A student can memorise the scientific method and still treat a confident claim as stronger than a careful one. Another can recite that science uses evidence and then accept a viral graph because it has numbers on it. A third can hear that scientific conclusions change and infer that science is unreliable because “they keep changing their minds.”
These are not merely failures of information. They are failures in beliefs about knowledge itself.
Every learner carries assumptions about what counts as knowing. Some of those assumptions are explicit: “I trust it because the teacher said it.” Others are almost invisible: “If two people disagree, the truth must be somewhere in the middle.” “If a result is scientific, it should be certain.” “If science changes, the earlier knowledge was fake.” “Personal experience is more real than a study.” “A published paper proves the claim.” “If I can think of an exception, the general pattern is false.”
These assumptions shape how students read evidence, interpret uncertainty, respond to correction and decide whether a claim deserves belief.
Education researchers often describe this territory using terms such as epistemic beliefs, epistemic cognition or beliefs about the nature of knowledge. The language can sound abstract. The classroom job is concrete: help learners understand that trustworthy knowledge is not simply a pile of statements delivered by authority. It is produced through methods, evidence, reasoning, criticism, comparison and revision.
OECD’s PISA 2025 reporting makes this especially timely. Its analysis of student attitudes includes beliefs about the nature of science and reports that science performance tends to be stronger where students more strongly recognise scientific knowledge as evidence-based, testable and open to scrutiny. It also highlights practical habits such as checking a claim from one source against other reliable sources. As always, the associations do not prove a simple causal pathway, and the measures are based on self-report. But the educational signal is important: what students believe about knowledge can affect how they use knowledge.
The 50-second answer
Epistemic beliefs work like a hidden rulebook for deciding what deserves trust. A student who believes knowledge is mainly fixed information from authority will study, argue and check differently from a student who believes claims should be supported by evidence, tested against alternatives and revised when stronger evidence appears.
In science, four ideas matter especially:
Claims need evidence. Evidence must be relevant to the claim. Methods affect how much the evidence can support. Conclusions can be strong without being infallible. New evidence can revise knowledge without making all knowledge arbitrary.
The educational goal is not to make students suspicious of everything. It is to make their trust better calibrated.
Begin with an ordinary claim: “But I saw it myself”
Imagine a student says, “Drinking this drink definitely improves concentration. I used it before a test and I scored higher.”
The experience is real. The conclusion may still be wrong.
Perhaps the student slept more. Perhaps the second test was easier. Perhaps revision was better. Perhaps the drink had caffeine and had a short-term alerting effect. Perhaps expectation changed effort. Perhaps the score difference was random. One observation can begin an inquiry; it rarely settles a general causal claim.
A learner with weak epistemic habits may hear this as an insult to experience: “So you are saying what happened to me did not happen?”
A learner with stronger epistemic habits can separate observation from inference. “The score improved” is an observation. “The drink caused the improvement” is an explanation. The explanation requires more support than the observation.
That distinction is foundational. It tells the student that evidence is not just “something that happened.” Evidence becomes evidence for a claim through a relationship between observation, method and inference.
The same distinction appears in school science, history, English argument, mathematics proof, media literacy and ordinary family decisions. What changes across domains is what counts as strong evidence and how it is evaluated.
Stage 1: students need to distinguish a claim from the evidence for the claim
Many reasoning problems begin because claim and evidence collapse into each other.
“Plants grow better with this fertiliser.” That is a claim.
“These treated plants were taller after four weeks than the untreated plants under the same measured conditions.” That is evidence relevant to the claim.
“This fertiliser contains nitrogen.” That may be background information or a plausible mechanism, but by itself it does not show that this product caused better growth under the tested conditions.
A classroom can teach this distinction explicitly. Ask students to mark the claim, the observation and the reasoning that connects them. Then change one part. What if the treated plants also received more light? What if the sample was two plants? What if only the largest plants were reported? What if the difference was tiny? What if another group could not reproduce it?
The learner begins to see that “evidence” is not a magic word. Evidence has quality, relevance, scope and uncertainty.
