Telling People You Use AI Might Lose You Their Trust

Grading a student's assignment, writing a job application, drafting an investment ad, composing a routine email. Across 13 separate experiments and more than 4,000 participants, Oliver Schilke and Martin Reimann tested what happens when someone says "I used AI for this" versus saying nothing at all. The work was identical in both conditions – only the disclosure changed. And in every context they tested, the person who disclosed was trusted significantly less than the person who didn't.

That's not a hunch or a hot take. It's a peer-reviewed finding published in Organizational Behavior and Human Decision Processes, and it should complicate a piece of advice most of us have been repeating without much scrutiny: when in doubt, be transparent about your AI use, and trust will follow. The research says trust doesn't follow. Not automatically, and not because the AI did a worse job.

It's about who's accountable, not the quality of the work

The researchers ruled out the obvious explanation first. If disclosure hurt trust because people assumed AI-assisted work was lower quality, you'd expect the penalty to shrink when the work was demonstrably good. It didn't. The study's authors trace the effect instead to something they call reduced perceived legitimacy: disclosure signals that the work deviates from the social norms people expect for how professional judgment gets produced, and that deviation itself – independent of quality – is what erodes trust. A secondary write-up of the research frames the same mechanism in plainer terms: role ambiguity. Once you say AI was involved, the evaluator can no longer cleanly locate who's actually responsible for the judgment call in front of them. Was it you, was it the model, or some blend nobody can quite specify? That uncertainty is what people are reacting to, not the output.

Here's the part that should really give you pause if you were hoping to solve this with better wording. In one of the study's experiments, researchers tested six distinct ways of framing AI involvement – including versions that explicitly cast the AI as "just a tool," emphasized human review, or explained the rationale for using it. All six failed to prevent the trust penalty. This wasn't a messaging problem you can write your way out of with the right sentence in your email signature.

It's also not algorithm aversion – which is the strange part

If you assumed this was really just people distrusting AI in general, the study has an answer for that too, and it cuts the other way. People trusted a fully autonomous AI agent, acting entirely on its own with no human in the loop, more than they trusted a human who disclosed using AI as a collaborator. The penalty isn't about the presence of AI; it's specifically about the ambiguity of a human-AI hybrid – the moment where you can't tell whose judgment you're actually relying on.

That reframes the whole problem for anyone doing advisory or consulting work with AI in the loop. The risk isn't "my client will find out I use AI and think less of my work." The risk is specifically in how disclosure is delivered, because a poorly framed disclosure creates exactly the accountability fog the research says people penalize.

This is not a case for staying quiet

I want to be precise here, because the tempting, wrong reading of this research is "so don't tell anyone." The study itself closes that door. One of its later experiments found that when AI use was exposed by a third party rather than disclosed voluntarily by the person who used it, trust dropped even further than it did with voluntary disclosure. Getting caught is worse than owning it. Disclosure remains the right call, professionally and often legally – the finding isn't that transparency is bad, it's that the naive, defensive, one-line version of transparency backfires, and silence is a worse bet still.

What the research actually supports is a distinction between whether you disclose and how you disclose. A one-off, apologetic caveat bolted onto a deliverable – "just a note, I used AI for part of this" – is precisely the framing the study tested and found ineffective, in six different variations. What it didn't test, and what remains a reasonable hypothesis rather than a proven fix, is disclosure at the level of how you run your practice: a standing, upfront description of where and how AI fits into your process, established before any single deliverable is on the table, rather than attached defensively to the output itself. That's a different disclosure altogether – closer to how a firm describes its methodology than how an individual explains an anomaly.

What this means if you run a boutique practice

If you're a boutique advisor or an independent consultant, this research lands somewhere you can't route around: your entire value proposition rests on being the trusted judgment behind the deliverable. Every AI-washing story so far has been about firms overstating their AI use or hiding weak work behind an AI label – that's already become a legal liability, not just a credibility one, and it's part of why due diligence has had to adapt to catching AI-washing in the first place. This research points at the opposite failure mode: doing everything right, disclosing honestly, and still taking a trust hit because the disclosure itself introduced ambiguity about who's actually accountable for the work in front of the client.

That's an uncomfortable position to be in, and I don't think there's a clean way out of it yet – the research on what actually works, rather than what fails, isn't there. But knowing which failure mode you're actually managing is worth something. It's not "will my client find out," it's "does my client know exactly who's answering for this judgment call, AI-assisted or not." Get that second part right, and the disclosure question gets a lot less scary. Get it wrong, and no amount of careful wording saves you – the study already checked six of them.

There's also a competitive angle underneath this that's easy to miss. The pressure boutique advisors actually face isn't the Big 4 – it's clients wondering whether they need an outside advisor's judgment at all once AI tools are sitting inside their own building. In that environment, the temptation is to over-communicate your AI use as a differentiator: look how efficient we are, look how current our tooling is. This research is a direct caution against that instinct. The thing that actually differentiates a boutique advisor was never "we also use AI" – every competitor can say the same sentence now. It was always the judgment nobody can quite automate, and every disclosure that muddies who's exercising that judgment chips away at the one thing that was never going to be commoditized.

Where I'd start

None of this means hiding AI use, and none of it means a moratorium on talking about how your practice works. It means being deliberate about the level at which you disclose. A standing description of your methodology, given upfront, as part of how you describe your practice, is a fundamentally different act than a defensive footnote attached to a specific deliverable after the fact – even though both are technically "disclosure." One establishes a norm before any judgment call is on the table. The other raises a question about a judgment call that's already been made. Only one of those seems likely, on this evidence, to leave your accountability legible instead of ambiguous. That's the test I'd apply before writing another disclosure sentence: does this make it clearer who's responsible, or does it just make it clearer that AI was involved?