Say someone tells you a task you agreed on is done. You ask to see the result, and it meets what was agreed. Then another doubt comes along: perhaps there’s something you haven’t thought to check.
A new worry can deserve attention. But what would let you decide the task is done? Each answer may bring a new demand that has nothing to do with the last. Then your doubt may have no way to end.
You don’t have to choose between believing everything and rejecting everything. You can say what would support an answer, what would go against it, and what would leave the question open.
This module is for working adults judging factual claims that can be checked, when the evidence is uncertain.
What you’ll get from this module
In this module
By the end of this module, you’ll have a clearer way to decide what evidence could change your mind. We’ll link research on training against false news, feedback and corrections with professional guidance on setting rules for a decision. You’ll think about what those findings mean and choose a situation from daily life. Then you’ll write a standard for accepting a claim, rejecting it or leaving it open.
Ask Jing
Become a Jingius.
What makes doubt useful?
Controlled experiments: true information can lose belief too.
Megan E. Graham and colleagues’ 2023 study asked whether a game about false news helped people tell true material from false. Four teams of undergraduates compared the Bad News game with Tetris at the University of Victoria.
The researchers studied 353 people after some were left out. People rated made-up Tweets before and after playing a game, with 24 false items and 8 true items in each team’s set.
Playing Bad News lowered how much people believed false Tweets. It also lowered belief in true Tweets by the same amount, or almost. An extra analysis, not planned in advance, found no better ability to tell the two apart.
Both kinds of judgement matter. If you look only at belief in false claims falling, the result looks like a success. Including true claims shows what that description leaves out.
Why it matters: These were quick rating tasks done by students straight after the game. There were more false items than true ones, and the teams’ methods differed. True news lost belief in these tasks. That doesn’t show people became unable to believe anything in daily life.
Read the experiments comparing Bad News with Tetris.
A review of many experiments: training can help people spot manipulation.
The wider evidence includes a different result. In a 2026 meta-analysis, Almog Simchon and colleagues combined data from 33 tests run by their wider research group. These tested games and videos that teach people about manipulation.
People got better at telling manipulative material from believable material. The analysis found no clear overall shift towards believing everything or doubting everything. The tasks asked whether material was reliable or manipulative. That’s different from checking whether a factual claim is true.
Why it matters: The studies used different training, materials and analyses. Their combined finding doesn’t cancel out the drop in belief in true items in the student tests. To know whether training helps, we need to know what people got better at judging.
Read the meta-analysis abstract on inoculation and discrimination. Read the full meta-analysis of manipulation recognition.
Online studies: spotting persuasion and judging truth are separate tasks.
Gordon Pennycook and colleagues’ 2024 paper reports five studies with 7,286 people online. The researchers used true and false news headlines and changed whether they used emotional manipulation.
A video helped people spot emotional manipulation, but didn’t improve their ability to tell true claims from false ones. Pairing the video with a reminder to think about accuracy did help. People then believed true content more. Those combined tests paired emotional false headlines with neutral true headlines; they didn’t test every mix of emotion and truth.
Spotting how a message tries to sway you answers a question about how it’s presented. Judging whether its factual claim is right answers a different question. In these studies, getting better at the first alone didn’t help with the second.
Why it matters: The combined result shows why the difference matters. Spotting emotional wording helped with the truth task when it came with attention to accuracy, under the conditions tested. Emotional wording on its own didn’t settle whether a claim was true.
Read the study abstract on inoculation and accuracy prompts.
Experiments on feedback: knowing your mistakes is different from knowing your rank.
Johannes Leder and colleagues’ 2024 research ran two online tests that added feedback after Bad News. People were placed in groups at random. The feedback was about people’s judgements of single items, including whether material was real or misleading.
Compared with playing without that feedback, the feedback groups did better on the task later, including a week after playing. Every type of feedback improved ratings of real news, but not every type clearly improved ratings of misleading items. People were judging how reliable news was, so the finding is about performance in those training tasks.
Knowing which judgement was wrong gives the learner something specific to think again about. Rejecting a true item matters here as much as accepting a misleading one.
Read the experiments on item-level feedback.
A different test shows why the kind of feedback matters. Benjamin Lyons and colleagues’ 2025 U.S. study gave people, at random, feedback on how they did compared with others on a task judging news headlines.
Why it matters: That feedback made people feel less sure of themselves compared with others. But the researchers found they were no better at telling true news from false afterwards. It told people where they stood, rather than explaining their judgements on single items. Feeling less confident and judging more accurately were different results.
Read the experiment on relative-performance feedback.
Experiments on corrections and a review of research: new evidence can help.
Thomas Wood and Ethan Porter’s paper, published online in 2018 and in a 2019 issue, tested factual corrections across five tests. More than 10,100 people online saw claims across 52 issues, with factual corrections given out at random.
The researchers found no backfire caused by corrections across those tests. Backfire here means ending up less accurate after a false claim and its correction than after the false claim alone.
That finding gives a reason to keep a way open for updating. An old belief can hang on even after a correction helps. That’s different from becoming less accurate because of the correction.
