Your team is wrong with confidence, and the AI agreed with them.
A team member asks ChatGPT a question. The answer sounds plausible. They paste it into a report. The report goes to a client. Two weeks later, somebody notices that the answer was wrong. Confidently, fluently, completely wrong. The client noticed before anyone internal did.
This scenario plays out somewhere in your organisation every week. Maybe every day. The AI is not the problem. The problem is that the staff member did not have the habit of asking “is this actually true?” before forwarding the output.
Critical thinking has always mattered. The arrival of large language models that produce confident, polished, plausible content has made it the single most important workplace skill of the next decade. And almost nobody is being trained for it.
This post is for the L&D, P&C, or operations lead who has watched a colleague trust AI output that should not have been trusted and is starting to suspect the same thing is happening across the organisation, quietly, every day.
Why “trust but verify” is not a training plan
The standard organisational response to AI risk is to tell staff to “verify everything.” This sounds sensible. It is also useless.
“Verify everything” is not a skill. It is a slogan. Staff hear it, nod, and then go back to copying AI output into documents because they do not know what verification actually looks like, how long it should take, or when it is worth doing.
What you need is not a slogan. You need a course that teaches:
- Which kinds of AI output are most likely to be wrong
- What the failure patterns look like, and what hallucinations actually sound like
- How to verify in proportion to the stakes (you do not need to fact-check a brainstorm; you absolutely need to fact-check a client deliverable)
- Where to push back on a colleague who is over-trusting AI
- How to do all of this without becoming the office sceptic that nobody listens to
This is a skills course, not a compliance course. It looks and feels different from the compliance training your staff are used to. That is the point.
What a good critical thinking course for the AI era should cover
A custom critical thinking course should do five things.
1. Teach the failure patterns
AI tools fail in recognisable ways. Hallucinated citations. Confident summaries of documents the model never actually read. Fabricated statistics. Subtly wrong legal interpretations. Code that looks correct but does the wrong thing. Translations that swap critical words.
Staff need to see these failure modes, with real examples, and learn to recognise them in the wild. The course should walk through six or eight of the most common patterns with side-by-side comparisons: here is what the AI said, here is what was actually true, here is what the warning sign was.
This is the section that makes the course memorable. Generic training cannot do this well. Custom training built for your context absolutely can.
2. Calibrate verification effort to stakes
The biggest practical problem with verification is that staff treat it as binary: either they check everything or they check nothing. The reality is that verification should scale with stakes.
A brainstorm session with ChatGPT does not need fact-checking. A draft email to a colleague might need a quick sanity check. A client-facing report needs every claim verified. A board paper needs sources triple-checked.
The course should teach this calibration explicitly, with examples drawn from your actual work. The goal is not paranoia. It is proportion.
3. Build the verification habits
What does verification look like, practically? Some of it is technical: cross-checking against authoritative sources, running a second AI tool to see if it gives a different answer, finding the original document the AI was summarising.
Some of it is behavioural: pausing before you paste, reading the output as if you were the recipient, asking yourself “would I stake my reputation on this being correct?” before sending.
The course should teach both. Concrete techniques, not abstract advice.
4. Cover the social dynamics
This is the part most courses miss. Critical thinking about AI is not just a private skill. It is a social one. Pushing back on a colleague who is over-trusting AI is awkward. Telling your boss the report they just sent is partly hallucinated is hard. Asking “are you sure?” can feel like a personal challenge.
Staff who lack this skill stay quiet, and bad output ships. Staff who have it can flag problems without creating conflict. The course should walk through this with real scenarios: how do you raise it? What language works? When do you escalate? When do you let it go? Building scenarios like these is its own design discipline. See our notes on writing engaging scenarios for the underlying approach.
5. Address the over-correction problem
There is also a failure mode in the other direction: staff who become so sceptical of AI that they refuse to use it at all, slowing down work that should be faster. Critical thinking is not anti-AI. It is the skill that makes AI use safe and productive at the same time.
The course should close with this framing. The goal is not to make people distrust AI. The goal is to make them use it with their eyes open.
How long should the course be?
Twenty to thirty minutes for the core course. Some organisations will want a longer version with more practice scenarios, which runs forty minutes.
This is one of the rare categories where the scenarios are most of the value. A short course with five well-built scenarios outperforms a long course with abstract content.
What it should cost
Critical thinking training is Tier 2 work: $7,500 plus GST. The reason is the scenarios. Building branching scenarios that feel realistic, with multiple paths and meaningful consequences, is where this course earns its money. A Tier 1 treatment would work for the awareness layer but would not actually change behaviour.
If you want the longer, role-differentiated version, with scenarios that branch differently for client-facing staff, back-office staff, and managers, that is Tier 3 at $10,500.
When to commission it
After acceptable use training. Before the first quarterly report cycle where AI-assisted work is going to client.
Critical thinking is the layer that catches the mistakes acceptable use training does not prevent. If you only have budget for one, do acceptable use first. If you have budget for two, do both.
If your organisation has already had an “AI got it wrong and we sent it anyway” incident, even a small one, even one that did not leave the building, commission this now.
What to brief your developer on
When you commission a critical thinking course for the AI era, bring:
- Three to five real examples of AI output that turned out to be wrong in your context. Anonymised, but real.
- The AI tools your staff are using most. Different tools fail in different ways.
- The high-stakes outputs in your organisation. Reports, proposals, client deliverables. The course needs to focus there.
- Your existing professional development content. So the new course extends rather than competes.
- Examples of the social dynamics. When have staff struggled to push back on bad AI output? What happened?
The bigger picture
Critical thinking is the connective tissue of the AI capability stack. AI acceptable use training tells people what they can and cannot do. AI training for managers tells managers how to lead through change. AI privacy training tells everyone what the regulator expects. This one tells your whole organisation how to actually catch mistakes before they cost something.
Of the five courses in this series, this is the one I would argue most aggressively for. The acceptable use and privacy courses are about preventing breaches. The manager course is about adoption. The critical thinking course is about competence. It is the one that pays for itself the fastest in mistakes avoided.
When you are ready, get in touch. Three weeks from kickoff to launch. Fixed price. No drama.