The Message Was Never the Problem

If a health education programme fails, the subsequent analysis typically centres on the message. Is the language clear? Is the translation accurate? Is the imagery relevant? These are legitimate questions, and they are almost never the right ones.

I have worked on communications across nations: in the media, in UN-linked institutions and for programmes operating in multiple countries. The pattern I see repeatedly is the same. The data is usually good. What varies is whether the audience believes the messenger.

Trust counts for more than anything else

According to a paper published in New Media & Society in 2025, 8,541 respondents from eight countries — Brazil, France, Germany, India, Japan, Pakistan, the United Kingdom and the USA — answered a survey question on what predicts having false beliefs about climate change. Researchers checked the usual suspects. Political orientation mattered. Media use did too, but mostly offline media use decreased false beliefs, while use of online media and social media showed almost no correlation with false beliefs. Specific exposure to different information sources explained 1 per cent of the variance.

Next, they checked trust, and it blew everything else out of the water. Trust in sources alone made up 15 percentage points of the explanatory power, compared to 3 per cent for media use. Trust in scientists, in environmental organisations and in international bodies reduced false beliefs. Trust in politicians, in celebrities and in energy companies increased them, each by a near-identical margin.

The differences in false beliefs by country are stark. In India, 47 per cent believed false claims about climate change. In Japan, only 12 per cent of respondents. The same planet, the same physics, the same science — and a quadrupled rate of misbeliefs.

This is not a translation problem. No messaging can close a gap like this.

Trust is not equally distributed, but we design campaigns as if it is

The 2018 Wellcome Global Monitor surveyed public attitudes to science and health in more than 140 countries. It found that 73 per cent of the world's population would trust a physician or a nurse over any other source for health advice. Family members and friends come second, at 14 per cent.

Underneath the global figure, there are dramatic differences by region. Worldwide, 84 per cent of the population have "a lot" of trust or "some" trust in physicians and nurses. In Norway, Denmark, Spain, Finland and Iceland, this figure is 97 per cent or higher. In Central Africa, it is the lowest observed anywhere. Trust in scientists specifically ranges from 54 per cent reporting high trust in Uzbekistan to 47 per cent reporting low trust in Gabon.

A campaign developed in a high-trust country carries an assumption so obvious, nobody ever thinks to question it: that an official source will be believed. Move that campaign to a place where this assumption does not hold and it will not just underperform. It will have no legs to stand on.

The AI layer makes things materially worse

According to an audit of five widely used AI chatbots published in BMJ Open in April 2026, half of all answers were problematic: 30 per cent of them somewhat problematic, and 20 per cent highly problematic. One system produced highly problematic responses 58 per cent of the time.

The bots performed best on vaccines and cancer, and worst on nutrition, stem cell and athletic performance questions. That is to say, worst on exactly the kind of everyday questions that people are most likely to ask a computer, rather than a doctor.

Two findings from the study concern me more than the accuracy rate itself.

The first is the tone. The answers were largely caveat-free and, according to the authors, were delivered "with confidence and certainty". The second finding is the citation. Fabricated references occurred often enough to leave every single chatbot without a complete reference list: on average, the completeness of the references was around 40 per cent.

Thus, we now have a source of information that is fluent, multilingual, confident in tone, and problematic in half of the answers it gives. And this source of information can also manufacture references on the spot. In addition, every answer in the audit was formulated at the college graduate reading level, thus putting it beyond the reach of many people who most need it.

Compare this to the trust findings. In places where institutional trust is already low, a source of information that sounds certain and needs no effort from the audience will not compete with the health system. It will replace it.

What this means for communicators

There are three conclusions to be drawn, and none of them has anything to do with wording.

Check who is trusted first, and then think about what to say. In most cases, this will be a physician, and in many cases, it will be a family member. It is rarely an institution, and almost never a foreign one.

Treat confidence in answers as a danger signal rather than a merit. The most damaging information out there today is not tentative. It is confident, immediate and unqualified, and we have been training our audiences to interpret such qualities as knowledge for decades.

Design communications for verification, not just for comprehension. A reader who knows the message, but cannot verify it, is a single misleading message away from abandonment.

Environmental and health literacy has always been a difficult task. But providing accurate information has ceased to be the obstacle. What remains scarce across the world, and what quietly decides the outcome, is whether anybody trusts the source it comes from.

Weirong Li serves on the Board of Directors of World Information Transfer as Secretary. She is a strategic communications advisor working on positioning, media strategy and cross-cultural message design, and is the author of the C.L.A.R.I.T.Y. Framework for intercultural communication.