Read the experiments on factual backfire.
Ullrich K. H. Ecker and colleagues’ 2022 review adds that corrections usually cut how much people rely on false news. They often don’t remove it. Detailed factual explanations usually work better than just saying the earlier claim was wrong.
Why it matters: An explanation gives people something to put in place of the mistaken account. The old account can still affect their reasoning, but that doesn’t mean the correction didn’t help. Neither result means every correction works.
Read the review of misinformation and correction.
Professional guidance: decide what would count before you read a result.
The NIST/SEMATECH statistics handbook describes how to test competing claims using rules set before you look at the results. It separates two mistakes: rejecting a claim when it’s true, and failing to reject it when it’s false.
This guidance is about statistical tests. For everyday judgements, it offers a useful rule. Say what evidence would support your decision, and what kind of mistake you could make.
Why it matters: A stricter rule can lower one risk while leaving another. That helps explain why “believe less” is incomplete advice: rejecting true news can have costs too.
Read the handbook explanation of statistical tests.
What this means for you
Aim for a judgement that responds to evidence. You can be careful about a claim with no support. You can still leave a clear way to accept it if good evidence arrives.
Keep both kinds of error in view
Ask what believing a false claim would cost and what rejecting a true claim would cost. They needn’t be equally serious. The costs help you choose how much evidence to ask for.
In the made-up task example, accepting unfinished work could leave a job undone. Rejecting finished work could mean insisting that someone still owes work they’ve already done. You need to look at the promised result, not count doubt itself as a success.
This is also how to read feedback. When a judgement turns out wrong, find the particular evidence you missed or misread. Include true claims you rejected, not only false claims you accepted.
A general feeling that you’re unusually doubting, trusting or unsure doesn’t tell you which judgement needs fixing. Specific evidence gives you a reason to change that judgement.
Separate the claim from its presentation
When a message uses emotional language, keep the persuasion technique apart from the factual statement. You might rightly spot an attempt to sway you and still have an open question about what happened.
Put the claim plainly enough that evidence could support it or go against it. Then ask what the evidence you have says about that claim. Spotting persuasion becomes a reason to look at the claim, not a rule to reject it automatically.
Set a standard that evidence could meet
Before you judge, name evidence you could really get that would support the claim. Also name what would go against it and what missing information would leave you undecided.
A standard that can be met still needs to fit the claim and the cost of being wrong. Asking to see a promised result fits a claim that a task is done. Asking for proof that deception is impossible leaves no practical way to an answer.
Say the claim is that a task you agreed on is done. You could accept it for now when you can look at the promised result and it meets what was agreed. That means accepting it for the moment, while staying open to relevant new evidence. A defect that can be shown would change that answer. If you can’t look at the result, stay undecided.
Looking at the result lets you compare it with what was agreed. If you can’t look at it, the comparison isn’t finished. Finding a defect is different: you now have evidence that it isn’t done.
| What you find in the task example | What it supports |
|---|---|
| The accessible result meets the agreed requirements | Accept completion provisionally, while remaining open to a demonstrated defect. |
| A demonstrated defect violates a requirement | Reject the completion claim on that basis and identify the defect. |
| You cannot inspect the result | Leave completion unresolved; missing evidence hasn’t shown the claim false. |
These conditions give your doubt a possible end without closing the door to a correction.
Update the judgment, or explain why the standard changed
When new evidence arrives, compare it with the conditions you set. If it meets them, change your judgement. If it shows that the conditions weren’t good enough, explain the concrete reason they need changing.
New evidence or higher stakes can justify a stricter standard. You may dislike a result that otherwise counts. That is a reason to look at your own reaction, not a reason on its own to ask for more.
Ask whether you would ask for the same evidence if you liked the answer. Use that question to check your own consistency, not to guess at someone else’s motives.
If relevant information is still missing, keep the answer open and name the gap. An open answer can be more accurate than either accepting too soon or rejecting without support.
How about your daily life?
Where do you need an answer but keep finding another reason to doubt it? You might be checking whether a task you agreed on is done. Or you might be thinking again about a claim after a correction.
Write one or more situations you’d like to handle more clearly. Choose a claim that can be checked, rather than a judgement about whether a person can be trusted overall.
| Situation or claim | What I want to understand |
|---|---|
| ____________________ | ____________________ |
| ____________________ | ____________________ |
Which situation would be most useful to explore now?
My chosen situation: ____________________
You can change your choice. Carry this situation into the exercise; choosing it doesn’t show whether the claim is true.
Write a workable belief standard
Start with your chosen situation. If you’ve come straight here, choose an ordinary claim that can be checked but where the evidence is unclear. One example is whether a task you agreed on is done.
Use this exercise to make your judgement clearer. It is not for judging another person, or for setting one level of trust for everything. Leave the question open when the relevant evidence isn’t available.
- State the claim and the consequences. Write exactly what would need to be true. Describe what could happen if you believed the claim when it was false, or rejected it when it was true.
