How to Improve Critical Thinking Skills at Work in the AI Era

What is critical thinking at work, really?

Definitions of critical thinking can sound abstract. The Foundation for Critical Thinking collects several, including the idea of disciplined thinking that conceptualises, applies, analyses and evaluates information against standards such as clarity and accuracy. At work, it looks much more ordinary than that.

It is the colleague who asks 'compared to what?' when someone says sales are down. It is the manager who notices that a request to 'run a time management course' might really be a problem with duplicated reporting. It is the analyst who checks the three figures in an AI summary that a decision depends on before the summary reaches the board.

In other words, critical thinking at work is the habit of slowing a statement or decision down long enough to see its parts, and then acting at a sensible speed. It is not endless doubt. Doubting everything is just a slower way of being stuck.

Why does critical thinking matter more now that AI is everywhere?

Employers already rated it highly. The World Economic Forum's Future of Jobs Report 2025 found analytical thinking to be the most sought-after core skill, with seven out of ten companies surveyed considering it essential.

Generative AI raises the stakes. Tools such as ChatGPT, Gemini, Claude, Copilot or your company's approved assistant produce clear, confident text in seconds. That fluency is useful, but it also makes errors harder to spot, because easy-to-read statements tend to feel more true. AI models can invent specifics, blend two documents into one, drop the exception that matters or simply agree with the premise of your question.

When answers are cheap, the scarce skill is judging which answers deserve trust. That is exactly what critical thinking trains.

How do you check a claim before acting on it?

Every argument has three layers: the claim, the evidence offered for it and the assumptions that connect the two. Most bad decisions come from the third layer. The evidence is often true; the bridge to the conclusion is made of guesses nobody said out loud.

Try this six-step check on any confident statement you meet this week:

• Restate the claim in one plain sentence. If you cannot, it is too vague to act on.
• List the evidence and label it: data, example, opinion, gut feeling or repetition.
• Write two rival explanations for the same evidence.
• Finish the sentence 'this only works if...' three times to surface assumptions.
• Pick the weakest link: the assumption that would sink the claim if false.
• Decide in advance what evidence would change your mind.

How do you frame a problem before solving it?

Problems often arrive wearing solutions. 'Hire faster', 'buy a new tool' and 'train the managers' are answers, not problems. Before starting work, rewrite the brief as a gap ('we see X, we want Y'), ask 'why does that matter?' twice to find the real goal, and ask 'what's stopping us?' twice to find causes and constraints.

Then write at least three framings that each point to a different kind of solution. A call centre with long wait times can be framed as a hiring problem, a call-handling problem or a problem of customers needing to call at all. The third framing often produces the cheapest, most durable fix.

When a problem keeps returning, dig for the root cause. The 5 Whys, associated with the Toyota Production System, and the fishbone diagram, associated with Kaoru Ishikawa, help you look past symptoms. Stop when you reach something you can change in the system, such as a process, template or policy, rather than a person to blame.

Which biases distort decisions at work, and what can you do about them?

Research by Daniel Kahneman and Amos Tversky from the 1970s onwards showed that people rely on mental shortcuts that produce predictable errors. Knowing the names helps a little. Changing your process helps far more. This table pairs common workplace biases with counter-moves that work.

How do you make a better decision with several options?

First decide how much process the decision deserves. In his 2015 letter to Amazon shareholders, Jeff Bezos distinguished one-way doors, decisions that are consequential and hard to reverse, from two-way doors, which can be changed if they turn out badly. Two-way doors deserve speed. One-way doors deserve care.

For one-way doors with several options, use a weighted decision matrix. Agree five to seven criteria and their weights before anyone argues about options. Describe what a score of 5 and a score of 1 mean for each. Include 'do nothing' as an option. Score independently, mark any dealbreakers, and record your gut feeling separately. If the matrix and the gut disagree strongly, find out why: either a criterion is missing or the gut is reacting to something irrelevant.

Express uncertainty in numbers. 'Likely' means very different things to different people; 'about 60 per cent' starts a real conversation. Start from the base rate (how often do projects like this land on time?) and adjust for the specifics.

Before committing, run a pre-mortem. Psychologist Gary Klein described the technique in Harvard Business Review in 2007: the team imagines the plan has already failed and each person writes down why. Turning failure into a given makes it easier for quiet doubters to speak, and it surfaces risks while they are still cheap to fix.

How do you verify AI output and online sources?

Treat AI like a sharp but unreliable junior colleague. Match the checking to the stakes: brainstorms need little, anything external, legal, financial or decision-driving needs a lot. Extract every specific claim (numbers, names, dates, quotes, citations), verify the important ones against the original documents, and never ask the same AI to confirm its own answer.

For online sources, borrow the habits of professional fact-checkers. A Stanford History Education Group study compared fact-checkers, historians and Stanford undergraduates evaluating websites. The fact-checkers were faster and more accurate because they read laterally: they left the page and opened new tabs to find out who was behind it, instead of judging it by its design.

Mike Caulfield's SIFT routine turns this into four moves you can remember:

• Stop: pause before reading deeply or sharing.
• Investigate the source: who made this, and what is their expertise and agenda?
• Find better coverage: what do trusted sources or expert consensus say?
• Trace claims to the original: follow quotes, figures and images back to where they first appeared.

What does a 20-minute critical thinking routine look like?

You will not always have time for a full workshop. When a significant decision lands and the clock is running, this compressed routine covers the essentials in about twenty minutes. It works alone or with one colleague, and it leaves you with a decision you can explain.

How do you keep improving your judgement over time?

Critical thinking improves fastest when you get feedback, and work rarely provides it. A decision journal fills the gap. For each significant decision, write down what you chose, what you expect to happen in specific terms, your confidence as a percentage and a review date. When the date arrives, record what happened and rate the quality of your process separately from the outcome.

Over a few months, patterns appear. You may find you are overconfident about timelines, or that decisions made while rushed turn out worse. You also avoid the trap of judging decisions only by results: a good decision can turn out badly through bad luck, and a poor one can turn out well by chance.

If you want a baseline before you start, the Critical Thinking & Problem Solving assessment on AssessAll gives you a measured starting point, and Jobulary can turn the results into a development plan you track over time.

To put all of this into practice, the Think Clearly When Machines Think Fast kit from Bodhih Training includes a Decision Toolkit workbook with a weighted matrix, pre-mortem register, assumption log, claim check log and a decision journal that scores your calibration automatically, along with worksheets for problem framing, root-cause analysis and AI output verification.