We have all been there. Listening to someone, and most often a business leader such as a CEO, who is physically present but communicatively absent. It feels like they are reading from a script they do not, or have forgotten, to believe in. It doesn’t feel real. But since we are in the same room, we can’t just get up and walk out, or change the channel, so we tune out. The clever messages disappear like particles caught in the rays of the sun, out of grasp and forgotten.
That communicative absence now has a number attached to it.
A Harvard Business Review analysis of more than 2,400 CEO communications tracked across eighteen months found that homogenized, AI-generated executive messaging correlates with a measurable and accelerating erosion of trust. Thirty-eight percent within ninety days. The number sounds clinical until you understand the mechanism behind it, which is anything but algorithmic. At first glance, I thought that it was because audiences are not detecting AI. It is not so much AI as it is them detecting the absence of a person. This distinction matters more than the statistic.
Aristotle devoted considerable effort to a concept most communication schools have either retired or conveniently ignored: ethos, which is the felt sense that the person speaking is present in the act of speaking, that something is at stake for them in the words they choose. You cannot fake ethos. You can simulate its markers such as warmth, directness, apparent vulnerability — but simulation is precisely what audiences have become extraordinarily sensitive to, even when they lack the vocabulary to name it.
The Harvard Business School researchers behind the Wade Test offered a useful frame for what is actually happening. They trained a generative AI system on a CEO’s prior communications and asked employees to identify which responses came from the CEO and which came from the machine. The machine passed 59% of the time, barely above random chance. What I found most interesting is what happened to messages that employees believed were AI-generated, regardless of whether they actually were. Those messages were rated as significantly less helpful, less credible, and less worth acting on. Perception alone was enough to blunt the communication’s function.
There is a name in academic communication studies for what large language models do to voice at scale. Language homogeneity. The flattening of idiosyncratic expression into a statistical average of every credible-sounding thing ever written. Words like “articulate,” “navigate,” “robust,” and “unwavering commitment” saturate executive communications now because they sit at the peak of the probability distribution for professional language. They are the most likely words. They are also the least alive.
When I encounter a CEO message (or anyone’s) that uses the phrase “navigate uncertainty” followed immediately by “resilient teams” and “shared values,” I know that I am not reading a person but rather the mean. This, as any statistician will confirm, is a number that often describes no actual individual at all.
The replicant effect, as researchers in the field have begun calling it, compounds the damage. When audiences encounter enough homogenized content in a shared ecosystem (LinkedIn, earnings calls, internal all-hands transcripts are all perfect examples, but there are many more), trust that was often teetering to begin with evaporates. not just in the AI-generated messages but in the human ones nearby. A single voice that sounds like the output of a language model recalibrates expectations for everything around it in a sort of verbal contagion. The cost, unfortunately, is not borne solely by the individual CEO. It is to the communicative environment they inhabit.
APCO’s research across the UK, Germany, France, and Italy sharpened this into a number that should concern every board: only 9% of respondents said they had been genuinely impressed or influenced by a business leader ( let me underline, ONLY 9%). Fifty-four percent believe executive statements shape company perception more than advertising. Thirty-one percent find executive communications authentic. These numbers point to a structural problem more than a perception one, luckily, because structure can usually be changed more easily than perception.
The communications failure here is not the use of AI. The failure is the substitution of AI for the irreplaceable act of a leader committing to a position in their own words.
The philosopher Byung-Chul Han (for anyone that reads me regularly, yes, him again) writes about the transparent society: a world so saturated with information that nothing retains opacity, and therefore nothing retains depth. CEO communications have entered that transparency trap from the opposite direction. They have become too frictionless. Every sharp edge has been polished away. Every uncomfortable silence has been filled. And the audiences receiving those communications experience the result not as clarity but as absence.
Voice authenticity is not a stylistic preference. It is a strategic infrastructure question. The organizations that understand this are building what I think of as voice architecture: a documented, stress-tested framework that captures not just tone guidelines but the actual reasoning patterns, failure modes, and idiosyncratic positions of the human being doing the communicating. Why is this so important? So that AI cannot accidentally overwrite it.
Cerence, a small automotive AI firm, held its earnings call using fully AI-generated voice clones of its CEO and CFO. The investor backlash was immediate and visceral. The company had, in effect, announced that the humans responsible for its direction were absent from the most important communication of the quarter. The technology performed exactly as designed. That was the problem. I know this is an extreme, but it is telling nonetheless.
The executives who are getting this right are doing something less dramatic and considerably more difficult. They are writing the hard sentences themselves. Crazy right? Actually, the ones who write it all themselves will allow their words to admit uncertainty, take a position a reasonable person could disagree with, and take the risk of being less smooth while introducing the friction of a mind that has not yet resolved everything, are the ones who created the greater connections. These sentences cannot be optimized by a language model because optimization precisely removes them, as they are statistically improbable, which ironically makes them tuned to the human ear and consciousness.
Fifty percent of US CEOs already report using AI for content creation, and the number is expected to rise. The structural question facing every communications leader is not whether to use these tools but where the human responsibility remains non-negotiable. I would argue that the answer is the same today as it was in Aristotle’s Lyceum: at the point where the speaker chooses what they actually believe, and commits to saying it in a way that costs them something.
The 38% erosion of trust figure is a warning about substitution. The moment a leader outsources meaning-making to a system that optimizes for reception rather than truth, the audience detects it.
Voice is not a communication channel. It is a proof of presence. The executives who remember that are the ones whose messages move rooms.
Leave a Reply
You must be logged in to post a comment.