Trust in leaders and AI adoption: what the data does and does not say

A Great Place To Work blog post dated 14 November 2024 argues in its headline that confidence in leaders builds a culture of innovation (Kitterman, 2024). The implied chain: staff who trust leaders innovate, and staff who innovate get excited about AI. Suppose you run 40 people and are weighing a quarter spent on "building trust" before you buy any AI tool. This post follows that claim to its numbers, then sets it beside other sources.

What two links sit behind the Great Place To Work number?

Two separate ones, and a direct link between trust in leaders and excitement about AI is not among them. Employees with a lot of confidence in their leaders were four times as likely to say they had many meaningful opportunities to take part in innovation. Employees who said they had many such opportunities were 82% more likely to be excited to use AI (Kitterman, 2024).

The post cites Great Place To Work's own global survey of nearly 43,000 employees, in which only 51% were excited to use AI to improve their work. Its text does not say which sample the "four times" and "82% more likely" figures come from, how that sample was drawn, what the comparison group's baseline was, or what was controlled for. "82% more likely" is a relative figure with no percentages on either side, and the outcome is "excited", an attitude, not AI use. The post ends by inviting companies to apply for Great Place To Work certification, so I treat it as commercial material: a starting point, not evidence.

What do the other sources show?

The same direction, measured against different things. The table sets six sources out by what each calls "leadership" and what each calls "AI".

Edelman looks from the respondents' side. In the developed nations of its five-country poll (Brazil, China, Germany, the UK and the US), two-thirds or more of respondents believed business leaders might not be fully honest with employees about what AI does to jobs (Edelman, 2025). I read Richard Edelman's commentary, not the report, so I have no per-country sample sizes and cannot tell whether "two-thirds" holds in each country or across them pooled.

How strong is trust next to a clear AI plan?

Small next to headline ratios, and on Makridis's reading a clear AI plan accounts for most of the link with AI use. Each source uses its own yardstick, so this part needs a slow read.

Korn Ferry reports r = 0.24 and restates it as "nearly 3 times". By my arithmetic, 0.24 squared is about 0.06, roughly 6% of shared variation. An odds ratio depends on where the cut-offs for "high trust" and "positive" are drawn, and the article does not say. It does state that its analyses show relationships, not causation (Korn Ferry, 2026).

Makridis puts both variables in one model, using the Gallup Workforce Panel. Employees who said their organisation had communicated a clear AI plan were about 27 percentage points more likely to use AI frequently (coefficient 0.268). Trust in leadership, counted in standard deviations of a single question, kept a coefficient of 0.016 (standard error 0.007), about 1.6 percentage points; on its own it was 0.019 (section 6). One is a yes/no contrast and the other is counted in standard deviations, so the two coefficients are not on one scale. The author notes that trust keeps some independent predictive content and is the only workplace-practice variable that survives once the clear plan is in the model (Makridis, 2026). The two also move together: regressing a clear plan on trust, with person and wave fixed effects, gives 0.041 (footnote 8, section 6). The author suggests that employees infer "strategy" from managers' behaviour rather than from formal policy. If so, part of trust's effect runs through the clear plan, and the model cannot separate the two routes.

Does trust lead to AI use, or does use lead to trust?

A cross-sectional survey cannot separate the two. The 60-year review by Oreg, Vakola and Armenakis (79 quantitative studies, 1948 to 2007, long before generative AI) gives two pictures. People who trust management and see it as supportive were more receptive to change (Oreg et al., 2011, p. 489). But in several other studies trust in management was treated as an indirect consequence of change (Oreg et al., 2011, p. 500). Most relied on self-report alone (Oreg et al., 2011, p. 502); a majority were cross-sectional, and almost all of those rested on a single self-report source, which invites bias from using one method of measurement (Oreg et al., 2011, p. 514).

Makridis says much the same: neither AI use nor organisational conditions are randomly assigned, and an organisation may communicate an AI strategy after use has spread. The author writes that the estimation strategy is not a claim of causal identification from randomised exposure (section 5.3) (Makridis, 2026).

PwC's figures also read both ways. Its 2026 survey (49,364 workers, 48 countries and territories) found trust in management down seven percentage points on the year. Its "front-runners" (14% of the workforce, 51% using generative AI daily) had 75% trusting senior management and 82% trusting their direct manager, against 33% and 40% in the largest cohort, the "engine room" (56%). The cohorts are built from the benefit workers report getting from AI and the demand for their skills. In my reading, workers who gain and have options may trust management more because they are doing well, not only the other way round. PwC claims neither direction, and the data cannot test either (PwC, 2026).

Experimental evidence is lighter. Xu and colleagues ran a two-factor design (explanation of the AI or none; low or high trust in leaders) with 426 people in an online survey in China, and found that a trustworthy leader raised trust in AI and intention to adopt (Xu et al., 2024). The abstract of McGuire and colleagues reports that a multi-wave field survey and an experiment support this model: leader-provided voice opportunities reduce negative affect about AI-related work, which raises perceived leader trustworthiness and, in turn, trust in AI (McGuire et al., 2026). Intention to adopt an HR-management AI system, and trust in AI, are not use, and I read only the abstracts of both. Oreg and Berson note that trust is one mechanism through which leaders affect change recipients (Oreg and Berson, 2019, p. 288), but they also observe that tests of such mechanisms are limited: in many quantitative studies of leaders' effects the mechanism said to carry the influence was not directly assessed (Oreg and Berson, 2019, p. 285).

