AI transformations run on trust, not reassurance

Anyone who has sat through a restructure knows the moment. The CEO stands at the front of the room, or beams in from a studio, and says that nobody needs to worry. Jobs are safe. This is about augmentation, not replacement. Everyone in the room nods, and half of them quietly update their CVs that night. The words were meant to calm people down. What they actually did was tell a room full of intelligent adults that the leadership team either didn't know what was coming or wasn't going to say.

I have watched this scene play out in some form at nearly every large organisation we have worked with at Collective Campus since we began running corporate innovation programmes in 2015, and more sharply since we started teaching AI skills in 2022. The pattern is consistent. Leaders reach for reassurance because they believe the main risk in an AI rollout is employee fear. They are half right. Fear is real. But reassurance is the wrong tool for it, and a recent piece of research from McKinsey explains why.

What the data says about trust

In August 2026, Aaron De Smet, Bonnie Dowling, Holly Price and Ignacio Fantaguzzi published AI transformations run on trust at McKinsey. Their central argument is that trust in an organisation and its leaders is one of the strongest predictors of whether employees are ready to use AI. Not the quality of the tools. Not the size of the training budget. Trust.

The starting position is better than most executives assume. The Edelman 2026 Trust Barometer found that 78 per cent of employees globally trust their employer to do what is right. That is a substantial reserve of goodwill, and it is the thing an AI transformation draws down on. McKinsey's own AI preparedness survey then shows what happens when that reserve is healthy: organisations with higher trust report capturing greater enterprise value from AI.

That finding deserves a pause, because it cuts against how most AI programmes are designed. The typical programme begins with a technology decision, adds a change management workstream somewhere in the middle, and treats trust as a soft output that gets measured in an engagement survey after the fact. The research suggests the sequence is backwards. Trust is an input. It determines how much of the technology's potential you will ever see.

The anxiety numbers make the point from the other direction. Roughly one in five respondents at every level of the organisation feel anxious about AI-related changes to their work. Among middle managers, it is closer to one in four. Those managers are the people we ask to translate strategy into daily practice, to coach their teams through new workflows, and to absorb the questions that never make it to the town hall. If a quarter of them are worried about their own position, we have a structural problem, not a communications problem.

The gap widens sharply where trust is weak. Employees who report low trust in their organisation's support during an AI transformation are 1.5 times more likely to feel anxious. Anxiety, in turn, shows up as hedging, quiet resistance, and the polite non-adoption that never appears on a dashboard but drains the value out of a licence purchase.

Why reassurance backfires

The McKinsey authors name three ways reassurance goes wrong, and each one matches what I have seen in practice.

First, it rings hollow. Employees read the news. They know that other companies have cut roles and cited AI as a reason. When a leader says nobody's job is at risk, the listener runs a quick check against what they know about the world and finds the statement implausible. The speaker loses credibility on the one topic where credibility matters most.

Second, it destroys trust when events contradict it. This is the expensive failure. A promise of job security works only until the first role changes, the first team is merged, or the first function is quietly redesigned around a new tool. At that point the original statement becomes evidence. People do not remember that you meant well. They remember that you said something that turned out not to be true, and they apply that discount to everything you say afterwards, including the useful things.

Third, it misses the chance to connect. Reassurance closes a conversation. It tells the other person how to feel and invites them to stop asking questions. Real worry is rarely about a single yes or no on job security. It is about whether my skills will still matter, whether I will be given time to learn, whether someone will tell me the truth early enough for me to act on it. A blanket promise answers none of those, and it signals that the harder conversation is off the table.

I would add a fourth, from my own observation. Reassurance is often a leader's way of managing their own discomfort. It is far easier to say that everything will be fine than to say that you do not yet know how roles will change, but that you will tell people as soon as you do. The second sentence is harder to deliver and far more trustworthy.

Four things leaders can do instead

De Smet and his colleagues propose four actions. None of them is exotic, which is rather the point.

The first is to set out a clear plan, and to accept that it will not be perfect. Employees do not need certainty. They need to see that someone is thinking carefully, that there is a direction, and that the plan will be adjusted in the open as evidence comes in. A plan that admits what it does not know earns more confidence than a polished one that pretends to know everything. When we run innovation programmes, the teams that trust their sponsors most are rarely the ones with the glossiest strategy decks. They are the ones whose sponsors say, this is our current best thinking, here is what would change our mind.

The second is to get out into the organisation. Leaders cannot build trust from the executive floor. They have to be in the rooms where the work happens, asking what people are actually doing with these tools, where they are stuck, and what they are afraid of. This is slow and unglamorous. It also produces the one thing a survey cannot: a leader who has seen the real friction and can speak about it from experience rather than from a slide.

The third is to invest in people, and here the research is pointed. The authors argue for sustained capability building, not one-time courses. I agree with this more strongly than almost anything else in the piece. A single workshop, however good, gives people a taste of what is possible and then returns them to a job that has not changed. Skills decay without practice. Confidence decays faster. If the organisation is serious about people being able to work well with AI, it needs to fund the long middle part: time to experiment, peer learning, coaching on real work, and a clear signal that building these skills is part of the job and not an extracurricular. This is where our own AI training work has shifted over the past few years, from delivering sessions to designing programmes that run for months and are tied to actual workflows. For leaders who want to see how that structure looks in practice, our AI training programme is a reasonable place to start.

The fourth is to equip a range of leaders. Trust is not built by the CEO alone. It is built, or eroded, by the direct manager, the team lead, the project sponsor and the person in HR who fields the difficult question. If only the top tier is briefed and the layers below are left to improvise, the message will fragment, and middle managers, already the most anxious group in the survey, will be asked to carry a story they do not believe. Equipping them means giving them real information, honest answers to the awkward questions, and permission to say, I don't know yet, I will find out.

What this means for the people running the programme

If you lead learning and development, innovation or digital transformation, you are likely the person asked to make an AI rollout land. That puts you in a delicate position. You will be pressured to produce a message that calms people down, and you will know that a calming message that overpromises is worse than silence.

My advice is to push back on the request for reassurance and offer something better. Offer a commitment to transparency instead. Tell people what is known, what is not, and when they will hear more. Build a cadence of honest updates, even when the update is that nothing has been decided. Treat anxiety as information about where the plan is unclear rather than as a morale issue to be managed away.

Also resist the pull towards a single, high-visibility training event. It photographs well and reports well, and it does little to change how people work in March. The more durable approach is smaller and longer: cohorts working on their own tasks, managers who have been through it themselves, and a steady flow of examples from inside the organisation that show what good looks like. That builds both skill and trust, and the research suggests the two feed each other.

Finally, measure the right thing. Adoption figures and licence activity tell you whether people have logged in. They do not tell you whether people believe their leaders will be straight with them when roles change. A short, honest pulse question about that belief, asked repeatedly and acted upon, will tell you more about your AI readiness than most technology dashboards.

The uncomfortable truth in the McKinsey findings is that trust cannot be announced into existence. It is accumulated through small, consistent acts of candour, and it can be spent in a single careless sentence. Most organisations begin an AI transformation with a generous balance, since nearly four in five employees still trust their employer to do right by them. The task is to spend it wisely. That means telling people the truth, giving them real capability, and leaving the empty promises out of the all-hands.