Why AI Companies Are Losing the Trust War

The same week that Anthropic was navigating the fallout from a US government directive that restricted access to frontier AI models for non-American users, two of the most significant datasets tracking public opinion and media behaviour arrived in my inbox. I looked at the Global Attitudes on AI 2026 Ipsos report and the Reuters Institute for the Study of Journalism Digital News Report 2026 to see if there is any correlation on the news environment affecting the level of trust, or lack of, that exists in AI..

The Ipsos AI Monitor 2026 surveyed 23,532 people across 32 countries between March and April of this year. The Reuters Institute for the Study of Journalism Digital News Report 2026 surveyed nearly 100,000 people across 48 markets between January and February. Neither was commissioned to study the other's subject matter. But when you look at them alongside each other, what emerges is a clear and uncomfortable picture for the AI industry: public perception of AI is under sustained pressure, and the channels through which that perception is formed are themselves in crisis.

This is not primarily a communications problem. It is a strategic one. And the difference matters enormously, for while AI needs to establish its position and shape perception, how it is done is ignoring how public audiences' opinions are being shaped today.

When Excitement and Nervousness Are Equally Matched

The headline from the Ipsos AI Monitor 2026 is deceptively simple. Globally, 51% of people agree that products and services using AI make them excited. Almost exactly the same proportion, 50%, say they make them nervous. This near-parity has been stable for years, and the researchers describe it as a persistent tension between Wonder and Worry. In many cases, the Ipsos researchers note, it is the same people experiencing both emotions simultaneously.

That is a genuinely important observation. The dominant assumption in AI industry communications is that nervousness reflects misunderstanding, and that better information will convert the nervous into the enthusiastic. But if the same individuals are holding both emotions at once, you are not dealing with an information gap. You are dealing with something more complex: ambivalence rooted in lived experience, economic anxiety, and a justified uncertainty about who is accountable for what.

In 17 of the 28 countries surveyed both this year and last, people are now less likely to think AI brings more benefits than drawbacks.

The direction of travel in the Ipsos data deserves particular attention. Germany saw an 8 percentage point fall in the proportion of people excited about AI between 2025 and 2026. Great Britain also fell 8 points. France fell 7. The United States fell 5. These are not marginal statistical wobbles. They represent a measurable and sustained withdrawal of optimism in some of the most economically and politically significant markets on the planet.

At the other end of the spectrum, countries in Asia like China (83%), India (79%), Thailand (77%), and Indonesia (76%) lead on excitement. The markets most enthusiastic about AI's potential are, broadly speaking, those where the perception of AI as a path to economic improvement is strongest. The Ipsos data show a strong correlation in these markets between excitement about AI and the belief that AI will improve the national economy. This is not irrational. For nations that see AI as a leapfrog opportunity rather than a disruption to established structures, the calculus is genuinely different.

Data: Ipsos AI Monitor 2026, 23,532 adults, 32 countries, March-April 2026.

The News Environment Through Which AI Is Understood

This is where the Reuters Institute Digital News Report 2026 becomes essential context, and why reading these two reports in isolation misses the most important signal.

The DNR 2026 records the lowest level of trust in news since the Reuters Institute began tracking this metric in 2015. Only 37% of people globally trust news most of the time. Trust fell in 29 of the 48 markets surveyed. In the United States, where the AI industry is primarily headquartered and where the most powerful AI companies are making their most consequential decisions, only 25% of people now say they trust the news most of the time.

Here is the compounding effect that is not being discussed enough. People who distrust news outlets are not therefore more receptive to corporate communications. The evidence from the DNR suggests that distrust tends to be generalised rather than selective. When institutional trust collapses, it collapses broadly. The Ipsos data show that globally, only 43% of people think business leaders tell the truth. That is the ambient environment in which your earnings calls, keynote addresses, and product announcements are being received.

For the first time in the history of the Digital News Report, social media and video networks are the most widely used source of news globally.

The structural change in how people access information is the other dimension that the AI industry is not taking seriously enough. The DNR 2026 documents a historic milestone: for the first time, social media and video networks are more widely used as a source of news (54% of all respondents weekly) than either television (52%) or news organisations' own websites and apps (51%). The proportion of people who say social media is their main source of news has risen from 22% five years ago to 30% today.

This is the architecture in which public understanding of AI is being formed. Not through considered, specialist journalism. Not through direct engagement with company communications. Through TikTok, YouTube, Instagram, Facebook, and a growing ecosystem of independent content creators who attract audiences precisely because they are personal, urgent, accessible, and unfiltered.