Who ran these surveys, and what do they sell?

Nearly all sell what the finding points to. That does not make the numbers wrong; it means I cannot see any result they chose not to publish. On their own pages: Great Place To Work invites certification and rankings (Kitterman, 2024); Gallup offers AI adoption services (Pendell, 2025; Kemp, 2025); Korn Ferry offers AI transformation consulting (Korn Ferry, 2026); PwC offers workforce services (PwC, 2026); Edelman is a communications firm and the commentary's author is its CEO (Edelman, 2025). Makridis lists four affiliations with Gallup Inc. first, uses Gallup data and survey items with permission and thanks Gallup, while stating "These views are our own" (Makridis, 2026). KPMG co-funded the University of Melbourne team, and the report states the team worked independently (Gillespie et al., 2025).

Least tied to selling are the academic reviews (Oreg and colleagues; Glikson and Woolley). Glikson and Woolley state that the work was sponsored by two US defence research bodies, DARPA and the Army Research Office (Glikson and Woolley, 2020, p. 627); Oreg and Berson state that it was supported by the Israel Science Foundation (Oreg and Berson, 2019, p. 272). For the three journal articles I read only in abstract, I have not seen the funding statements.

If the data stop here, what should an owner do?

Do not treat "build trust first" as a precondition, and do not skip it either: try a few cheap steps and measure them on your own team. This is my suggestion, not any source's conclusion. For the hypothetical 40-person company above:

1. Write the AI plan on one page and ask staff one yes/no question: "Has the company communicated a clear AI plan?" Gallup reports that only 22% of US employees say their organisation has communicated a clear plan or strategy for AI, although 44% say it has begun integrating AI (Pendell, 2025); where it has begun, only 28% strongly agree their manager actively supports AI use (Kemp, 2025).
2. Ask the people who do not use AI why. In Gallup's survey, 44% of non-users gave as their main reason that they do not believe AI can help with their work, and 16% that they lack access to tools (Kemp, 2025). If your answers fall mostly in the first group, raising trust is not the first job.
3. Measure use from tool logs or activated accounts, not only self-report, and repeat each quarter with the same people. Makridis compares each person with themselves across survey waves to remove fixed personal traits (Makridis, 2026). With 40 people you cannot detect a difference of about 1.6 percentage points; treat the result as an observation log, not a test.

One more point: trust is not the goal. In the KPMG data, among employees in organisations that use AI, higher trust in AI went with less critical engagement with its output (standardised coefficient −0.24) (Gillespie et al., 2025, p. 84). Glikson and Woolley put the underlying problem plainly: low trust in capable technology leads to disuse, and high trust in incapable technology leads to misuse (Glikson and Woolley, 2020, pp. 629–630). The post on three studies of AI productivity shows why heavy use is not the same as good use.

Going deeper: how does trust in a leader differ from trust in AI, and what do these sources leave out?

They are different objects. The most cited definition, by Mayer, Davis and Schoorman (1995), as Glikson and Woolley quote it, is "the willingness of a party to be vulnerable to the actions of another party based on the expectation that the other will perform a particular action important to the trustor, irrespective of the ability to monitor or control that other party" (Glikson and Woolley, 2020, p. 630). For a leader, the other party is a person who can explain and answer for a decision. For AI it is a system whose behaviour is not deterministic and whose decision logic is often not transparent (Glikson and Woolley, 2020, p. 628). McGuire and colleagues argue that trust in AI is shaped by the social context, leaders included, as well as by technical features (McGuire et al., 2026).

An older critique belongs beside it. Oreg and colleagues cite Dent and Goldberg: when employees "resist", the fault sometimes lies with the change agent or the change itself, not with the employees (Oreg et al., 2011, p. 515). Applied here: if two-thirds or more of respondents in developed markets think leaders might not be fully honest about AI and jobs (Edelman, 2025), low trust may be an accurate reading of a real situation, and "building trust" while the situation stays the same risks becoming public relations. Four theories of resistance to change takes this view further.

What these sources do not cover:

• Vietnam and small firms. Vietnam is not among the 47 countries in the KPMG survey (Gillespie et al., 2025); Gallup is US data; Edelman covers five countries (Edelman, 2025). For the Vietnam study by Nguyen and colleagues (2026) I read only the abstract, and it does not give firm size (Nguyen et al., 2026). For PwC's 2025 Vietnam report I saw only the summary page; the PDF returned a 403 error.
• Real use. Every figure above is self-reported frequency or attitude, not usage logs.
• Dates. Figures are as of 3 November 2026. The Makridis paper is the April 2026 version of a working paper and I did not find a journal version, so its numbers may change.

What did I read? In full: the pages from Great Place To Work, Gallup (two articles), Korn Ferry and PwC (2026 survey), and Richard Edelman's commentary in Fortune; two Edelman PDFs returned a 403 error, so I did not read the underlying report. In part: the KPMG and University of Melbourne report (executive summary, method, the employee section, appendix notes); Makridis's working paper (abstract, introduction, data, estimation strategy, determinants and the start of the results, not the rest of the 90 pages); Oreg and colleagues 2011 (internal context, consequences, method and limitations); Oreg and Berson 2019 (the first half and the passages on trust); Glikson and Woolley (introduction, trust in organisations, discussion; the sections on robots, virtual agents and embedded AI only skimmed). Abstract only: Xu, McGuire and Nguyen, whose full texts at Wiley, Elsevier and Emerald returned a 403 error. I searched in English and Vietnamese.