This is also why good critical thinking is more than disagreement. A student who says “I don’t believe it” has not yet reasoned well. Skepticism becomes useful when it can specify what evidence is missing, what alternative explanation remains possible, or what test would distinguish competing claims.
Stage 2: evidence is only meaningful relative to a question
The same observation can be strong evidence for one claim and weak evidence for another.
Suppose a school tests a reading programme and finds that students report enjoying the lessons. That is evidence about enjoyment. It is not automatically evidence about reading comprehension. If comprehension scores improve, that is evidence about performance, but the strength of the causal conclusion depends on the design. If the programme also increased teaching time, reduced class size and changed assessment, the programme label may hide several active ingredients.
Students should therefore learn to ask: Evidence for what, exactly?
This question is deceptively powerful. It prevents the common move in which any data connected to a topic are treated as support for every claim about that topic.
A graph can show that two variables move together. That is evidence of an association. It may not establish cause. A survey can show what respondents report. It may not show what they actually do. An experiment can support a causal conclusion under specified conditions. It may not show that the effect generalises to every population or setting.
When learners understand this, they stop treating the word “study” as a seal of certainty. They begin to inspect the job the study can actually do.
Stage 3: methods change what evidence can tell us
Scientific knowledge is powerful partly because methods are designed to reduce particular kinds of error.
Controls help distinguish an intervention from background change. Replication tests whether a result survives repetition. Blinding can reduce expectation effects. Random assignment can reduce systematic differences between comparison groups. Measurement standards reduce ambiguity about what was observed. Statistical methods help estimate uncertainty. Peer criticism can expose weaknesses that the original researchers missed.
Students do not need advanced methodology before they can understand the principle: how you look affects what you are entitled to conclude.
A child comparing two paper aeroplane designs can understand the need to keep launch conditions similar. A secondary student can understand why asking only satisfied users gives a distorted view of a service. A science class can understand why measuring temperature with different instruments without calibration makes comparison difficult.
This is epistemic education at ground level. The lesson is not merely “follow the procedure.” It is “the procedure protects the claim from specific alternative explanations.”
When students see that relationship, method becomes meaningful rather than ceremonial.
Stage 4: scientific knowledge can be reliable without being untouchable
One of the hardest ideas in science education is that knowledge can be both well supported and revisable.
Students often receive mixed cultural messages. School textbooks present polished conclusions. News reports announce that “scientists have discovered” something. Later, recommendations change. A new study challenges an old result. A public-health guideline is updated. Learners can interpret this as contradiction: if science was trustworthy, why did it change?
The answer is that revision is not a defect added to science from outside. Revision is part of the mechanism that makes science trustworthy.
A claim earns confidence because it has survived evidence, testing, criticism and comparison with alternatives. That confidence can be strong. It is not metaphysical certainty. If better evidence changes the balance, the responsible action is to update the conclusion.
This does not mean every claim is equally uncertain. “Scientific knowledge is revisable” should not be turned into “nothing is really known.” The orbit of Earth is not on the same evidential footing as an early-stage claim from a tiny exploratory study. Confidence should track the quality and convergence of evidence.
A good epistemic education therefore teaches graded confidence rather than certainty versus doubt.
Stage 5: one source is rarely enough when the claim matters
OECD’s current PISA analysis highlights a practical belief: information from one source should be checked against other reliable sources before it is accepted, especially when the claim is important or uncertain.
This sounds like ordinary fact checking. The deeper mechanism is source independence and evidence convergence.
Ten websites repeating the same press release are not ten independent confirmations. A video that cites an article that cites another article that cites the same original claim has created repetition, not replication.
Students need to learn to follow evidence chains.
Who is making the claim? What is their source? Is the source primary, secondary or merely repeated? Does another independent source report compatible evidence? Are there authoritative organisations with relevant expertise? Is the claim recent enough that updated evidence matters? What would count as a serious disagreement?
