What online health misinformation does to behaviour, and what blunts it
A randomised trial put a number on how much misinformation lowers vaccination intent. A systematic review mapped how common it is. And inoculation studies show the skill of spotting it can be trained at scale.
| Group | Value (pp) |
|---|---|
| United Kingdom | 6.2 (3.9 to 8.5) |
| United States | 6.4 (4 to 8.8) |
Online health misinformation has a measurable effect on behaviour, but a smaller and more specific one than the alarm around it implies: in a randomised trial, exposure to recent misinformation cut the share of people who would "definitely" take a COVID-19 vaccine by 6.2 percentage points in the UK and 6.4 points in the USA [s1]. That is a real, replicated shift in stated intent — not a mass conversion, and not nothing.
The honest reading sits between two familiar postures: the claim that misinformation is rewiring the public wholesale, and the dismissal that it is just noise people ignore. The better-designed studies support neither. They show a modest average effect, an uneven distribution of exposure, and — importantly — that resistance to manipulation can be taught.
The one number that comes from an experiment
Most claims about misinformation's effects rest on correlation: people who believe false things and people who behave a certain way tend to appear together, which cannot establish that one caused the other. The Nature Human Behaviour trial is valuable precisely because it intervened. Researchers randomly assigned people in the UK and USA to see either factual content or circulating misinformation about COVID-19 vaccines, then measured intent to vaccinate [s1].
Relative to factual information, recent misinformation induced a decline in intent of 6.2 percentage points (95th percentile interval 3.9 to 8.5) in the UK and 6.4 percentage points (95th percentile interval 4.0 to 8.8) in the USA, among those who had said they would definitely accept a vaccine [s1]. As of September 2020, the trial also found that fewer people would definitely take a vaccine than is likely required for herd immunity — before any exposure was added [s1].
Two details matter more than the headline figure. The effect was concentrated among people who started out willing, which is where a few points can tip a population target. And scientific-sounding misinformation was more strongly associated with declines in intent than crude falsehood was [s1] — the material that mimics the form of evidence does more work than the material that looks like a rant. That is the same property AI-generated content optimises for, which is why the finding has not dated.
How much of it is out there
If the per-exposure effect is modest, the other half of the question is dose: how much health misinformation circulates at all. A systematic review of 69 studies gave the first broad map [s2]. The most-studied topics were vaccines (32% of studies), drugs or smoking (22%), noncommunicable diseases (19%), pandemics (10%), eating disorders (9%) and medical treatments (7%) [s2].
Where prevalence was measured, it was high. Posts carrying misinformation reached 87% in some studies, misinformation about vaccines was very common at 43% — with the human papillomavirus vaccine the most affected — and it was most prevalent on Twitter and around smoking products and drugs [s2]. Those figures are not a single population rate; they are the range across very different samples and methods, which the review is careful to present as heterogeneous rather than as one number [s2]. The useful takeaway is directional: on contested health topics, a large fraction of what circulates is false or misleading, so exposure is not rare.
The part that is actually encouraging
The reflex response — remove the false posts — runs into scale and speed problems. A more durable approach is to make people harder to fool, and here the evidence is stronger than the debate usually credits. In seven preregistered studies, researchers tested five short videos that "inoculate" viewers against common manipulation techniques: emotionally manipulative language, incoherence, false dichotomies, scapegoating and ad hominem attacks [s3].
Across six randomised controlled studies (n = 6,464) and an ecologically valid field study run on YouTube (n = 22,632), the videos improved recognition of manipulation techniques, boosted confidence in spotting them, increased people's ability to discern trustworthy from untrustworthy content, and improved the quality of their sharing decisions [s3]. The effects held across the political spectrum [s3] — a rare property in this literature, and the reason the approach can be deployed as a platform-wide campaign rather than a partisan one.
What inoculation does not do is worth stating plainly. It trains recognition of technique, not a verdict on any specific claim, and the studies measured discernment and sharing, not long-run behaviour change or vaccination itself [s3]. It is a resilience intervention with demonstrated reach, not a cure.
What the three studies license together
Put in order, the evidence says something narrower and more actionable than either side of the public argument. Health misinformation is genuinely prevalent on the platforms where people discuss contested topics [s2]. A single dose of it produces a real but modest drop in a health intention, largest among the already-willing and largest when it mimics the sound of science [s1]. And the skill of resisting the techniques behind it can be trained at scale, across political lines [s3].
None of that supports the strongest "misinformation is destroying public health" framing, which tends to lean on the prevalence numbers while ignoring the small per-exposure effect. Nor does it support the dismissal, which ignores that a few points of intent can decide whether a coverage target is met. The same measured stance applies to the related debate over social media and adolescent well-being, where the effect is also real, negative and small — and, as here, routinely asked to carry more weight than its size supports.
Sources
- Measuring the impact of COVID-19 vaccine misinformation on vaccination intent in the UK and USA — Nature Human Behaviour, 2021-02-05
- Prevalence of Health Misinformation on Social Media: Systematic Review — Journal of Medical Internet Research, 2020-10-30
- Psychological inoculation improves resilience against misinformation on social media — Science Advances, 2022-08-24
Sources
- Measuring the impact of COVID-19 vaccine misinformation on vaccination intent in the UK and USA — Nature Human Behaviour , February 5, 2021
- Prevalence of Health Misinformation on Social Media: Systematic Review — Journal of Medical Internet Research , October 30, 2020
- Psychological inoculation improves resilience against misinformation on social media — Science Advances , August 24, 2022
When AI symptom-checkers disagree with patients, patients tend to walk away
In 6,772 users of a US health system's AI triage tool, people engaged about twice as often when the AI matched what they already planned — raising a hard question about what the tools are steering.
The FDA's blueprint for AI medical devices: what a maker must show, cradle to grave
A January 2025 draft guidance lays out the documentation the agency expects across an AI device's whole life, including bias checks across demographic groups and postmarket monitoring. It is not yet binding.
The EU AI Act makes most medical AI 'high-risk.' The hard part starts in 2027
Under Article 6, AI that is or sits inside a device already needing an independent safety check counts as high-risk. A 70-study review finds the trouble is overlap with existing device law.
How the FDA lets an AI device update itself without a new clearance
A predetermined change control plan is the FDA's answer to a model that keeps learning: pre-approve a bounded set of future changes, and the manufacturer can ship them without returning for a fresh review.