The DNR shows that around a quarter of respondents globally now get some news from individual creators or influencers who focus on news. Respondents describe these creators as more entertaining and easier to understand than traditional news outlets. They also describe them as less trustworthy and less impartial. People are knowingly choosing this tradeoff. Convenience and relatability are winning over reliability.

Data: Reuters Institute Digital News Report 2026, YouGov, ~100,000 respondents, 48 markets, January-February 2026.

The Disinformation Paradox

One of the most strategically significant findings in the Ipsos AI Monitor 2026 sits in the data on disinformation. Globally, only 27% of people think AI will make the amount of disinformation on the internet better. Some 45% think AI will make it worse. This finding does not cluster neatly in one geography or age group. It runs across markets, including those that are otherwise broadly positive about AI.

Now hold that alongside the DNR finding that concerns about fake news and misinformation rose 4 percentage points globally to 62% of respondents. Jumps of more than 5 percentage points were recorded in 11 markets. The platforms through which people now primarily consume news are the same platforms they trust least for reliability. This is a structural feature of the current information environment, not a temporary problem that will resolve itself.

The intersection of these two datasets creates a specific reputational exposure for AI companies that very few appear to be addressing proactively. If a majority of people believe AI will make disinformation worse, and if they are simultaneously consuming news about AI through platforms they consider unreliable, the narrative about AI's societal impact is being written in a fragmented, low-trust, high-anxiety information environment. The people most likely to trust AI chatbots for news, the DNR shows, are those who already use them and have built direct experience. That is 10% of the global population. The other 90% are forming their views through secondary sources they do not fully trust.

Why Poor Tactical Communications Makes This Worse

There is a pattern that has emerged in AI industry communications over the past three years that this data should finally put to rest. The dominant approach has been to announce capabilities, host demonstrations, publish ethics frameworks, and issue statements responding to concerns. This is tactical communications. It is reactive, episodic, and fundamentally unsuited to the environment the data describe.

Tactical communications assumes a relatively stable and accessible public information environment where accurate messages, distributed at sufficient volume, will reach and persuade target audiences. The DNR and Ipsos data show that this environment does not exist. The people who most need to be persuaded are the ones least reachable through traditional channels. They are getting their information about AI through social media and content creators whose economic model rewards controversy and simplification. They are doing so in a context of generalised institutional distrust. And they are making sense of AI at a moment when their own economic anxieties about job displacement are high.

The Ipsos data show that across all 32 countries surveyed, more than one in three people are concerned that AI will replace their job in the next five years. Among those under 35 who are actually using AI tools at work, the rate of job displacement anxiety remains significant even as they acknowledge productivity gains. This is a cohort experiencing the productive reality and the disruptive threat of AI simultaneously. A campaign will not resolve that.

What the data actually point toward is not a communications strategy but a trust strategy, and the two are not the same thing.

A trust strategy starts not with message development but with an understanding of the general populous and audiences and their established institutional behaviours. It asks: what decisions can we make, what partnerships can we form, what standards can we commit to, and what accountability mechanisms can we accept, that would credibly reduce the distance between what we say and what people experience? It is slower and harder than a campaign. It is also the only thing that works in an environment defined by generalised institutional scepticism.

The Ipsos data are instructive on one specific point. Across all 32 countries, 80% of people believe that products and services using AI should have to disclose that use. In Great Britain the figure is 85%. In France it is 85%. In the United States it is 82%. This is among the most consistent findings in the entire 48-slide dataset. Companies that are still treating disclosure as an optional or aspirational commitment are not misreading a communications environment. They are misreading a strategic one.

The Geographic Segmentation Imperative

One of the most useful things the Ipsos data do is reveal how geographically differentiated the AI perception challenge really is. The Ipsos researchers plot countries on a scatter chart mapping levels of excitement against levels of nervousness. What emerges are three distinct clusters.

The Anglosphere sits at high nervousness and moderate excitement: Canada, Australia, the United States, Great Britain, and New Zealand all cluster together with nervousness levels of 62 to 67% and excitement levels of 26 to 40%. Europe sits at lower nervousness but also lower excitement: Germany, Netherlands, France, Sweden, and Belgium tend toward a subdued rather than anxious relationship with AI. A third cluster, what the researchers label as 'More Positive,' encompasses India, Thailand, Indonesia, China, and South Korea, where high excitement coexists with relatively lower nervousness.