The point is not to force students into endless research for every ordinary statement. Verification should be proportional to stakes, novelty and uncertainty. A homework definition can be checked quickly. A health claim, financial claim or sensational scientific claim deserves more scrutiny.
The learner is building an internal threshold for “enough checking.”
Stage 6: common sense is useful — and not a court of final appeal
PISA 2025 also asks students about relying on common sense versus scientific studies. This matters because common sense has a complicated educational role.
Common sense is not worthless. It contains accumulated experience, rough causal models and practical expectations. Scientists use intuition to generate hypotheses. Engineers use plausibility checks. Students should absolutely ask whether an answer makes sense.
The error occurs when common sense becomes immune to evidence.
Many scientific findings are unintuitive precisely because human intuition was built for ordinary scales and experiences. Randomness feels patterned. Exponential growth is routinely underestimated. Objects at astronomical, microscopic or quantum scales do not behave like familiar household objects. Medical outcomes can be distorted by regression to the mean, placebo effects and selection biases that ordinary experience does not automatically correct.
A strong epistemic stance is therefore: use intuition to generate questions and check plausibility; use evidence and method to adjudicate claims when intuition and systematic evidence diverge.
That is more sophisticated than either “trust your gut” or “ignore common sense.”
A worked example: “This revision method works for me”
A student says that rereading works because it makes the chapter feel familiar. Another student says retrieval practice works because a teacher recommended it. A third says flashcards are useless because she once used them and did badly.
How should a learner reason?
First separate personal preference, immediate experience and learning outcome. Familiarity may measure ease of processing, not later recall. A teacher’s recommendation is a reason to investigate, not proof. One failed attempt can reflect poor implementation, unsuitable material or insufficient practice.
Then ask what outcome matters. If the goal is long-term recall, the useful evidence is delayed performance, not how fluent the material felt during study. If the goal is understanding a complex argument, a different set of methods may be needed.
Then consider converging evidence: classroom experience, credible summaries of research, the learner’s own carefully observed results, and the fit between the method and task.
The conclusion need not be absolute. “For factual retrieval, active recall appears more reliable than repeated rereading, but I still need to understand the material and use the method well.”
That sentence is epistemically stronger than “This works for me” because it distinguishes evidence, outcome and boundary.
A worked example: the dramatic graph
A student sees a graph online showing two lines rising together: smartphone use and anxiety among teenagers. The caption says phones cause anxiety.
The graph may be important. It does not by itself establish the causal claim.
The learner can ask what each variable measures, whether the time periods match, how the sample was selected, whether other factors changed simultaneously, and whether studies using different designs reach similar conclusions. They can also ask whether the graph is even sourced accurately.
None of this requires the learner to deny that smartphones might affect anxiety. Good epistemic reasoning is not a ritual of saying “correlation is not causation” and walking away. It is a route toward a better causal judgement.
The learner might conclude: “This graph supports the possibility of a relationship. It does not isolate the causal effect. I need stronger designs and converging evidence before claiming causation.”
That is what calibrated belief sounds like.
A worked example: when the textbook is wrong or incomplete
Students are often taught to trust textbooks, and usually that trust is reasonable. Textbooks are curated educational resources. The problem is turning reasonable trust into absolute authority.
Suppose a textbook simplifies a scientific model for the learner’s level. Later, a teacher introduces exceptions or a more advanced model. A student may feel betrayed: “So the old thing was wrong.”
A better epistemic frame is that models have scope. Early models can be useful approximations for certain questions. Later models add variables, mechanisms or boundary conditions.
This matters beyond science. In mathematics, a rule that works in one domain may need qualification in another. In grammar, a school rule may be a teaching simplification rather than a universal description of language. In economics, a model can illuminate one mechanism while holding others constant.
Students who understand scope are less likely to interpret complexity as inconsistency.
Epistemic beliefs are not the same as fact checking
Fact checking asks whether a specific claim is accurate. Epistemic beliefs shape how the learner thinks accuracy should be established in the first place.