Claim and consequences: ____________________ - Write your conditions before deciding. Name evidence you could really get that would support the claim. Then name what would go against it and what missing information would leave it open. Make the conditions fit the claim and the costs of being wrong.
Accept provisionally if: ____________________
Reject if: ____________________
Remain undecided if: ____________________ - Compare the available evidence with those conditions. Write down what the evidence actually supports. Accept for now when your condition for accepting is met. That means you stay open to good new evidence. Reject when the evidence goes against the claim. Otherwise, stay undecided and name the missing information.
Evidence, current judgment and any gap: ____________________ - Keep a route to revision. Name something you might see that would change your answer. When a correction arrives, compare it with your first conditions and update when it’s warranted.
If you change the conditions, write down the new evidence or changed costs that justify it. Check whether you’d use the same standard for an answer you liked.
Revision condition, or reason for changing my standard: ____________________
Ask Jing
Become a Jingius.
Your next step
You’ve finished when you have a claim, a reason for your current judgement, and a condition for changing it. A gap left clearly open is a useful result too.
A note on the evidence
The studies look at particular judgements, results of training and corrections. They don’t show that cynicism always does more harm than believing too easily. Nor do they show how common lasting doubt of everything is.
The meta-analysis is about spotting manipulative material. The combined video-and-prompt tests used emotional false headlines and neutral true ones. These findings shouldn’t be treated as swappable tests of everyday accuracy. Feedback on single answers is also different from feedback on where you rank.
How we prepared this module
This module links tests, a meta-analysis, a review of research and statistics guidance to everyday factual judgement. The made-up example of a finished task shows different answers. The written exercise helps you state your own conditions for evidence.
Sources
- Megan E. Graham, Brittany Skov, Zoë Gilson, Calvin Heise, Kaitlyn M. Fallow, Eric Y. Mah, and D. Stephen Lindsay. (9 October 2023). Mixed News about the Bad News Game. Journal of Cognition, 6(1), 58, 1–14. Final publisher PDF; Abstract p. 1; Method pp. 2–4; pooled Results pp. 5–7.
- Almog Simchon, Tomer Zipori, Louis Teitelbaum, Stephan Lewandowsky, and Sander van der Linden. (issue 1 February 2026; publisher copyright 2025). A signal detection theory meta-analysis of psychological inoculation against misinformation. Current Opinion in Psychology, 67, 102194. Author abstract, “Results show” sentence; issue 1 February 2026. Full published article, author-uploaded copy, Method pp. 3–4 and Results and discussion pp. 4–6.
- Gordon Pennycook, Adam J. Berinsky, Puneet Bhargava, Hause Lin, Rocky Cole, Beth Goldberg, Stephan Lewandowsky, and David G. Rand. (4 November 2024). Inoculation and accuracy prompting increase accuracy discernment in combination but not alone. Nature Human Behaviour, 8(12), 2330–2341. Author abstract, study-design sentence and final two sentences; accepted author manuscript, Experiments 1–5, especially Experiment 4 Results and Experiment 5 Materials and Procedure/Results.
- Johannes Leder, Lukas Valentin Schellinger, Rakoen Maertens, Sander van der Linden, Breanne Chryst, and Jon Roozenbeek. (2024; accepted 25 March 2024; repository copy deposited 27 October 2025). Feedback Exercises Boost Discernment of Misinformation for Gamified Inoculation Interventions. Journal of Experimental Psychology: General, 153(8), 2068–2087. Published version with repository cover; Experiments 1–2, pp. 2071–2077; General Discussion, p. 2083.
- Benjamin Lyons, Ariana Modirrousta-Galian, Sacha Altay, and Nikita Salovich. (10 June 2025). Reducing Blind Spots? Performance Feedback Reduces Relative Confidence but Does Not Improve Subsequent News Discernment. Collabra: Psychology, 11(1), 138652. Publisher article; Methods, Sample and Design and Procedure; Results, The Effect of Feedback; Appendix Table A1; Discussion, opening paragraphs.
- Thomas Wood and Ethan Porter. (online 16 January 2018; issue March 2019). The Elusive Backfire Effect: Mass Attitudes’ Steadfast Factual Adherence. Political Behavior, 41(1), 135–163. Online-first publisher PDF; Abstract, PDF p. 1; definition of backfire, PDF p. 2.
- Ullrich K. H. Ecker, Stephan Lewandowsky, John Cook, Philipp Schmid, Lisa K. Fazio, Nadia Brashier, Panayiota Kendeou, Emily K. Vraga, and Michelle A. Amazeen. (12 January 2022). The psychological drivers of misinformation belief and its resistance to correction. Nature Reviews Psychology, 1, 13–29. Publisher-typeset review via mirror; “Debunking interventions,” p. 21; Box 2, p. 18.
- NIST/SEMATECH. (undated; accessed 11 October 2026). 7.1.3. What are statistical tests?. e-Handbook of Statistical Methods. Official handbook; “Concept of null hypothesis,” “Significance levels” and “Errors of the second kind.”
