI answer have to look like before you trust it?

   ✓It can be verified, and it matches what they already know.

   ✕Trusts it when it sounds confident.

8. 8

   What have you built to reuse: templates, custom instructions, a workflow, an agent?

   ✓Something that saves them time every week.

   ✕Starts from scratch every time.

9. 9

   What did you try to do with AI that didn’t work? What did you do instead?

   ✓Honest about the limits, and changed approach.

   ✕Can’t name one.

10. 10

    How has your work changed in the last year because of AI?

    ✓More time on decisions, less on first drafts.

    ✕“I’m just faster.”

Can they tell a real fix from one that only looks right?

✓Listen for ✕Watch out for

1. 11

   The AI’s fix makes the failing test pass. What do you check before you merge it?

   ✓Whether it fixed the cause or only hid the symptom. Edge cases. Reads the whole diff.

   ✕“The test passes, so it’s fixed.”

2. 12

   How do you review a 400-line pull request the AI wrote?

   ✓Reviews it like a stranger’s code, asks for it to be split, runs it.

   ✕Skims it and approves.

3. 13

   When do you let an agent run on its own, and when do you go step by step?

   ✓Depends on the risk and whether it can be undone. Tests as the guardrail.

   ✕Always lets it run, or never does.

4. 14

   How do you give an agent context on a codebase it hasn’t seen?

   ✓Points it to files, conventions and examples. Keeps a project instructions file.

   ✕“It figures it out.”

5. 15

   Tell me about a bug the AI introduced that you caught late. What would have caught it earlier?

   ✓Names the gap, and what they changed afterwards.

   ✕Can’t think of one.

Can they make the call the AI won’t make for them?

✓Listen for ✕Watch out for

1. 16

   You have 15 user interviews. How do you use AI to pull out themes, and how do you know it’s right?

   ✓Reads some themselves, traces each theme back to quotes, watches for rare but important signals.

   ✕Uses the summary as it comes.

2. 17

   The AI gives you a roadmap that keeps every stakeholder happy. What’s wrong with it?

   ✓There are no trade-offs. A plan that says no to nobody isn’t a plan.

   ✕“Nothing, it looks good.”

3. 18

   Which parts of a spec do you let AI write, and which do you write yourself?

   ✓Writes the problem and the why themselves. AI for structure and edge cases.

   ✕Lets AI write all of it.

4. 19

   How would you test an idea with AI before engineering builds anything?

   ✓A quick prototype or mock-up in front of users.

   ✕Waits for engineering.

5. 20

   A stakeholder sends you an AI-made analysis that supports their feature. What do you do?

   ✓Checks the inputs and the method before the conclusion.

   ✕Judges it by who sent it.

Do they know which claims, audiences and channels are real?

✓Listen for ✕Watch out for

1. 21

   The AI’s draft launch post has a great statistic in it. What do you do before publishing?

   ✓Finds the source, or cuts the number.

   ✕Keeps it because it sounds right.

2. 22

   How do you stop AI-written copy from sounding like everyone else’s?

   ✓Customer language, their own examples, a clear point of view.

   ✕Doesn’t think it does.

3. 23

   How do you use AI for competitor or customer research, and where does it mislead you?

   ✓Knows it fills gaps with confident guesses, and checks real sources.

   ✕Treats it as fact.

4. 24

   Walk me through planning a campaign with AI. Where do your own decisions come in?

   ✓Picks the audience, channel and budget themselves, and has a plan B.

   ✕Lets the AI pick the channel.

5. 25

   You have 20 AI-written ad variants. How do you test them without wasting budget?

   ✓Narrows down first, tests a few, decides beforehand what counts as a win.

   ✕Runs all 20 at once.

Do they know when a clean number is wrong?

✓Listen for ✕Watch out for

1. 26

   The AI writes a query that returns a clean number. How do you know it’s right?

   ✓Checks row counts, joins that duplicate rows, and known totals.

   ✕“The number looks reasonable.”

2. 27

   When wouldn’t you let AI choose the analysis method?

   ✓When the method decides the answer, or the data has quirks it can’t know about.

   ✕Never thought about it.

3. 28

   The CFO asks “how do you know?” about a finding you made with AI. What do you say?

   ✓Explains the method and the checks, without leaning on the AI.

   ✕“The AI calculated it.”

4. 29

   How do you give AI the context of your data model and metric definitions?

   ✓Shares definitions and schemas, plus examples of correct output.

   ✕Pastes in the raw table and asks.

5. 30

   Tell me about a number that looked right and wasn’t.

   ✓A specific story, and the check they now always run.

   ✕Can’t name one.