Workers Are Using AI to Learn on the Job — But 65% Don't Trust What It Teaches Them

A new Fast Company report on a survey commissioned by the American College of Education found that 63% of workers now rely on AI tools like ChatGPT to learn skills their employer never formally taught them. But there's a catch: 65% of those same workers worry the AI's answers may be inaccurate. That tension — using the tool anyway, while quietly doubting it — is rewriting how careers are built in 2026. Metaintro takes a look at what the data means for everyday workers and what it takes to learn safely from a machine that sometimes makes things up.

What does the data actually say about AI learning at work?

The American College of Education survey polled 1,046 U.S. workers who use AI on the job. The headline numbers are striking. Sixty-three percent say AI has taught them something their employer never did. Twenty-three percent say AI is now their first stop when they need to learn something new — nearly tied with asking a colleague (24%) and well ahead of asking a manager (13%). Almost a third (29%) said they use AI precisely because it lets them learn without admitting they didn't already know.

In other words, the modern workplace has a new backchannel. It isn't the breakroom or a Slack DM. It's a chatbot window. And it's open all day.

The productivity gains are real. Sixty-nine percent of workers said AI improved their output. More than half — 55% — said it made them feel more confident doing their jobs. For a generation told to "always be upskilling," a tool that teaches on demand, judgment-free, is hard to resist.

Why do 65% of workers distrust AI's answers?

Because the technology still gets things wrong in ways that matter. Research published through 2026 consistently documents what engineers call "hallucinations" — moments when an AI invents a fact, a source, or a step. Recent benchmarks show GPT-5 makes roughly 45% fewer factual errors than earlier versions and six times fewer fabricated answers, but even the best models hallucinate around 4.8% of the time with advanced reasoning turned on. On short factual Q&A tasks, some studies have measured accuracy as low as 49%.

Citations are an even rougher picture. Academic analyses of AI-generated references have found that as few as 14% reliably link to real, verifiable sources. More than half of tested citations were either fabricated or contained errors. For a worker trying to learn a new compliance rule, a safety protocol, or a tax regulation, that's the difference between getting it right and getting in trouble.

Workers seem to sense the gap. Only 7% said learning from AI alone is "enough." Thirty-nine percent said they treat it as a starting point and go verify elsewhere.

What are workers actually using AI to learn?

The survey doesn't name specific tools, but the pattern across workplaces in 2026 is consistent. Workers are using AI for:

• Software and tool fluency — How to write an Excel formula, build a pivot table, or format a slide deck in Microsoft Office or Google Workspace.
• Writing and communication — Drafting emails, summarizing meetings, rewriting awkward sentences, and translating jargon.
• Coding and data tasks — Debugging scripts, explaining unfamiliar code, or drafting SQL queries for non-engineers who suddenly need to pull a report.
• Policy and process lookups — "What's the rule for expensing travel?" "How do I structure a PIP conversation?" — the questions people used to ask a manager or HR.
• Industry background — New hires using AI to quickly ramp on acronyms, frameworks, and history they didn't learn in school.

That last category is why 48% of workers said AI sparked enough curiosity to push them into formal training they'd either enrolled in or seriously considered. AI isn't replacing learning. For many, it's the on-ramp.

Where does AI most often get it wrong?

The accuracy gaps tend to cluster in predictable places, and knowing them makes the tool safer to use.

Anything requiring current, specific facts — a recent regulation, a company's latest org chart, a newly released product spec — is risky. AI models are trained on past data and can confidently report outdated information as current. Anything involving numeric precision — a salary range, a tax bracket, a dosage, a deadline — should be cross-checked against a source of truth. And anything that sounds like a citation ("according to a 2024 Harvard study...") deserves scrutiny. If the AI can't produce a working link, treat the claim as unverified.

AI also tends to flatten nuance. Ask it "is this policy legal?" and it will often give a clean-sounding yes or no when the real answer depends on your state, your industry, and your specific contract. For anything with legal, medical, financial, or safety consequences, AI is best used as a briefing document before talking to a human expert — not as a substitute for one.

How can you verify what AI teaches you?

The workers who get the most from AI learning share a few habits worth borrowing.

Ask for sources, then check them. Most major chatbots now return citations when asked. If a link doesn't resolve or a quoted study can't be found on the publisher's site, assume the claim is unreliable.

Cross-reference two tools. If ChatGPT and Claude or Gemini give materially different answers to the same question, that's a flag. Reconcile them against an authoritative source before acting.

Test it before you ship it. For code, run it. For formulas, spot-check the math. For a policy explanation, read the actual policy. AI is best used as a translator and first pass, not a final draft.

Keep a learning trail. When you ask AI how to do something new, save the prompt, the answer, and the verification step that confirmed (or corrected) it. Over time, you build a personal knowledge base that's more reliable than any single session.

Escalate for high-stakes decisions. If the consequence of being wrong is legal, financial, medical, or career-altering, AI gets you the question, not the answer. Take it to a qualified human.

What does this mean for employers?

The survey is, in many ways, a report card on formal training programs. When 63% of workers say AI is filling gaps their employer left open, that's a clear signal to HR and L&D leaders: the need is there, the spend isn't keeping up, and workers are routing around the system.

Smart employers in 2026 are responding in three ways. First, they're publishing approved AI usage guidelines so employees know which tools are sanctioned and what data they can and can't paste into them. Second, they're investing in verification training — teaching people how to spot hallucinations, check citations, and when to escalate. Third, they're treating AI-assisted upskilling as real learning by surfacing it in performance reviews and career paths rather than pretending it isn't happening.

Ignoring it isn't a strategy. Twenty-nine percent of workers said they're already learning from AI quietly, precisely to avoid looking underprepared. That's a culture problem disguised as a tech problem.

What should workers take away from this?

AI is the fastest on-ramp to new skills that most workers have ever had — and it's still unreliable enough that blind trust is a career risk. The workers who thrive aren't the ones using AI the most or the least. They're the ones who use it with a verification habit built in.

Treat AI like a well-read colleague who sometimes misremembers dates. Useful for brainstorming, rough drafts, and "where do I even start" moments. Not a substitute for primary sources, experts, or your own judgment. The 7% who think AI learning is "enough" are the ones most likely to get caught out. The 39% who see it as a starting point are the ones building durable skills.

People Also Asked

Q: Is it safe to use AI to learn job skills?

A: Yes, when you verify what it tells you. AI is strong for explanations, examples, and first drafts, but current research shows it still hallucinates facts and citations at meaningful rates. Use it as a starting point, then cross-check important details against authoritative sources or a qualified human before you act on them.

Q: What percentage of workers don't trust AI's accuracy?

A: According to a 2026 American College of Education survey of 1,046 U.S. workers, 65% worry that AI's answers may be inaccurate or incomplete. Despite that concern, 63% still say AI has taught them skills their employer never formally trained them on — a sign that workers are using the tool and verifying it rather than avoiding it.

Q: Should I tell my manager I'm using AI to learn on the job?

A: In most cases, yes — especially if your employer has published AI usage guidelines. Being transparent about how you use AI builds trust, protects you if something goes wrong, and often opens the door to formal training budget. The 29% of workers who hide their AI use are more exposed when mistakes happen than those who learn openly.

Ready to future-proof your career?

Future-proof your career by pairing AI-driven learning with roles that actually value it. Metaintro curates openings from companies investing in upskilling, AI fluency, and modern L&D — so you can put your new skills to work somewhere they'll be recognized. Sign up to get matched with employers building for the way people actually learn in 2026.