AI interview skills: the 30-day plan that closes the gap your score revealed

Your score told you which signals are soft. Find them below, start with the highest-weight gap on your list, and run that play for 30 days on your actual work before moving to the next. If you scored 80 or above, you may have one gap or none. If you scored below 60, plan on two to four plays.

1. 1

   Signal 1 (25 pts): the live demo

   **What employers test:** can you drive a real task from your role, screen-share, no slides, and brief the AI with context instead of typing a one-line prompt? **What 30 days looks like:** pick one workflow from your actual job, run it AI-first every week, and rehearse narrating your choices out loud until the muscle is there. The interviewer's test is watching the first five minutes; fluent people brief the AI like a context document instead of treating it like a search bar. **What done looks like:** you can say yes when asked to demo live, and the demo is boring to you because you have run it twenty times. **Derails when:** you prepare a demo on a toy task instead of the real work the role actually does. Interviewers spot this in the first minute.

2. 2

   Signal 2 (20 pts): verification habits

   **What employers test:** can you describe specific checks you run on AI output before it ships, beyond 'I read it over'? **What 30 days looks like:** write your verification checklist for the two or three output types you ship most, then use it out loud in every session this month. The AI verification habit is learnable but only builds through repetition on real work. **What done looks like:** you can name the exact check that catches the subtle error in your role, and you have a story about a time it fired. **Derails when:** you describe process in the abstract instead of having a specific catch story. A real story from real work closes the gap a general answer never can.

3. 3

   Signal 3 (15 pts): outcomes with numbers

   **What employers test:** does your resume show before-and-after numbers instead of tool name-drops? **What 30 days looks like:** rewrite every AI bullet as an outcome with a number. 'Cut report prep from four hours to 40 minutes' beats 'used ChatGPT' in every screen. Track one before-and-after this month on real work so the number is yours. The resume guide has the exact patterns. **What done looks like:** every AI claim on your resume has a number you can defend instantly when asked. **Derails when:** you estimate or round generously and then can't defend the number in the room. The number earns trust; a wrong number destroys it.

4. 4

   Signal 4 (15 pts): a real automation

   **What employers test:** have you built a workflow automation that runs without you, something beyond chat sessions? **What 30 days looks like:** automate one recurring task end-to-end, even a small one. The interview story is the artifact: what it does, what it gets wrong, and how you find out. Failure modes are the proof point, because only people with real automations have watched one break. **What done looks like:** something runs on schedule, you have found one thing it gets wrong, and you can sketch its failure mode in bullets. **Derails when:** you describe an automation you designed but never shipped. Interviewers who have built automations can tell.

5. 5

   Signal 5 (10 pts): agent direction

   **What employers test:** have you directed AI agents on multi-step work, delegating a task and reviewing the result, instead of only running single prompts? **What 30 days looks like:** hand an AI agent one multi-step task per week for a month: scope it in writing, let it run, review the output like a manager checking a report. Directing is a habit, and the language it produces (scope, checkpoints, review) is different from prompt-and-accept language. **What done looks like:** you can describe a multi-step delegation with what you scoped, what came back, and what you changed. **Derails when:** you conflate 'I used a long prompt' with 'I directed an agent.' The distinction shows in the vocabulary.

6. 6

   Signal 6 (10 pts): currency

   **What employers test:** have you adopted a new AI tool or capability in the last 90 days, with a concrete example? **What 30 days looks like:** adopt one thing this month and use it on real work. Recency beats certificates from two years ago; a 2023 story in a 2026 interview is a quiet red flag. **What done looks like:** you can name what you adopted, what it does that your old setup could not, and where you used it. **Derails when:** you follow the news without hands-on use. 'I have been watching this space' is the answer that signals you have not crossed the line from interested to capable.

7. 7

   Signal 7 (5 pts): proof by shipping

   **What employers test:** do you hold a credential or public proof that required shipping something real? **What 30 days looks like:** ship one portfolio piece, a before-and-after on a real task, or earn a certification that requires real work instead of completion clicks. **What done looks like:** there is something linkable or showable that a skeptical interviewer can look at. **Derails when:** a watch-the-videos certificate is the only proof. Completion certificates predict the live demo going well about as well as a swimming badge predicts a race.