You got certified, you built something, you used AI at work. Now someone is going to ask you about all three in a room. The exact questions, the answer that works, the answer that loses it, and how to hold your number when they push.
Doing the three moves gets you the interview. Answering the questions about them gets you the job. Almost nobody preps the second one.
The moves are simple: get certified, build something real and put a number on it, pair what you know with AI and check the work. People do all three and fall apart on the follow-up, because they never said it out loud. "I automated our weekly report" on paper is not the same as saying it to someone who asks, quietly, "how do you know it saved time?"
Read This With Its Companion
This guide is what to say in the room. The part before it, the outreach, the proof, the first 90 days, is 3 Claude Moves That Actually Get You Hired. That one is what to do; this one is what to say.
A certification gets your resume read, not hired. What they are checking is whether the badge means anything, because they have met six people this month who had one and could not name a thing they learned from it.
Question One
"What did you actually learn?"
Answer shape: one concept, one place you used it. Do not list the modules.
Works: "How much context these tools need. Instead of one line I now give it our brand rules and three past emails, and a draft went from an hour to fifteen minutes." Loses it: "It covered prompt engineering, model types, and use cases." That is the table of contents.
Question Two
"What did it not teach you?"
Admitting a gap makes everything else more believable. Works: "The vocabulary and the shape of things, and nothing about a real messy company, like our data living in four places that do not match." Loses it: "It covered everything." Nobody believes that.
The Probe Most People Fail
"Explain one term from that course like I have never heard it." A hallucination, a token, fine tuning. Plain words, twenty seconds. If you can only explain it using other words from the same course, you memorized it. Pick your three likeliest terms tonight and say each to the wall.
Which cert to actually take. You do not need one to get hired, but a credible name helps. Take one of these, all built for non-engineers:
Full walkthrough with enroll steps: 3 AI Certifications For Career Switchers.
This is the part you can actually win. Most candidates talk about what they would do. You have a thing that exists.
Question One
"Walk me through something you built."
Answer shape, four sentences: what was annoying, what you made, what changed, what it cost you.
Works: "We answered the same forty questions all day. I built a helper that reads our old tickets and drafts a reply for the agent to approve. Nothing goes out untouched. Two weekends, and I rebuilt the search twice." Loses it: starting with the tools. "So I used Claude and connected it to a spreadsheet and then." They do not know the problem yet, so nothing lands.
Question Two
"What broke?"
Have a real failure ready. Works: "The first version summarized a file missing two weeks of data and said nothing. The number looked fine and was wrong. Now it checks the row count and dates first, and stops if anything is missing." Probe, "what would you do differently?" wants a rule, not a feature: "I build the checks before the output. If I cannot tell when it is wrong, I do not trust it when it looks right."
When you show it, bring a two to three minute screen recording, not a live demo that dies on guest wifi. Record it at home with a real error case, and ask before you share your screen.
"You said it saved six hours a week. How do you know?"
This is the question that decides it, and nobody preps for it. Everyone tells you to put a number on your work; nobody warns you a good interviewer asks where it came from. If your answer is "it just feels faster," every other number you gave gets quietly downgraded. Not because you lied, but because you cannot show your work.
The fix takes about twenty minutes. Measure it honestly, before the interview, in four parts.
Part One
The baseline
What it took before, and how you know. Best case you timed it; realistic case you reconstruct it from the calendar block or the file timestamps. Write it as a range: "two and a half to three hours" beats "three hours", because it sounds like you paid attention instead of rounded.
Part Two
The after
Time it for real, at least three times, including the runs where you had to fix something. End to end: if the tool takes four minutes but you spend eleven checking its output, your after number is fifteen, not four. Interviewers ask about checking time, so have it in there.
Part Three
The sample size
How many times you ran it, over how long. "Eight weeks, once a week" is a real claim; "I did it twice" is a guess with a number attached. If your sample is small, say so: "Over five runs, about forty minutes instead of three hours. That is not many, so I would want another month before I put it in a report."
Part Four
What you did not count
One sentence, and it wins the room. Name the cost you left out: "That does not count the two weekends I spent building it, or the ten minutes a week I still spend checking. Count the build and it paid for itself around week seven." Almost nobody does this, and it makes you the person who does the math properly.
How to say it, in one sentence. Number, how you got it, how long. "It went from about three hours a week to about forty minutes, timed across eight weeks, and that forty includes checking." Say it out loud until it is smooth, because the first time in the room it comes out wobbly, and wobbly sounds unsure.
Then the pushback that decides it. "That seems high." Do not defend, do not fold, narrow it: "The before is my reconstruction, not a measurement, so call it two to three hours. Even at the low end that is over an hour a week, and I would rather quote you the low end."
Never measured it? Run it three times this week with a timer: "I only started timing recently, so I have three runs: 38, 45 and 41 minutes. Before, it was most of a morning." That beats a confident six that cracks on the first push. Never round up: one number they can poke a hole in costs you every other number.
The least flashy move, and the one that separates people. Anyone can get output; the person who gets hired knows when it is wrong. AI plus a beginner gets fast wrong answers. Someone who knows the work catches them.
Question One
"How do you check what AI gives you?"
Answer shape: a specific routine, not a value. Works: "Three things every time. I check the totals against the source, because a summary that adds wrong still reads fine. I spot check three line items. And anything with a name or a date gets verified, because that is where it invents things." Loses it: "I always review it carefully." A promise, not a process, and everyone says it.
Question Two
"What would you never hand to AI?"
Having a line is the answer. Not having one is the red flag. Works: "Anything going to an upset customer goes out in my words. It can draft, I do not send its draft. Tone is the whole job in that moment, and it does not know what happened to that person last week."
The Probe Most People Fail
"How would you know if it was wrong?" Not what you do when you catch it. How you would catch it at all. Most people freeze, because they check by feel. The answer is a thing that fails loudly: totals that must match a source, a row count, a date range, a link you can click. It is what a manager is really asking: can I hand you something and not check behind you.
One line to end on. If they ask what makes you different from everyone else using the same tools: "The tool is the same for everyone. What I bring is knowing what right looks like in this job, so I can tell in ten seconds whether what it handed me is usable." Then say it all out loud first, to a wall or a friend. The moves got you the meeting; the words are what you owe yourself now.
© 2026 Mariah Brunner. All rights reserved.