These three clusters require fundamentally different approaches. In the Anglosphere markets, where economic anxiety about AI job displacement is acute and trust in institutions is at historic lows, the communications challenge is not to generate enthusiasm. It is to reduce the perception that AI development is proceeding without accountability. In the European markets, where regulatory frameworks are more advanced and the cultural expectation of governance is stronger, the communications challenge is different again. In the high-excitement markets of Asia-Pacific, the opportunity exists to build substantive track records that can then be referenced in more sceptical markets.

Most AI companies are running global communications programmes that treat all these markets as variations on the same challenge. The data suggest they are not.

Data: Ipsos AI Monitor 2026, scatter analysis pages 14-15.

The Young Are Not the Allies You Think They Are

The assumption that demographic change will resolve the AI perception problem is one of the most persistent strategic errors in the sector. It runs roughly as follows: younger people are digital natives, more comfortable with technology, more likely to be using AI already, and therefore the passage of time will naturally convert sceptics into supporters as the older, more nervous generation is replaced.

The Ipsos data challenge this directly. People under 35 are the most likely to be both nervous (52%) and excited (56%) about AI. This is the highest dual-sentiment score in the dataset. High ambivalence is not the same as high acceptance. And the DNR adds an important layer: interest in news among people under 35 has fallen by an average of 13 percentage points since 2021. Younger audiences are simultaneously the most intensive users of AI tools, the most anxious about AI's impact on their livelihoods, and the most disengaged from the traditional journalism that might offer context and accountability.

The DNR also documents something that the AI industry should find genuinely concerning. The report notes that at graduation events across American universities this year, speakers attempting to generate enthusiasm about AI were jeered by audiences of graduates entering an uncertain labour market. The Ipsos data confirm that those under 35 are more likely than any other age group to believe AI will replace their job in the next five years. The generation growing up with AI is not uniformly enthusiastic about it. They are using it under conditions of economic insecurity and watching the adults responsible for its development move faster than the governance structures designed to protect them.

What Strategy Actually Looks Like Here

The data point toward a set of strategic priorities that go well beyond the communications brief.

The first is to commission serious, disaggregated perception tracking, by market, by income segment, by employment sector, and over time. The global average is a distraction. What matters is understanding the specific shape of trust and scepticism in the specific markets where your regulatory exposure, your customer base, and your institutional relationships are most significant.

The second is to take the disinformation question off the corporate responsibility team's desk and put it on the strategy table. The Ipsos finding that 45% of people think AI will make disinformation worse is not primarily a reputational risk. It is a signal about where the public conversation about AI accountability is heading. Companies that develop credible, substantive positions on AI and information integrity before they are forced to will be in a significantly stronger position than those that respond reactively.

The third is to treat disclosure not as a regulatory compliance matter but as a trust-building tool. The near-universal public support for AI disclosure, documented in the Ipsos data, is an opportunity as much as an obligation. Organisations that go further than required and make disclosure meaningful and accessible are building the kind of trust that surveys cannot easily manufacture.

The fourth is to reframe the relationship with journalists and news organisations, not as platforms for corporate announcements, but as partners in helping publics understand a genuinely complex technology. The DNR documents the ongoing collapse of traditional news consumption. But it also notes that trust in individual established news brands is holding up better than trust in news overall. The organisations still capable of producing trusted, in-depth AI coverage are a finite and valuable resource. The AI industry's current relationship with serious journalism is largely transactional and episodic. It should be something more substantive.

The fifth, and perhaps the most important, is to accept that the current window for building durable trust is time-limited. Both reports document a dynamic in which people continue using technologies they do not entirely trust because the convenience and competitive pressure to do so is currently high. This is not a stable equilibrium. A significant AI failure, a sustained economic disruption, or a major regulatory intervention in a key market could convert accumulated ambivalence into active resistance faster than any communications team can respond. The strategic question is not how to manage perception today. It is how to build the institutional credibility that will survive whatever comes next.

Reputation Matters

Both of these reports are, at their core, about the gap between what institutions say and what people experience. The Ipsos AI Monitor 2026 shows that 63% of people globally say they do not always trust AI tools, but use them anyway. The Reuters Institute Digital News Report shows that people continue consuming news from sources they distrust because they fear missing out or becoming irrelevant.

These are not healthy patterns. They are the behavioural signatures of an environment in which trust has been so degraded that people no longer expect the things they use to deserve it. The AI industry did not create that environment. But it is operating inside it, and it is making decisions that will either contribute to it or begin to change it.

The companies that understand the distinction between a communications challenge and a trust challenge, and that are willing to invest in the harder, slower work of the latter, are the ones that will still have public permission to operate when the current cycle of anxiety and ambivalence eventually resolves.

The ones that treat this as a messaging problem will find out what happens when the say-do gap finally closes.