# How to Build Critical Thinking — and Where It Stops Working > Critical thinking improves with training but does not arrive on its own. A meta-analysis by Abrami and colleagues (2015) pooling 341 effect sizes found an effect of g+ = 0.30 for critical thinking instruction. The effect is largest when principles are taught explicitly and practiced on real material through dialogue. Memorizing categories and terms does not fire at the moment of judgment. - URL: https://brainrank.jp/en/blog/critical-thinking - Published: 2026-08-22 - Publisher: BrainRank Editorial Team (https://brainrank.jp) - Tags: Psychology, Cognitive abilities Critical thinking improves with training. Sitting through a lecture and memorizing terms, however, does not improve it. What works is teaching the principles explicitly, practicing them on real material, and arguing them out with other people. This article covers how large the effect is, a procedure you can use, and why knowing about bias is not enough. ## What critical thinking is Critical thinking is defined as reasonable, reflective thinking focused on deciding what to believe and what to do. It is not the posture of rejecting what someone says. In practice it means separating three things: the claim (the conclusion), the evidence (the facts supporting it), and the assumptions (conditions that must hold but are left unstated). Left mixed together, a claim with support and a claim without it become indistinguishable. ## How large is the training effect The reference point is the meta-analysis by [Abrami and colleagues (2015, Review of Educational Research)](https://journals.sagepub.com/doi/10.3102/0034654314551063). Pooling 341 effect sizes, instruction in critical thinking produced an effect of g+ = 0.30 (p < .001). An effect of 0.30 is small by statistical convention but substantial for an educational intervention. The same analysis showed that conditions matter. Effects were larger where instruction combined the following. - Principles of reasoning taught explicitly rather than left embedded in subject content - Practice on authentic material and concrete problems - Dialogue and discussion built into the work - Instructors themselves trained in teaching critical thinking Put the other way: designs that leave the principles implicit and assume thinking will develop from exposure tend not to produce much. ## The procedure: separate claim, evidence, assumption The smallest workable version has four steps. 1. State the claim in one sentence: write what is being asserted, without evaluating it 2. List the evidence: what facts support it. For figures, include the source and how it was measured 3. Dig out the assumptions: what must silently be true for this claim to hold 4. Look for disconfirmation: if your conclusion were wrong, why. Write at least three reasons Step four is the one most often skipped and the one that does the most work. People preferentially gather information supporting what they already think (confirmation bias), so unless the search for counter-evidence sits outside you as a procedure, it does not happen. These four steps apply the checking phase of [metacognition](https://brainrank.jp/en/blog/metacognition) to the moment of judgment. Both share the structure of treating your own thinking as the object. ## Why knowledge alone does not fire Memorizing a taxonomy of biases barely operates when a judgment is actually being made, because bias operates where judgment is fast and automatic. For knowledge to interrupt, something has to stop you — and rather than expecting that signal from inside, it is more realistic to place it outside as a procedure (see [cognitive biases and how to counter them](https://brainrank.jp/en/blog/cognitive-bias)). The harder problem is that judging the quality of your own judgment is itself difficult: evaluating performance in a domain requires competence in that domain (see [the Dunning-Kruger effect](https://brainrank.jp/en/blog/dunning-kruger-iq)). That is exactly why dialogue with others appears among the conditions for effective training. ## Critical thinking in the age of AI Generative AI delivers its output as a finished claim with the evidence and assumptions stripped out. The more readable it is, the easier it becomes to skip all four steps above. What helps is reversing the order: write your own answer and your reasons first, then set the AI's response against it. Done that way, the output becomes something to verify rather than a conclusion to adopt. We cover this in [AI and cognitive offloading](https://brainrank.jp/en/blog/ai-and-cognitive-offloading), and how to read figures about AI capability in [what is an AI's IQ](https://brainrank.jp/en/blog/ai-iq). ## Measure the reasoning underneath Critical thinking is a procedure, but the reasoning that runs it sits underneath. [BrainRank's free IQ test](https://brainrank.jp/en/start) scores 20 questions across spatial reasoning, pattern recognition, logical reasoning, and classification (about 10 minutes) using item response theory (IRT) and returns domain tendencies. The logical reasoning result is useful material for thinking about which of the steps above you are most likely to stumble on. Results are estimates for entertainment and self-understanding, not a medical or diagnostic assessment. ## Frequently asked questions ### What is critical thinking? Reasonable, reflective thinking focused on deciding what to believe and what to do. It is not criticizing other people's views; it means separating claim, evidence, and assumption, and holding your own judgments to the same standard. ### Can critical thinking be trained? Yes. A meta-analysis pooling 341 effect sizes found instruction produced an effect of g+ = 0.30. That is on the small side but real, and it grows when principles are taught explicitly and practiced on authentic material. ### Do smart people automatically think critically? They are different capacities. Strong reasoning still produces biased judgments unless you actually run the step of looking for evidence against your conclusion. People skilled at argument can be especially good at assembling support for the position they already hold. ## Sources - [Abrami and colleagues (2015, Review of Educational Research)](https://journals.sagepub.com/doi/10.3102/0034654314551063) --- Statistics and percentages are calculated from a normal distribution model with a mean of 100 and a standard deviation of 15. The tests on this site provide estimates for entertainment and self-understanding; they are not medical or clinical assessments, nor official intelligence tests. When citing, credit "BrainRank (brainrank.jp)" and link the article URL.