The AI Sentiment Gap, measured

Net negative, and worsening sharply. Using a Net Sentiment Score (NSS) – a standard formula for turning positive and negative survey shares into a single number from −100 to +100 – the current public mood on AI works out at −62.5, down from −44.2 a year earlier (a −18.3 point shift).

This is a single-source calculation, not a composite: the Bentley University-Gallup Business in Society survey asks Americans one direct question – does AI do more good than harm, more harm than good, or an equal amount of both – every year, to the same population, on a consistent methodology. In 2025: 12% good, 31% harm, 57% equal. In 2026: 9% good, 39% harm, 52% equal.

The "equal amounts of harm and good" middle ground (57% in 2025, 52% in 2026) is excluded from the NSS calculation itself, which is why the score reads more starkly negative than the raw "39% think AI does more harm than good" headline – the formula measures the balance of opinion among people who have taken a side, not the whole population. Both are legitimate readings of the same data; this page uses NSS because it isolates the direction of travel among those with a firm view. This is a distinct measure from the Expert–Public Gap Clock below: NSS captures the public's overall mood in isolation, while the Gap Clock captures the distance between how experts and the public each answer the same specific question. The two numbers are not directly comparable, and shouldn't be read as two readings of the same thing.

The AI Sentiment Gap is the measurable, widening difference between how the people building and deploying AI feel about it, and how the people using it, working alongside it, or living downstream of it actually feel. It shows up wherever the same question is asked of two groups sitting on either side of an AI deployment decision – executives versus employees, experts versus the public, older generations versus younger ones – and produces two very different answers.

Stanford put a number on the broadest version of it: a 50-percentage-point gap between AI experts and the US general public on whether AI's impact on jobs, the economy, and healthcare will be positive – roughly 1.5 times the scale of US political polarisation on a single technology question.

No – and treating it as one is precisely the mistake driving the gap wider. A PR problem is solved by better messaging about a reality both sides already agree on. A sentiment gap this size means the two sides are not disagreeing about the same facts; they are living different realities inside the same deployment.

CARMA's analysis of 6,006 media articles and 6,300 respondents across 19 countries found public trust in AI is now shaped more by direct product experience than by media narrative – which means better coverage cannot close a gap that is being opened by lived experience of the technology itself. The fix has to run through what AI actually does to people's work and lives, not through how confidently that is described afterwards.

Substantial, and growing. Tricentis' 2026 Quality Transformation Report, surveying 2,501 IT and QA leaders across six countries, found the confidence gap runs straight down the org chart.

The report's own conclusion is instructive: this is not AI underperforming, it is organisations discovering the governance gap between confident sponsorship at the top and hard-won caution on the front line.

Considerably less than those executives may assume. A 2026 CNBC Generation Labs survey of over 1,000 US adults aged 18–34 asked who they trust to act responsibly on AI, offering nine leading AI executives as options.

The best-scoring name, Microsoft's Satya Nadella, still only reached a 35% trust rating. Palantir's Alex Karp scored lowest, with 81% of respondents selecting "don't trust". This is not diffuse scepticism about a technology category – it is a specific, personal verdict on named leadership, and it lands hardest with exactly the workforce cohort AI adoption depends on retaining.

By the clearest generational tracker available, it is widening, and quickly. A Gallup poll conducted for the Walton Family Foundation and GSV Ventures surveyed 1,572 people aged 14–29 in February and March 2026.

This shift happened even though roughly half of Gen Z uses AI daily or weekly. Gallup's own researchers attribute the anger to AI's dimming effect on entry-level job prospects, with the oldest, most labour-market-exposed members of Gen Z registering the strongest negative shift.

No – and the disagreement is itself part of the story. Stanford HAI's global tracking shows the share of people worldwide saying AI's benefits outweigh its drawbacks actually rose from 55% in 2024 to 59% in 2025. But in the same window, the share saying AI products make them nervous rose to 52%.

Optimism and anxiety are rising together, not trading off against each other – a pattern consistent with a public that is adopting the technology out of necessity while trusting the people and institutions deploying it less. CARMA's parallel data shows a similar split: measured public trust in AI ticked up to 30% in 2026 (from 26% in 2025) even as trust signals in media coverage declined to 26% (from 35%) – trust rising even as the Bentley-Gallup Net Sentiment Score above turns sharply more negative over the same period, underlining how differently "trust" and "net good vs. harm" can move.

Yes, and it interacts with national politics in ways that make a single global response impossible. Stanford HAI's 2026 Index found the United States has the lowest public trust in its own government to regulate AI of any country surveyed.

OECD's 2026 Trust Survey across member states found scepticism about public-sector AI use rises from 24% among 18–29 year-olds to 41% among those over 50, and is consistently higher among people who feel financially insecure or at risk of discrimination – meaning the sentiment gap is not evenly distributed even within the sceptical half of the public.

Three things the data points to consistently. First, measure the gap internally before it becomes visible externally – the Tricentis CEO/QA Director split shows the gap exists inside organisations well before it reaches customers or the press, and it is measurable with an ordinary staff survey.

Second, treat sentiment as a lagging indicator of lived experience, not a messaging output – CARMA's finding that direct product experience now outweighs media narrative means the gap closes through what AI is actually allowed to do to people's jobs, not through how that is described afterwards.

Third, disaggregate the public – age, financial security, and proximity to AI-exposed roles all predict sentiment more strongly than demographics alone.