32% of New Hires Faced an AI Test. Non-Coders, Rehearse a Live Demo

ZipRecruiter's latest New Hire Survey found that 32% of recently hired workers were tested on AI proficiency during hiring, and 35% ran into AI somewhere in the process. If you're a marketer, analyst, designer, or operator who keeps hearing "show me how you use AI," your resume line won't be enough on its own. There's a real chance you'll be asked to use AI while someone watches.

What does an AI proficiency test look like if you don't code?

Outside engineering, the best-documented AI proficiency test is a live task, such as designing a workflow for an ops role, paired with the question "How do you currently use AI in your daily work?" The interviewer is watching how you prompt, push back on bad output, and change course, not just what you produce.

An AI proficiency test is any hiring step where the employer watches or scores how you actually work with AI tools, as opposed to reading that you know them.

Zapier has published the most detail on how this works. Brandon Sammut, Zapier's Chief People and AI Transformation Officer, and St. Dic have described AI fluency being checked at four separate points:

1. The initial application.
2. An AI-powered interviewer at the recruiter screen. This is where "How do you currently use AI in your daily work?" comes up.
3. A live-iteration skills test. For ops roles, this might mean designing a workflow.
4. The executive interview.

Zapier CEO Wade Foster published an AI Fluency Rubric in May 2025 and a V2 on March 31, 2026. In V2, the "Capable" level ("I use AI to operate at a meaningfully higher level") is the minimum bar for every hire, not a nice-to-have. Growth Method has published a version of that rubric adapted for marketing teams. If you're preparing for an AI test for a marketing job interview, read it before anything else.

How common are AI skills assessments in job interviews now?

About one in three recent hires was tested on AI proficiency, and the share is climbing fast. In ZipRecruiter's Q4 2025 survey, 22.1% of new hires had some form of AI assistance in their interview process and only 20.8% were asked or encouraged to use AI. Now it's 35% who encountered AI and 32% who were tested on it.

The funnel shows why each test matters so much. ZipRecruiter's median new hire sent 16 applications, sat 5 interviews, and got 2 offers over 5 weeks. Most applications never turn into a conversation. When one finally does, and that interview includes a live AI task, you're spending one of very few real chances. Freezing in front of a blank chat window wastes weeks of applying.

Greenhouse Chief People Officer Sharawn Tipton put the mood bluntly: "For the first time I can remember, neither side thinks hiring is working." The live AI test is one way employers are trying to see past the paperwork.

Why won't an "AI" line on your resume carry you?

A resume keyword won't carry you because almost no non-technical professionals signal AI in their headline anyway, and the live test grades your process, which a keyword can't show.

In Refolk's index of professional profiles, there are 88,110 current U.S. Marketing Manager profiles. Only 530 of them mention "AI" in their headline.

The index matches headlines only, not skills sections, so 0.60% measures how many people signal AI, not how many use it.

Who does say so is the interesting part. In a 25-profile sample of the U.S. AI-headline group, only 15 were actual Marketing Managers. The rest were founders, CEOs, a "Fractional CMO," and an "Artificial Intelligence Consultant." A big share of AI branding in headlines belongs to self-employed people, not in-house staff. The employed manager you're up against for the same role probably hasn't branded at all.

That leaves you with a wide signal gap on one side and a live test on the other. Zapier's process watches how people prompt, push back on bad output, and adapt. A keyword can't show any of that, and a rehearsed demo can.

Why are AI-required roles riskier, and how do you set the bar yourself?

Roles with AI requirements fall apart more often: new hires in those roles had an offer pulled 39% of the time, versus 16% overall. That suggests many employers can't define what they actually want, so a candidate who brings a clear definition fills a real gap.

The employer data shows the same confusion:

• 74% of employers call AI skills a strong advantage or a requirement for new hires.
• Half expect candidates to already be practical or advanced users.
• Only 22% provide mandatory AI training for all employees.

So companies expect fluency they don't teach and often can't describe. It's easy to see how that ends in a pulled offer: a hiring manager says yes, then someone above them asks "but are they actually good with AI?" and nobody has an answer.

Give them one. Map yourself to a public framework like Zapier's rubric levels, Capable and Adoptive, and say which level you're at, with evidence. "I'd put myself at Capable on Zapier's rubric. Here's the reporting workflow I rebuilt, and here's what it cut" is a sentence a hiring manager can repeat to their boss. You've done their definition work for them.