A student can follow a fact-checking checklist mechanically while still holding poor assumptions. They may believe that an official-looking site is automatically true, that two sources always beat one, that experts never disagree, or that disagreement means there is no knowledge.
Epistemic education asks deeper questions: What makes a source relevant? Why does method matter? How should uncertainty be represented? When is authority a useful proxy? What kind of new evidence would justify revision? How do independent lines of evidence change confidence?
These beliefs influence fact checking, but they are not reducible to it.
Epistemic beliefs are not the same as epistemic emotions
Curiosity, surprise and confusion can trigger inquiry. They are emotional responses to knowledge gaps, contradictions or novelty. Epistemic beliefs are the assumptions that guide what the learner does with those feelings.
Confusion can lead to productive investigation if the learner believes contradictions deserve examination. It can lead to disengagement if the learner believes a correct subject should never feel uncertain. Surprise can prompt revision if the learner sees models as testable. It can be dismissed if the learner treats prior belief as untouchable.
For the neighbouring emotional mechanism, see How Epistemic Emotions Work | Curiosity, Surprise, Confusion and Boredom Change the Learning Route.
Epistemic beliefs are not the same as epistemic framing
Epistemic framing concerns what kind of knowledge work a learner thinks a situation requires. Is this a task of recall, explanation, proof, interpretation, design, critique or evidence evaluation?
Epistemic beliefs operate at a more general level: what counts as a justified claim, whether knowledge can change, how authority and evidence relate, and what role testing should play.
The two mechanisms interact. A student may know that science depends on evidence but misframe an exam question as a recall task when it actually asks for evaluation. Or a student may correctly identify that a task requires source comparison while still believing that the most authoritative source should automatically win.
For that neighbouring owner, see How Epistemic Framing Works | Know What Kind of Knowledge Work the Situation Requires.
Why this matters in the age of generative AI
Generative AI makes epistemic beliefs more important, not less.
A fluent answer can arrive before the learner has decided what would make the answer trustworthy. The surface qualities that humans often use as shortcuts — grammar, confidence, detail, speed, apparent synthesis — can be generated without a corresponding guarantee of truth.
Students therefore need to separate presentation quality from evidential quality.
Who or what is the source of the factual claim? Can the claim be checked against primary or authoritative material? Does the answer distinguish established knowledge from uncertainty? Are citations real and relevant? Is the explanation internally coherent but empirically unsupported? Does the system have access to current information where recency matters?
The important habit is not “AI is unreliable.” That belief is too crude. The habit is: the method of verification must fit the stakes and the claim.
A low-stakes brainstorming suggestion may need little checking. A health, legal, financial or scientific factual claim needs much more. The epistemic standard travels with consequence.
The common failure modes
The first failure mode is authority absolutism: a claim is true because a teacher, textbook, expert, institution or prestigious journal said it. Authority can be useful evidence about credibility, but it is not infallibility.
The second is authority nihilism: because experts can be wrong, expertise does not matter. This throws away one of society’s most valuable signals of accumulated knowledge.
The third is one-study certainty: a single published result is treated as the final word rather than one contribution to an evidence base.
The fourth is all-opinions equality: disagreement is treated as proof that every position deserves equal weight regardless of evidence.
The fifth is common-sense supremacy: personal intuition overrules systematic evidence whenever the evidence feels strange.
The sixth is method blindness: data are treated as evidence without asking how the data were produced.
The seventh is revision cynicism: updates in scientific understanding are interpreted as proof that science cannot be trusted.
The eighth is source-counting: several repeated versions of one claim are mistaken for independent confirmation.
The ninth is uncertainty inflation: because perfect certainty is impossible, the learner concludes that no conclusion is justified.
The tenth is skepticism as performance: the student learns to challenge every claim theatrically but never states what evidence would change their own mind.
Strong epistemic reasoning avoids both gullibility and reflexive cynicism.
A practical route for learners
When you encounter an important claim, use five questions.
First: What exactly is the claim? Make it specific enough to test.
Second: What is being offered as evidence? Separate observations, data, authority, mechanism, anecdote and opinion.
