Public trust in science has not collapsed; the survey record is clear on that. But trust is not transmission. In all 125 countries measured, people badly misjudge what their neighbours think — and what circulates online is selected for negativity, not accuracy.
Eighty-one per cent of British adults say they generally trust scientists to tell the truth; 13% say they do not.7 That places scientists fifth of the 28 professions in Ipsos's 2025 Veracity Index — behind nurses, engineers, doctors and teachers — and well clear of the 52% average across the full list.7
The international picture points the same way. A preregistered survey of 71,922 people in 68 countries put mean trust in scientists at 3.62 on a five-point scale, where 3 is neither high nor low, and found no country with low overall trust.1 Its authors set out to test the popular claim of a crisis of trust, and reported that the claim is not supported by the evidence.1
The index's newest entry sits at the opposite pole. Social media influencers, added in 2025, are trusted to tell the truth by 6% — the lowest score any profession has recorded since the series began in 1983.7 The gap between scientists and influencers is 75 percentage points.
Britons are not credulous about the channel. In a separate Ipsos survey, 31% said they trusted online news from influencers a great deal or a fair amount, while 74% said misinformation was prevalent in it — the only source measured where the doubting share exceeded the trusting one.8
Trusting scientists is not the same as absorbing what they establish — and the sharpest measure of that distinction is not what people believe about science, but what they believe about each other.
The Gallup World Poll asked 129,902 people in 125 countries whether they would give 1% of their household income every month to fight global warming. Sixty-nine per cent said yes; 86% said people in their country should act, and 89% wanted their national government to do more.2 Respondents were then asked how many of 100 fellow citizens would do the same. The average guess was 43% — a gap of 26 percentage points against the measured figure.2
The gap runs the same way in every country surveyed: in all 125, the average estimate falls below the real proportion, and 81% of individual respondents underestimated their own country's figure.2 This matters because the belief is load-bearing. The same study found that a one-point rise in the perceived share of others willing to contribute is associated with a 0.46-point rise in a respondent's own probability of contributing — so a pessimistic misreading of one's neighbours is self-fulfilling.2
The study's authors do not treat the gap as inexplicable. They attribute it in part to media and public debate giving disproportionate prominence to sceptical minority opinion, and to organised campaigning — an attribution, not an experimental result.2
Distrust is no barrier to reach. In 12,448 randomised trials run on live news headlines — 53,699 headline variants, 2.78 million clicks — each extra negative word in a headline of average length raised the click-through rate by 2.3%, while each positive word cut it by 1.0%.3 Across 2.73 million posts from news accounts and US legislators, each term referring to a political opponent raised the odds of a share by 67%, an effect about 4.8 times that of negative-affect language.4
The mechanism is narrower than the shorthand suggests. In the same headline trials, sadness raised click-through, but fear lowered it, the coefficient for anger was not statistically significant, and explicitly moralised wording reduced clicks outright.3 "Outrage" is not what the data shows winning; a flat, bleak register is.
One reading — the desk's interpretation, not a finding in any paper cited here — is that an attention market selects for a downbeat framing of the world, and that this is precisely the framing under which a well-supported majority position looks like a contested minority one. The evidence for the selection effect is strong. Evidence that it causes the perception gap does not exist.
Nothing cited here establishes that attention effects cause the perception gap. The two are measured in different populations, on different subjects, by different methods, and the link remains an attribution rather than a result.2 The headline trials ran on one US site between 2013 and 2015, and their topic model covered entertainment, politics, parenting and rights coverage — no science category — so the negativity effect is unmeasured on science communication.3 The perception gap is documented for one issue, climate, and should not be assumed to generalise to every scientific question. The Veracity Index is a quota sample, not a probability sample, so no formal margin of error applies; Ipsos also cautions that its 2020 switch from face-to-face to telephone interviewing makes comparison with earlier years unsafe.7 And "trust to tell the truth" is blunt: the 68-country study put scientists' perceived openness to feedback (3.33) well below their perceived competence (4.02).1
Whether algorithmic ranking changes what people think, as opposed to what they see, is unsettled among researchers running comparable field experiments. Both trials below measured political attitudes; neither measured trust in scientists or beliefs about other people.
US users switched onto X's algorithmic feed for seven weeks in 2023 shifted toward more conservative positions on policy priorities and current affairs, and went on following the accounts they had picked up after the algorithm was switched off again.5
Three months on reverse-chronological Facebook and Instagram feeds during the 2020 US election substantially changed what users saw and how much they used the platforms, but did not detectably alter issue polarisation, affective polarisation or political knowledge.6