A New Kind of Third Party: When AI Sits In on Trust-Sensitive Communication
More professional exchanges now involve a third presence that never speaks.
AI may help draft an email, soften a demand, translate a term, check tone, or pressure-test a position before anything is sent. It may shape a single sentence or influence much more.
The other side may never know it was there, or may find out.
This is still new enough that the evidence remains mixed. Two 2025 studies illustrate the tension.
One finding: disclosing AI use can reduce trust.
Across thirteen separate experiments, people who disclosed using AI in work tasks were consistently trusted less than people who did not disclose doing so, even though disclosure is often promoted as the more transparent or ethical course.
Another finding: in some settings, disclosure may matter less than assumed.
A separate study, using incentivized trust-game exchanges between participants, found that writing with AI assistance had little to no effect on trust overall, including when AI use was disclosed. The authors note that their findings are particularly relevant to one-shot, transactional interactions.
The gap between these findings may reflect differences in context, in what kind of trust was being measured, and in study design. At this stage, the evidence does not yet allow a confident single explanation.
What this means for now.
Whether to mention AI assistance in trust-sensitive communication is not yet a settled question, and treating it as settled goes further than the current evidence supports.
Message quality and substance still matter, but the research suggests that knowledge of AI involvement can, in some settings, affect how the person behind the message is perceived.
Where trust is already fragile, decisions about what to disclose should be made carefully, rather than assuming that either silence or complete transparency is always the right approach.
Sources
Schilke, O., & Reimann, M. (2025). The transparency dilemma: How AI disclosure erodes trust. Organizational Behavior and Human Decision Processes, 188, 104405. https://doi.org/10.1016/j.obhdp.2025.104405
Purcell, Z. A., Jakesch, M., Dong, M., Nussberger, A.-M., & Köbis, N. (2025). Writing with AI boosts trust-building efficiency. iScience, 28(12), 114092. https://doi.org/10.1016/j.isci.2025.114092
A working document — not legal advice. Use alongside counsel.