Third: How was the evidence produced? Ask whether the method is capable of supporting the conclusion.
Fourth: What alternative explanation remains? A strong claim should survive plausible alternatives, not merely sound convincing by itself.
Fifth: What would make me revise my belief? If the honest answer is “nothing,” the process is no longer evidence-led.
These questions do not need to be used equally for every claim. Use more scrutiny when the stakes are high, the claim is surprising, the source is weak, or the evidence conflicts with established knowledge.
A practical route for parents
Parents can support epistemic development without turning dinner into a debate club.
When a child makes a strong claim, ask, “What makes you think that?” Then listen to the kind of evidence offered. If the answer is “My friend said,” the next step is not necessarily “Your friend is wrong.” Ask, “What could we check?”
When parents themselves are uncertain, model calibrated confidence. “I think this is probably right because these two reliable sources agree, but I have not checked the original study.” That is more educational than pretending certainty.
When recommendations change, explain revision rather than accusing experts of inconsistency. “The advice changed because the evidence changed or the balance of evidence became clearer.”
Most importantly, let children see adults update beliefs without humiliation. If changing one’s mind is treated as losing, evidence will struggle to compete with identity.
A practical route for teachers
Teachers can build epistemic beliefs through ordinary curriculum rather than separate lectures about “the nature of science.”
Show students a claim and ask what evidence would discriminate between two explanations. Compare a strong and weak study design. Present a historical scientific model and ask what evidence forced revision. Give two sources that agree for different reasons and ask whether they are independent. Use error bars, uncertainty ranges or model limitations where appropriate instead of presenting every result as exact.
When correcting misconceptions, explain why the better account deserves confidence. “Because the textbook says so” may fix the answer and leave the epistemic model unchanged.
Ask students to distinguish confidence levels: possible, plausible, supported, strongly supported, established under specified conditions. This vocabulary helps them move beyond true/false thinking.
In argument writing, require students to explain not only what their evidence says but why it is probative. In science practical work, connect procedural controls to the alternative explanation each control blocks. In mathematics, distinguish example from proof. Across subjects, show that different domains have different evidential standards.
What schools should avoid measuring too simply
Because epistemic beliefs can be surveyed, schools may be tempted to create a single score for “scientific thinking.” That is risky.
Students may endorse desirable statements on a questionnaire because they know what school expects. A learner can agree that claims should be checked and still fail to check an emotionally attractive claim in real life. Cultural differences can affect how survey items are interpreted. Beliefs may vary by domain: a student can be sophisticated about evidence in science and naïve about evidence in social media.
Better classroom evidence includes decisions. Does the learner ask what claim the data support? Do they distinguish correlation from causation when relevant? Can they state limitations? Do they revise a conclusion when counterevidence appears? Can they compare source independence? Do they avoid treating uncertainty as ignorance?
The goal is not an epistemic label. It is better reasoning under real conditions.
Evidence and caveats
OECD’s PISA 2025 analysis reports that students who more strongly recognise scientific knowledge as evidence-based, testable and open to scrutiny tend to perform better in science across participating systems. The report also finds positive relationships between science performance and rejecting the idea that common sense should simply replace scientific studies. It discusses cross-source checking as part of more discerning information evaluation.
These findings should be interpreted carefully. PISA measures student beliefs largely through self-report. Students with stronger science knowledge may have learned more sophisticated beliefs because of successful science education; sophisticated beliefs may support learning; high-quality schools may produce both; and other factors may influence the relationship. The data do not justify a claim that teaching one slogan about evidence will directly raise scores.
The stronger educational case is conceptual and practical. Science requires learners to understand why evidence, testing, criticism and revision matter. These ideas help students interpret experiments, models, uncertainty, changing guidance and competing claims. PISA 2025 adds current international evidence that these beliefs are meaningfully related to science performance and student agency.
The ordinary-weekday test
It is a weekday evening. A student sends a family group chat a video claiming that a common food “destroys memory.” The video has a confident narrator, animated brain graphics and three screenshots of scientific papers.
