False stories spread because people share them, not mainly because bots push them. A large study of Twitter from 2006 to 2017 found falsehood outran the truth, and bots made little difference to the gap. What slows it down also works on people. Each method helps a little; none is a cure.
- Two questions sort the words. Is it false, and is harm meant? Misinformation is false without intent to harm, disinformation is false on purpose, malinformation is true but used to hurt.
- False news outran true news in a large study of Twitter, and people, not bots, made the difference.
- Novelty is the leading explanation, not a proven cause.
- Three habits have evidence behind them: learning the tricks in advance, leaving a page to check who is behind it, and asking whether something is accurate. The effects are small to modest.
- AI makes persuasive falsehood cheaper to write. Getting real people to share it is still the hard part, on the evidence so far.
Three words for three problems
The clearest definitions come from a 2017 report for the Council of Europe by Claire Wardle and Hossein Derakhshan. “Mis-information is when false information is shared, but no harm is meant.” “Dis-information is when false information is knowingly shared to cause harm.” “Mal-information is when genuine information is shared to cause harm, often by moving information designed to stay private into the public sphere.”Council of Europe, Information Disorder: Toward an interdisciplinary framework for research and policy making, 2017, read at source 18 Sep 2026.
One post can be both. It can start as disinformation and become misinformation when an honest person passes it on. Of rumours shared after an attack in Paris, the report says: “The people sharing this type of content are rarely doing so to cause harm.”Council of Europe, Information Disorder, 2017, read at source 18 Sep 2026.
Plenty of people who pass on a false story mean no harm. Treat every sharer as a liar and you get the problem wrong, and the fix.
Why false stories travel faster
A 2018 study in Science by three MIT researchers followed roughly 126,000 stories on Twitter from 2006 to 2017, tweeted over 4.5 million times by about 3 million people, each marked true or false using six fact-checking organisations.MIT News, Study: On Twitter, false news travels faster than true stories, 8 March 2018, read at source 18 Sep 2026: “tracked roughly 126,000 cascades of news stories spreading on Twitter, which were cumulatively tweeted over 4.5 million times by about 3 million people, from the years 2006 to 2017.”
Falsehood spread “significantly farther, faster, deeper, and more broadly than the truth in all categories of information”. False stories were 70% more likely to be retweeted, and a true story took about six times as long to reach 1,500 people. Bots did not explain it: “robots accelerated the spread of true and false news at the same rate”.Vosoughi, Roy & Aral, The spread of true and false news online, Science, 2018, MIT Media Lab copy, read at source 18 Sep 2026: “falsehoods were 70% more likely to be retweeted than the truth”; “It took the truth about six times as long as falsehood to reach 1500 people”.
The share button is the engine. Anything that slows false stories has to work on the people pressing it.
What that study does not say
It is one platform, ending in 2017. The authors “emphasize that careful studies are needed” before assuming other platforms behave the same way.MIT News, Study: On Twitter, false news travels faster than true stories, 2018, read at source 18 Sep 2026.
It covers stories big enough to be fact-checked. The authors flag “a selection bias” from that, then report that a second sample no fact-checker had examined gave “results nearly identical to those estimated with our main data set”.Vosoughi, Roy & Aral, The spread of true and false news online, Science, 2018, read at source 18 Sep 2026.
It does not prove why. False news was more novel, but “we cannot claim that novelty causes retweets or that novelty is the only reason why false news is retweeted more often”.Vosoughi, Roy & Aral, The spread of true and false news online, Science, 2018, read at source 18 Sep 2026.
It gets stretched in retelling. MIT’s own list of press coverage summarises one opinion piece as describing false news “traveling six times faster than factual news”. The paper measured something narrower: time to reach 1,500 people.MIT News, Study: On Twitter, false news travels faster than true stories, press mentions, read at source 18 Sep 2026: “traveling six times faster than factual news”. See how to read a number before you repeat it.
Research about misinformation gets misquoted too. Quote the measure, the platform and the years.
What slows it down
Prebunking: learn the trick before you meet it. Cambridge-led researchers made five short videos, each explaining one manipulation technique, such as emotional language or a false either-or. In seven preregistered studies, one run as YouTube ads with 22,632 responses, the videos “improve manipulation technique recognition” and “improve the quality of their sharing decisions.” On YouTube the gain was “about 5% on average”; the authors “were unable to study how long the inoculation effect remains significant”; and “This study was funded by Google Jigsaw.”Roozenbeek et al., Psychological inoculation improves resilience against misinformation on social media, Science Advances, 2022, read at source 18 Sep 2026: “an ecologically valid field study on YouTube (n = 22,632)”.
