Can You Spot AI-Generated Work? A Team Exercise

AI-generated text is now everywhere, and most of us can't reliably tell it from human writing. That matters at work: you might be reading an AI-drafted report, reviewing AI-written copy, or receiving AI content presented as someone's own thinking. Being unable to spot it leaves you vulnerable to confident-sounding errors and shallow work dressed up as substance. But the skill is rarely taught, so people either trust everything or distrust everything — both wrong.

Learning how to spot AI-generated content isn't about catching people out. It's about reading more critically, knowing AI's characteristic weaknesses, and judging content on its merits rather than its polish. A blind comparison exercise builds that eye fast.

How can I tell if content was written by AI?

There's no certain test, and you should be honest about that upfront — confident "AI detection" claims are often unreliable, and humans can write generically while AI can be edited to sound human. What you're building is informed suspicion, not proof. With that caveat, certain tells recur. AI text often defaults to generic, hedge-everything phrasing and a smooth, even tone that never quite takes a position. It can be confidently vague — lots of words asserting little. It sometimes repeats sentence structures or transitions. And most usefully, it can state plausible-sounding "facts" that fall apart on checking, because it generates what sounds right, not what is right.

The deeper skill is shifting from "who wrote this?" to "is this actually good and true?" Specific detail, lived experience, a genuine point of view, and verifiable facts are hard to fake — and they're what you should value whether a human or an AI produced the draft. Judge the substance, and the source matters less.

How to run the AI vs. Human exercise, step by step (about 20 minutes)

You need a few text samples — some AI-generated, some human-written — on similar topics, and a group.

1. Prepare a blind mix. Gather, say, six short passages on a theme: some written by people, some by AI. Don't label them. Keep an answer key.
2. Have everyone guess, individually. For each passage: human or AI? And crucially, why — what's the tell? The reasoning matters more than the guess.
3. Reveal the answers and tally the score. People are usually humbled by how hard it is. That surprise is the point — it dismantles overconfidence in both directions.
4. Discuss the tells that worked and the ones that fooled you. Where AI passed as human, and where humans seemed "too perfect." Build a shared list of honest signals, not myths.
5. Stress-test a "fact." Take a confident claim from one AI sample and check it together. Seeing a plausible statement turn out wrong is the most useful lesson of all.
6. Agree on a practical norm. Not "ban AI" or "trust AI," but "judge the content: is it specific, true, and useful?" That's the durable takeaway.

A worked example

A team runs the exercise with six product-blurb passages. Most people score barely better than a coin flip, which surprises them. In discussion, the tells that actually helped weren't stylistic — they were substance: the human-written blurbs had a specific anecdote and an opinion, while the AI ones were smooth but said nothing concrete. Then they fact-check a confident line from one AI sample and find the statistic it cited doesn't exist anywhere. The team leaves with a sharper rule than "spot the AI": demand specifics and verify claims, no matter who or what wrote it.

When this is most useful

This exercise is valuable for any team that reviews written work, evaluates content, or increasingly receives AI-assisted material — editors, marketers, managers, researchers. It's great digital-literacy training and a healthy antidote to both AI hype and AI panic. It's less useful as a reliable gatekeeping tool: don't use "I think this is AI" to accuse someone, since you can't actually prove it and you'll be wrong sometimes. The goal is critical reading, not detection theatre.

The takeaway

You probably can't reliably tell AI from human writing — and pretending otherwise is its own risk. Run a blind comparison exercise to learn the real tells, then shift the question from "who wrote this?" to "is it specific, true, and useful?" Stress-test a confident claim and watch it crumble, and you'll build the only defence that actually holds: judging content on substance, and verifying facts regardless of their source.

This is one of Funstorming's 100 quests — bite-sized soft skills methods you actually put into practice, not just read about. Try it, then bring your result (or your sticking point) to the Funstorming community of practice (CoP), FunHub | Your Soft Skills Playground.

#funstorming #softskills