Five AI interview questions for your next marketing hire

Most marketing candidates writing “AI fluent” on their resume in 2026 cannot produce evidence of it. They’ve used ChatGPT to write a few subject lines. They’ve watched a couple of YouTube videos about Claude. Maybe they wrote a long prompt once. None of that translates into running marketing systems with AI agents.

For a founder hiring a marketer at a 5-30 person B2B SaaS, this matters more than almost any other skill check. A marketer who actually uses AI ships three to five times the output of one who doesn’t. A marketer who claims to and doesn’t is paying the same salary for half the work.

Generic interview questions about “how do you use AI” produce generic answers. You need questions that force a specific demonstration, and answers that show real reps.

Here are the five I’d ask.

1. Walk me through a workflow you automated with AI in the last 90 days

Specific. Recent. Concrete. The candidate should walk you through a real artifact, not a theoretical one. A good answer names a tool (Claude, Zapier, Make, a custom script), a trigger (when X happens), and an output (Y gets created or sent or posted). Five minutes of detail.

What you’re listening for: the candidate had a problem, built a thing, used the thing, improved the thing. Bonus points if they can show you the prompt or the workflow on a screen-share.

Red flag: vague stories with no artifact. “I use ChatGPT for first drafts” is a habit, not a workflow.

2. Show me a prompt you wrote and tuned over time

This is where the AI tourists fall off. A prompt used twice and abandoned looks nothing like a prompt used twenty times and refined. The latter has structure: explicit role, constraints, output format, examples, edge cases handled.

The candidate doesn’t need to walk in with a perfect prompt. They need a prompt with edits. Show me the v3 and tell me what changed from v1.

Lenny Rachitsky has been writing about how prompt craft is becoming a hireable skill on its own. A good answer feels closer to “operator with reps” than “person who knows about AI.”

3. Tell me about something AI got wrong for you and how you caught it

This separates people who trust the output blindly from people who treat the output as a draft. A good answer is a specific story: the AI produced something that looked right and was wrong. The candidate caught it through their own knowledge of the topic, a fact-check, or a downstream signal. They adjusted the prompt or the process.

a16z has written about why this judgment layer is the differentiator between AI-augmented operators and people who think AI replaces judgment. The candidates worth hiring have stories about when AI failed them, and they sound annoyed about it, not surprised.

Red flag: “I haven’t really run into issues” or “AI hasn’t been wrong for me.” Either they aren’t using it for anything that matters or they’re not checking the work.

4. What’s your read on the difference between Claude, ChatGPT, and the other models for marketing work

A working operator’s read is what you want. A good answer is opinionated: “Claude is better for long-form voice work because of how it handles tone. ChatGPT is faster for image generation and search-grounded answers. Gemini is my default for visual assets right now.” Specific, with reasoning.

The candidate doesn’t have to agree with your read. They have to have one.

Red flag: “They’re all pretty similar.” That answer comes from someone who’s tried one model and assumed the others are the same.

5. Describe a task you stopped doing because an agent now does it

This is the highest-signal question on the list. It’s a version of the agents-as-employees framing applied to the candidate’s own week. A real AI-fluent marketer has retired specific tasks from their personal queue and handed them to a system.

A good answer: “I used to write a weekly performance digest by hand every Monday. I built a Claude workflow that pulls the numbers, drafts the commentary, and writes the Slack post. I review and ship it in 10 minutes. That used to be 90 minutes of my Monday.” Specific task, specific time saved, specific judgment they kept.

Red flag: they can’t name a task they’ve stopped doing. That means AI is supplementary for them, not structural. They’ll work the same way they did in 2022, with a few more shortcuts.

What good answers add up to

A candidate who answers all five well looks more like a marketing engineer than a traditional marketer. They have reps, opinions, artifacts, and a working memory of what AI is bad at. They’ve fired tasks from their week. They run prompts the way a senior engineer runs scripts.

Reforge has written about this hiring shift in detail. The takeaway from operator interviews: the marketing hire that compounds in 2026 is the one who can show their stack, not the one who lists tools on their resume.

What to do this week

If you have a marketing hire coming up, three concrete moves:

1. Add these five questions to your interview rubric. Score them on specificity, not enthusiasm.
2. Ask the candidate to send a sample prompt and a five-minute screen recording of an AI workflow they actually use. Most won’t be able to. The ones who can are the shortlist.
3. Cut “AI fluent” from your job description. Replace it with “ships marketing systems using Claude, ChatGPT, or agents.” It self-filters the resume pile.

If you want help mapping which functions in your business should be staffed with AI-fluent operators versus which need a different skill profile entirely, the AI Opportunity Sprint is the four-week version of that conversation.

Jared Castronova is the founder of JAC Growth Marketing, where he builds AI-powered GTM systems for B2B companies.