Lateral reading: leave the page to judge the page. A Stanford study watched 10 historians, 10 professional fact checkers and 25 undergraduates judge unfamiliar websites. Historians and students often stayed on the site and studied it. Fact checkers “read laterally, leaving a site after a quick scan and opening up new browser tabs in order to judge the credibility of the original site.” The authors warn that with samples this small, “we can’t rule out the possibility that doubling or tripling sample size would have produced different results.”Wineburg & McGrew, Stanford University, Lateral Reading and the Nature of Expertise, 2018 manuscript in press at Teachers College Record, read at source 18 Sep 2026: “We sampled 45 experienced users of the Internet: 10 Ph.D. historians, 10 professional fact checkers, and 25 Stanford University undergraduates.” It shows what skilled checkers do; it did not test teaching it.
Accuracy prompts: ask whether it is true before sharing. A 2021 Nature paper found that “sharing does not necessarily indicate belief”, and that “subtly shifting attention to accuracy increases the quality of news that people subsequently share.”Pennycook et al., Shifting attention to accuracy can reduce misinformation online, Nature, 2021, read at source 18 Sep 2026: “subtly shifting attention to accuracy increases the quality of news that people subsequently share.” Pooling 20 of their own experiments, the researchers found the prompts worked mainly by cutting sharing intentions for false headlines, by 10% against control, and called the effects “modest in size”. Those are intentions in surveys, not observed shares.Pennycook & Rand, Accuracy prompts are a replicable and generalizable approach for reducing the spread of misinformation, Nature Communications, 2022, read at source 18 Sep 2026: “primarily by reducing sharing intentions for false headlines by 10% relative to control in these studies”.
All three work on the person, not the post, and help at the margin. They are cheap, and they stack.
What AI changes, and what it does not
It lowers the cost of writing. In a December 2021 survey of 8,221 US adults, researchers compared real foreign propaganda with articles written by an OpenAI model, GPT-3. With no article, 24.4% agreed with the propaganda’s claim; after the real thing, 47.4%; after the AI version, 43.5%.Goldstein et al., How persuasive is AI-generated propaganda?, PNAS Nexus, 2024, read at source 18 Sep 2026: “While only 24.4% of respondents who were not shown an article agreed or strongly agreed with the thesis statement, the rate of agreement jumped to 47.4%”; “43.5% of respondents who read a GPT-3-generated article agreed”. That measured one article read on request, not whether anyone would have found or shared it.
It has not, so far, solved distribution. Of five covert influence campaigns it caught using its models, OpenAI reported: “As of May 2024, these campaigns do not appear to have meaningfully increased their audience engagement or reach as a result of our services.” That is a company describing campaigns it caught; it cannot speak for ones it missed.OpenAI, Disrupting deceptive uses of AI by covert influence operations, 30 May 2024, read at source 18 Sep 2026. The Council of Europe report made the older point: “Without amplification, dis-information goes nowhere.”Council of Europe, Information Disorder, 2017, read at source 18 Sep 2026. For how much of the web is now automated, see the dead internet theory.
Expect more convincing falsehood for less money. The bottleneck is still a person deciding to pass it on.
Before you share
- Ask whether it is accurate, not whether it is interesting.
- Leave the page. Search for who is behind the site or account before reading it closely.
- Name the technique: outrage, a false either-or, a convenient scapegoat.
- If it is a screenshot, find the original — see how to check a viral screenshot.
The habit underneath all four is on critical thinking.Reasoning, September 2026 — summarises this page’s section on what slows misinformation down.
What this page could not verify
- Whether the Twitter findings hold on today’s platforms. This page did not find a repeat at the same scale.
- How long prebunking lasts outside a study. The YouTube test measured recognition within a day.
- How much AI-written misinformation circulates now. This page did not find a reliable measure.
- Two publisher pages. Science and SAGE refused automated access; both papers were read in university-hosted copies.
Council of Europe, Information Disorder, 2017 · Vosoughi, Roy & Aral, The spread of true and false news online, 2018 · MIT News, Study: On Twitter, false news travels faster than true stories, 2018 · Roozenbeek et al., Psychological inoculation improves resilience against misinformation on social media, 2022 · Wineburg & McGrew, Lateral Reading and the Nature of Expertise, 2018 · Pennycook et al., Shifting attention to accuracy can reduce misinformation online, 2021 · Pennycook & Rand, Accuracy prompts are a replicable and generalizable approach, 2022 · Goldstein et al., How persuasive is AI-generated propaganda?, 2024 · OpenAI, Disrupting deceptive uses of AI by covert influence operations, 2024. All read at source on 18 September 2026.