AI and Workplace Communication: Keeping the Human Voice

Every organization has a story. Some are deliberate, others accidental or inherited, and some are undirected but form the essence of the organization. Problems arise when intentional narratives become misaligned. Katherine Tuominen calls this narrative debt. She was talking about this phenomenon in a recent Signal Room discussion. She elaborates trust diminishes as the gap between leaders' communications and team's experience widens, and mistrust is incredibly hard to recover.

AI is driving an even larger wedge to this gap, in an area that is still being unmonitored by leaders: everyday communication. Tuominen's argument suggests that, at this stage, AI is primarily a human problem and has a secondary impact as a technical instrument. The first impact is loss of the authenticity of our everyday communications.

Subtle Signs Begin the Narrative Debt

Tuominen noted that early warning signs are usually very subtle. One such subtle clue could be someone leaving a note on a fridge magnet or on a notepad rather than entering the note into the system that they were instructed to use. It is not necessarily a good or a bad thing. Rather it indicates that the communication did not result in the expected action. Although subtle, these signs indicate that there is a need for a deeper evaluation of the unofficial communication methods. This is especially the case when staff members use unofficial methods to communicate issues to the upper management, because staff members believe that the provided solutions do not aid in the daily operations of the organizational activities.

It needs to be understood that this practice rarely comes from a malicious or bad intent. Staff members utilize loopholes because they lack the necessary managerial trust to voice their concerns. The distortion of the boardroom's interpretation of the front-line's reality travels up to the leaders, leaving them to manage a story that no longer reflects the actual state of their operations.

When Two Assistants Talk and Humans Skim

The most common communication breakdown, according to Tuominen, originates in the most mundane of places. Busy individuals rely on AI to quickly compose or revise emails. They do not bother to double-check if the AI has misunderstood the email context. Subsequently, the recipient may send the AI-generated email to their assistant for analysis. You believe you are interacting with a coworker. Unfortunately, as Tuominen articulates, you have two AI systems communicating with each other, and your assistant is confirming the messages on their end.

In the context of a more straightforward, singular, time-bound task, this may be acceptable. However, these small tasks begin to accumulate, and the same small deviations from clarity in the messaging will eventually lead to confusion about the course of a project. It is indeed a translation problem, with the exception that it is not a traditional instance of a model translation with a human. It is an unintentional translation problem between two people, with the machines smoothing over a misunderstanding instead of surfacing it.

The Lexicon Itself is Shifting

Tuominen pointed out a more subtle effect. Because of the pervasiveness of AI, people observe that language is changing and AI is reflecting it back at us. She referenced a publication that pointed out how a term like 'delve', which was not common in the majority of writings outside of the digital space, has begun to appear in spoken communication. AI-generated sentences have begun to influence people's writing and speaking.

She cited comparative phrasing as the most apparent sign. This is the construction "it is not this, it is that." Copywriting tools used to rely heavily on this structure, and now, it is found everywhere. From LinkedIn to texting and emailing, this structure is used because most tasks are completed using ancillary tools. This is not the method of communication most people are accustomed to, and most people are now so accustomed to this structure that they fail to realize it. The concern is not the syntax. The concern is that a team's communication to a leader is no longer informing them on the message's comprehension. The communication is no longer in the leader's control, and the nuances of the team's communication now have a greater impact on the meaning than the words themselves.

The practical test she uses was a welcome change to the high-tech initiatives, and the host also agreed with that idea. Before you send the message, read it out loud. If it does not sound like you, change it. A communication expert is generally able to identify the automated writing in a draft that someone adjacent to it has stopped noticing.

In Health Care, the Risks Are Not Superficial

The most relevant example Tuominen presented was clinical. Without providing nurses with training or a communication framework to understand the resources in front of them, someone could think it harmless to look up a patient's diagnosis or the subsequent steps in an open large language model. What they fail to realize is that these systems are designed to learn from their interactions, and that information can resurface when someone else asks a similar question.

Asking a question quickly does not exist with the considerations of HIPAA, privacy, and confidentiality. Documents of best practices are not enough. The effects are both immediate and long term. The effort is in making the guidance actionable. This means changing the way employees do their jobs by showing them what to ask and what not to ask. This is a much better solution than creating a SOP that no one will implement.

Build the Guardrails with Communicators, Not Just IT

Tuominen's solution talked about a sequence that starts with listening. Before putting AI-assisted communication in place, leaders need to gather ground level feedback (not theorized in an executive meeting) and understand where communication is breaking down. Only then should it be addressed by designing an appropriate infrastructure.

She spoke plainly about a mistake that people are prone to making. They give all of the responsibility of building guardrails to a developer or a person in IT. That person is not always a communicator. Getting AI assisted communication correct requires a lot of collaboration from the people who sit at the contact points in the organization. The people who sit in contact points in the organization will ensure the communication lines stay aligned with the organization's values instead of just saving time. The preliminary work is the cost of admission and is completely feasible from her experience, as long as nobody skips it.

AI as Co-Pilot, Not the Thinker

When asked what one thing she would say to CEOs, Tuominen emphasized the utmost importance of establishing critical thinking around AI enterprise-wide. She suggested AI should serve as the co-pilot, the partner in thinking, and that it should never be the source of thinking. She noted how people still treat AI as a search box, not thinking about the implications of treating AI in that way, due to their lack of understanding as to what a large language model really is.

Tuominen is implying that there is an available market for training, regardless of organizational level. This encompasses the shared enterprise definition of proper tool usage, a critical thinking framework that abstains from tool reliance, and safe-use measures that protect the organization and its stakeholders from unnecessary compliance risk. Tuominen noted that this is a must do, and not when it is convenient. This is not a blanket ban, as people will rely on the tool when it is the easiest option available. The best option is to prepare for tool reliance.

How Hutchins Approaches Workplace Communication and AI

We meet this problem where Tuominen locates it: in the gap between what an organization says and what its people experience day to day. Our work helps health systems put structure around AI use before the habits set — defining acceptable use, the guardrails that protect confidentiality and compliance, and the human review that keeps AI-assisted messaging sounding like the people who are accountable for it. That is inseparable from AI literacy, which gives staff the judgment to use these tools well, and from responsible AI, which provides the governance those judgments are made within. These themes run throughout The Signal Room podcast, where practitioners describe what trustworthy communication and adoption take in practice.

A note on scope: the conversation with Tuominen also touched on the broader trust and leadership dynamics behind AI transformation, which we treat separately. This piece stays on communication specifically — the language layer where over-reliance on AI does its quietest damage.

Authoritative sources

• WHO — Ethics & governance of AI for health
• NIST AI Risk Management Framework
• U.S. Department of Health & Human Services — HIPAA
• HHS Office for Civil Rights
• NEJM AI