What Will You Refuse to Automate in Internal Comms?

Chick-fil-A's CEO, Andrew Cathy, told CNBC the company will not put AI voice ordering in its drive-thru lanes, because "we really want that hospitality to be human to human." A company named what it refuses to hand over, and Stephanie Nivinskus of Ragan's Center for AI Strategy, writing in PR Daily, turned that into a question comms teams should take to leadership before employees ask it first. Which parts of the organization will remain human-led?

For internal communicators, deciding what to automate in internal comms comes down to two jobs. Sort your messages into tiers by how much AI each one can carry, then publish that line in plain sentences and confirm the people it was written for read it.

The AI question in internal comms has changed

Most AI governance work so far has covered approved tools and data access. The sharper question now is what you will not automate, and for an IC team it lands twice, once for the messages you send and again as the line leadership has to explain to the workforce.

The research on how employees read AI-written messages shows why the line matters. In a study of 1,100 working professionals by researchers Cardon and Coman, covered by Canadian HR Reporter, people rated AI-assisted messages as clear and professional, yet trust perceptions fell as the share of AI writing went up. The University of Florida's release on the study puts numbers on it. Only 40% to 52% of employees viewed supervisors as sincere when they used high levels of AI, compared to 83% for low-assistance messages.

Draw your AI line in three tiers

Nivinskus argues that a short, specific list holds up longer than a sweeping pledge, and the point deserves a spot above your monitor. "We use AI responsibly" commits nobody to anything. Tiers with examples give a writer something to check before sending.

Tier 1 is where AI drafts and a human edits

This is the routine, informational traffic that fills most of an IC calendar, such as event logistics, IT maintenance windows, benefits-deadline reminders, and the weekly newsletter roundup. Linjuan Rita Men of the University of Florida told the same HR Reporter piece that AI on a broad, generic CEO announcement may be received better than AI on a crisis message or an apology, which is the logic of this tier. Let AI draft and a person check it, then spend the saved hours on the tiers below.

Tier 2 is where AI assists and a named human owns it

Policy changes and executive updates can use AI for structure and early drafts, but a specific person approves the final version under their name and can answer questions about it. This tier makes sure someone with authority stands behind every word. Gartner's Design Engaging Executive Messages says traditional executive communications treat employees as passive recipients, and positions the tool as a way for heads of comms to shift toward an action-oriented messaging architecture. The approver should check that the message asks employees to do something specific and that the leader will follow through.

Tier 3 is what you never automate in internal comms

Layoff announcements, condolence and bereavement notes, safety incidents, apologies, performance conversations, and personal recognition belong here. Coman told HR Reporter he "would not encourage" managers to use AI for performance reviews and similar high-stakes discussions, and one respondent in the study asked the question that defines this tier, "If my manager has to use AI to write a message like this, why are they my manager at all?" If employees would feel cheated to learn AI wrote it, the message belongs in tier 3.

Write the line down where employees will see it

A line that lives in the comms team's head protects nobody. Nivinskus recommends writing an approved statement that explains where AI is used and which interactions are human-led, and adding "Where will your company refuse to use AI?" to the briefing materials for the next town hall. Rehearse it with your CEO, because an improvised answer will sound like the sweeping pledge you were trying to avoid.

Keep the statement short enough for employees to repeat, with one line per tier. The stakes for leaders are personal, since 73% of respondents to Mission North's New Rules of Trust survey, as reported by O'Dwyer's, said undisclosed AI executive messaging would reduce their trust.

A policy nobody reads is theater

Publishing the line is half the job. It protects trust only if employees know it exists, and the tier 2 and tier 3 messages protect trust only if they reached the people they were meant for.

Measure reads for each audience group, such as the plant floor or the night shift, on the statement and on every tier 2 and tier 3 message. A healthy company-wide average can hide a site that missed it entirely. Follow up with the groups that didn't read it through their manager or a second channel. Many send tools, including tools like Cerkl Broadcast, can report who read a message by name, and with that report you can verify the commitment you published.

Start with five messages this week

Pull last month's sends and sort five of them into the tiers. You will probably find one that was drafted like tier 1 but belonged in tier 2 or 3, which is the most useful finding of the exercise. Then have your leader approve the statement and send it under their name, and check who read it before the next town hall.

Drawing the AI line is a values decision, and making it hold is a comms job.