Do this before you apply, not after. Postings word their AI requirements in very different ways, and you want to know where you actually match before you spend a weekend rehearsing. Refolk scores how well you fit each posting, so you can see whether the AI line is a real gap or just wording before you put in the prep.

Why should you hide AI in your documents but show it in your process?

Written materials get judged by AI detection, and live tests get judged by process. So your resume and cover letter should sound unmistakably like you, and your AI fluency should show up when someone watches you work.

Greenhouse's October 21 candidate report shows candidates already working this out:

• 77% of U.S. job seekers used AI in their search in the past year.
• 60% changed or considered changing how they write so their application isn't mistaken for AI-generated.
• 39% rewrote their entire application for that reason.

Separately, 62% of job seekers tried tactics to get past AI screening, including using AI to work out what the system wanted (54%) and adding skills or keywords they didn't have (40%). That 40% is a trap for live tests. If you claimed a tool you don't use, the skills test is where it comes out.

Polished, generic writing reads like a machine wrote it. A rough live session with clear reasoning reads like a person who knows the tools. Keep the two jobs separate. Your documents should be specific to your own history, with numbers and details only you would know. Refolk writes your resume from your actual work history, tailors it to each posting, and drafts a cover letter that reads like you instead of a template. Then put your AI fluency into the demo.

How do you build a repeatable live AI demo?

Pick one real task from your job, run it with AI from start to finish several times, and rehearse the story of what broke and how you fixed it. Truffle's breakdown of these assessments puts it directly: a rough result with visible reasoning and iteration beats a polished result with no process behind it.

The structure that wins is version history: I tried X, it failed at Y, so I switched to Z. That's a story you can prepare in advance.

Build it in six steps

1. Pick a task you really do weekly. Real work, not a toy prompt.
2. Run it from scratch three times. Use a fresh chat each time. Notice where the output goes wrong in the same way every time.
3. Write down the failure and the fix. "The first draft invented a statistic, so I now paste the source data and tell it to cite only that."
4. Time it. Get your demo under ten minutes, with a two-minute version ready.
5. Narrate as you go. Say why you're rejecting an output. Silent prompting looks like guessing.
6. Prepare an answer to the screen question. "How do you currently use AI in your daily work?" deserves a 60-second answer with one named tool, one task, and one result.

Demo ideas by role

For marketers specifically, the Growth Method rubric adaptation gives you language for the level you're demoing. Name it out loud during the task.

Should you use AI live without asking first?

Don't use AI in any interview stage until you've confirmed the rules, because unapproved use can get flagged as cheating. Fabric's data, as reported by Truffle, shows AI-flagged interview cheating jumped from 9% of interviews in July 2025 to 38.5% by January 2026.

The cheating figure quadrupled over a stretch when only 20.8% of candidates were actually invited to use AI. Detection is rising faster than permission. A candidate who quietly opens a tab during a "talk me through it" exercise can look identical to one who's cheating.

Greenhouse CEO Daniel Chait advises asking about AI rules directly because they're "often not published." Refolk has a full list of AI policy questions for every interview stage, but the core ones are:

• "Is AI use allowed in this exercise, and which tools?"
• "Do you want to see my prompts, or just the result?"
• "Should I share my screen while I work?"
• "Is this stage scored by a person or an AI interviewer?"

Ask once, in writing if you can, at the start of each stage. Asking is itself a fluency signal. People who use AI seriously know the rules differ by context.

FAQ

How do I show AI skills in an interview if I'm not technical?

Bring one rehearsed task from your actual job and run it live with narration. Explain what you tried, where the output failed, and what you changed. Map yourself to a public framework like Zapier's AI Fluency Rubric so the interviewer has a shared definition of "good." A rough result with visible reasoning beats a polished one with no process.

Should I put AI in my resume headline?

It won't hurt, but don't expect it to carry you. In Refolk's index only 530 of 88,110 U.S. Marketing Managers mention AI in their headline, and in a sample of that group many were founders and consultants rather than in-house staff. The live test grades process, so a specific bullet with a measurable result plus a rehearsed demo does more than a headline keyword.

What are common interview questions about using AI at work?

The one that's publicly documented is Zapier's screen question: "How do you currently use AI in your daily work?" Beyond that, prepare to walk through a recent task you did with AI, to explain a time AI gave you a wrong answer and how you caught it, and to do a live exercise. Have a 60-second answer for the first, a version-history story for the middle two, and a timed demo for the last.