The family has several possible responses.
“Fake news.” Too fast.
“Sounds scientific.” Too trusting.
“Everything is bad for you nowadays.” Cynical.
A stronger route is ordinary and teachable. What is the exact claim? Are the papers real? Do they study the same food, population and outcome? Are the screenshots reporting laboratory mechanisms, observational associations or controlled human outcomes? Do independent reviews reach the same conclusion? How large is the effect? What are the limitations?
No one needs to become a professional scientist at the dining table. The family only needs a better rule for moving from presentation to belief.
That is the practical value of epistemic education. It changes the pause between “I saw this” and “I believe this.”
FAQs
What are epistemic beliefs in simple language?
They are beliefs about knowledge: what counts as knowing, where knowledge comes from, how claims are justified, how certain conclusions can be, and what should make us change our minds.
Are scientific facts always temporary?
No. “Revisable” does not mean equally unstable. Some findings are supported by many converging lines of evidence and are extremely robust. Revision usually refines scope, mechanism or precision rather than resetting all knowledge to zero.
Should students distrust authority?
No. Expertise and reputable institutions are useful credibility signals, especially when learners cannot inspect all primary evidence themselves. The goal is calibrated trust: authority matters, but claims remain connected to evidence, methods and accountability.
Is personal experience evidence?
Yes, but usually limited evidence. It can establish that something occurred to a person. It is often weak for general causal claims because many uncontrolled factors can influence the outcome.
Is checking two sources enough?
Not necessarily. Two sources may repeat the same original material. The important question is whether the evidence is independent, relevant and credible, not merely whether the page count reached two.
Does science change because it is unreliable?
Science changes partly because it has mechanisms for detecting error and incorporating new evidence. A system that never changes despite contrary evidence would be less trustworthy, not more.
How does this connect to critical thinking?
Epistemic beliefs provide some of the rules critical thinking relies on. They help determine what counts as a reason, how evidence should affect confidence, and when revision is justified.
What is one useful classroom question?
“What evidence would make this claim stronger, and what evidence would make you revise it?” That question connects justification and openness to change.
The final idea
Students leave school carrying more than facts. They carry a model of how facts become worthy of belief.
If that model is weak, knowledge can become brittle. The student memorises correct statements but is vulnerable when sources conflict, evidence changes, authority disagrees or a polished falsehood arrives first. If the model is stronger, the learner has a way to navigate uncertainty without collapsing into either gullibility or cynicism.
The goal is not permanent doubt. It is earned confidence.
A claim becomes more trustworthy because relevant evidence supports it, the method can bear the inference, alternative explanations have been tested, independent evidence converges, and the conclusion remains open to revision if something better appears.
That is not a footnote to science education. It is part of what science education is for.
Surgical internal links
- How Epistemic Emotions Work | Curiosity, Surprise, Confusion and Boredom Change the Learning Route
- How Epistemic Framing Works | Know What Kind of Knowledge Work the Situation Requires
- How Studying Works | Belief-Consistency Validation — Why a Claim Can Feel True Before You Check the Evidence
- How Transfer of Learning Works | When Knowledge Survives a New Situation
Sources
- OECD. PISA 2025 Results (Volume I): Student Attitudes Towards Learning. 2026. https://www.oecd.org/en/publications/pisa-2025-results-volume-i_73451bc5-en/full-report/student-attitudes-towards-learning_a3cc05a7.html
- OECD. PISA 2025 Results (Volume I): Singapore Country Note. 2026. https://www.oecd.org/en/publications/pisa-2025-results-volume-i-country-notes_2d4ff9ea-en/singapore_80497e2f-en.html
- Education Endowment Foundation. Metacognition and Self-Regulated Learning. Second edition, 2025. https://educationendowmentfoundation.org.uk/education-evidence/guidance-reports/metacognition
- Institute of Education Sciences / What Works Clearinghouse. Organizing Instruction and Study to Improve Student Learning. https://ies.ed.gov/ncee/wwc/practiceguide/1