Your manager forwards the summary you wrote and asks one question: “Where’s this number from?”
You know the answer. AI gave it to you. You didn’t verify the AI output. It sounded right, so you moved on. Now you’re opening tabs, looking for a source that may not exist.
That moment is the AI skill gap most people miss. It isn’t about whether you can use AI. It’s about what you do after the answer shows up.
The gap isn’t where you think
We’ve been watching this in our own data. Among the people who’ve taken the free GAIL AI Literacy Assessment so far, students score close to working professionals on the pillars about operating AI: 82 vs. 84 on AI Foundations & Limits, and 81 vs. 83 on AI Interaction, Collaboration & Task Design.
Critical Thinking & Verification is where it opens up: 77 vs. 86. About 1 in 5 students scored below 60 on that pillar. For professionals, it’s about 1 in 20.
That’s early, first-party data from over 100 people who chose to take the assessment (students and professionals), not a representative study. But it lines up with what larger surveys are finding.
In a 2026 survey of UK university students, 85% of those who use AI said they’d found problems with what it produced. Only 43% said they usually or always verify the output, and 19% said they rarely or never do (King’s College London Policy Institute, AI and the Future of Work, 2026).
So most students have seen AI get it wrong. Fewer than half check every time.
77 vs. 86
Average Critical Thinking & Verification score, students vs. working professionals (GAIL AI Literacy Assessment, early first-party data)
43%
of UK university students who use AI usually or always verify its output (King’s College London, 2026)
1 in 3+
entry-level jobs now require AI skills, per employers (NACE, April 2026)
Why this is the skill employers notice
More than one-third of entry-level jobs now require AI skills, nearly triple the share employers reported in fall 2025. Nearly 60% of employers say they’re assigning interns projects that use AI tools (NACE, Job Outlook 2026 Spring Update).
Faculty see the risk from the other side. In a survey of 1,057 faculty, 90% said generative AI will diminish students’ critical thinking skills, and 63% said their spring 2025 graduates were not very or not at all prepared to use it at work (Elon University and AAC&U, January 2026).
Put those together and the picture is clear. Employers expect you to use AI. What earns their trust is whether your work holds up when someone checks it.
Why smart people skip the check
Fluency is not accuracy. AI writes in the same confident, polished voice whether it’s right or wrong. And the smoother a draft reads, the less likely most of us are to question it.
Deadlines make it worse. When the draft looks done, checking feels like extra work. But the moment you send it, every claim in it is yours. AI doesn’t sign its name to your work. You do.
A 10-minute verification habit
Take five checks before you hit send on any work you create and collaborate on using AI.
- Circle the claims that would hurt if they’re wrong. Numbers, quotes, names, dates, and citations. Start with the one that would embarrass you most.
- Trace each one to a primary source. The report itself, not the blog that quoted it. If you can’t find it, it doesn’t go out.
- Make sure every citation exists. AI can invent a plausible journal, attach a real author to a paper they never wrote, or hand you a link that goes nowhere.
- Cross-check with a second tool or a quick search. Use it as a checker, not a co-author.
- Make the call. Ship it, add a caveat, or start over.
These five checks come straight from the verification skills in Pillar 3 of the GAIL framework, Critical Thinking & Verification. For a real-world look at what happens when a fake source slips through, read The $440,000 Citation.
Calibrated, not paranoid
None of this means distrusting everything AI gives you. A brainstorm or a rough first draft doesn’t need a fact-check. A class paper, a client deck, or a post with your name on it does.
The goal is calibrated trust: knowing what to check, and checking it every time it matters. That’s a habit, and habits get built with reps.verify AI output
So here’s your rep for this week. Pick one thing you’re about to send that AI helped you write. Find the claim you’d least want a manager or professor to question, and trace it to its source before you hit send.
