You Can't Interview for AI Fluency

Resumes, interviews, and certifications all fail to measure the skill that matters most. Here's what the research says actually works.

Everyone agrees that AI fluency matters. Companies are screening for it, candidates are claiming it, and training providers are certifying it. But there's a problem nobody wants to talk about: the tools we use to assess AI fluency don't work.

Resumes can't verify it. Interviews can't surface it. Certifications can't prove it.

How do we know? Because 79% of tech workers admit they pretend to know more about AI than they actually do. That's not a rounding error. That's a survey of 1,200 technology professionals confessing, anonymously, that they're faking it. C-suite executives are the worst offenders at 91%.

And 65% of organizations have already abandoned AI projects because the people they hired didn't actually have the skills they claimed.

The assessment gap is real. And it's expensive.

Resumes are self-reported fiction

A resume is a marketing document. It tells you what someone wants you to believe about their skills, not what they can actually do.

This has always been true, but AI fluency makes it worse. There is no standardized definition of "proficient in AI tools." One candidate means they've used ChatGPT to rewrite emails. Another means they've built AI into every stage of their workflow. Both put the same line on their resume.

TestGorilla's 2025 State of Skills-Based Hiring report surveyed over 1,000 hiring decision-makers and found that one in three recruiters can't tell from a resume whether a candidate actually has the skills they claim. 36% can't determine whether a resume is even accurate. 86% of employers report having problems with resumes as a hiring tool.

The market is responding. Resume usage in hiring dropped from 73% to 67% in a single year. 85% of companies now use some form of skills-based hiring. But for AI fluency specifically, most haven't figured out what "skills-based" even means.

Interviews measure confidence, not competence

The standard hiring interview is an unstructured conversation. The interviewer asks whatever comes to mind. The candidate answers however they want. Both walk away feeling like they learned something.

They didn't.

The largest meta-analysis of hiring research ever conducted - Sackett et al. (2022), re-analyzing 85 years of personnel selection data - found that unstructured interviews have a predictive validity of just .19. That's on a scale where 1.0 is perfect prediction and 0 is random chance. An unstructured interview is barely better than flipping a coin.

For AI fluency, it's even worse. When you ask "how do you use AI in your work?", you're measuring storytelling ability. The candidate who gives a polished, confident answer about their AI workflow sounds more fluent than the one who pauses and says "it depends on the task." But the second answer reveals more judgment than the first.

This is compounded by a confidence problem that runs in both directions. ManpowerGroup's 2026 Global Talent Barometer, surveying 14,000 workers across 19 countries, found that regular AI usage jumped 13% - but confidence in using technology fell 18%. More people are using AI while feeling less sure they know what they're doing. In an interview, the over-confident candidate who barely uses AI will outperform the genuinely fluent one who understands enough to doubt themselves.

Certifications test the wrong skills

AI certifications are booming. Every platform, every vendor, every university is offering one. But they share a fundamental flaw: they test knowledge, not judgment.

A certification can verify that someone knows what a large language model is, how to structure a prompt, or which API to call. It cannot verify that they know when not to trust the output, when to push back on an AI recommendation, or when to abandon AI entirely and think for themselves.

The OECD's 2025 report on the AI skills gap found that only 1% of AI-exposed jobs require specialized, complex AI skills. The vast majority need general AI fluency - the ability to use AI effectively within the context of a specific role. Yet most AI training programs target the 1%: engineers and data scientists with advanced prerequisites. The 99% who need practical judgment get certifications that don't match their actual work.

Gartner sees where this is heading. Their 2025 strategic predictions forecast that by 2026, half of global organizations will require "AI-free" skills assessments during hiring - testing what candidates can do without AI, not just with it. Why? Because AI is creating what Gartner calls skill atrophy: workers who appear competent because AI is doing the thinking for them, but who can't function independently when the tool breaks or the task falls outside its capability.

A certification tells you someone studied. It doesn't tell you someone can think.

What the research says actually works

If unstructured interviews score .19 and resumes are unreliable, what predicts job performance?

The same Sackett et al. meta-analysis found the answer: structured observation. Structured interviews - where every candidate faces the same questions, evaluated against the same rubric - score .42, more than double unstructured interviews. Work sample tests, where candidates perform actual job tasks, score .33.

The pattern is clear. The closer your assessment gets to observing real work behavior, the better it predicts real work performance.

For AI fluency, this means one thing: you have to watch people use AI in context. Not ask them about it. Not test them on vocabulary. Not check whether they have a certificate on the wall. You have to put them in a realistic work scenario - the kind of situation they'd actually face in the job - and observe what they do.

These behaviors can't be self-reported. They can't be certified. They can only be observed.

That's exactly what hiAIre does. Role-specific work simulations where candidates interact with AI in realistic scenarios - then scored across behavioral dimensions that reflect how they actually think, not how they describe themselves. Observable evidence of judgment, not claims about it.

The hiring process needs to catch up to the skill it's trying to measure. AI fluency is a behavior, not a credential. Assess it like one.