# What to tell your teams about AI, and how to say it.

> In this field note, veridive explains how to communicate AI changes to employees: why vague reassurance backfires, what a first message must contain, how managers can answer five common questions, how to talk honestly about jobs with HR and employee representatives involved, and how regular updates on what changed and what was fixed keep trust.

People fill silence with the worst case. Say early and specifically which tasks change, which decisions stay with people, what the time saved is for and what training comes, and give managers words for the questions they will get.

## Key takeaways

- Silence and vague reassurance invite the worst reading; say specifically which tasks change and which decisions stay with people.
- The first message names the workflow, what the system does and doesn’t do, the data it sees, training and who to ask.
- Brief managers first with answers to the hard questions, and never promise anything about jobs you can’t keep.
- Follow up on a regular rhythm: what changed, what people reported and what was fixed.

The rumor usually arrives before the announcement. Someone saw a demo, someone heard “automation” in a budget meeting, and by the time the official email goes out, the team has already decided what it means for them.

Silence invites the worst reading. The answer is not more reassurance but more specifics, earlier: which tasks change, which decisions stay with people, what the time saved is for and what training comes. Then give managers words for the questions they will be asked, because they will be asked first.

## Why does vague reassurance backfire?

“AI will support you, not replace you” can’t be checked, so people check it against the one fact they do have: the company is paying to make work faster. Where does the saved time go? If the message doesn’t say, everyone supplies their own answer.

Vague reassurance also borrows against future trust. If “nothing changes for you” later proves untrue for one team, every other message loses credit. Specific statements are safer because they are smaller. “The assistant drafts replies to delivery-status emails; an agent checks and sends every one” is easy to believe and easy to keep.

## What should the first message say?

Keep it short, and make sure it contains seven facts:

1. **The workflow and team** affected, and when the pilot starts.
2. **What the system does,** in one sentence, and what it doesn’t do.
3. **Which decisions stay with people.**
4. **What data it sees,** and what it doesn’t.
5. **What the time saved is for.**
6. **What training comes,** and when.
7. **Who owns it,** and where questions go.

Illustrative wording for a customer service team:

*“From [month], our team will pilot an assistant that drafts replies to delivery-status emails. It reads the customer’s email and the order record and prepares a reply; you check, edit and send every one. Complaints, refunds and anything unusual stay entirely with you. It sees only the email and the order, under our data protection rules. The aim is less time on routine replies and more on the cases that need judgment. Before the pilot, you’ll have two short sessions on your own emails. Questions go to your team lead or [owner], and we’ll share what we learn as we go.”*

## What should managers be ready to answer?

Brief managers before the message goes out. These are the questions they will hear first, with illustrative talking points:

| Question | What to say | What to avoid |
|---|---|---|
| Will this replace me? | Which of your tasks change, which stay, and what is known for this scope | Promises about the future you can’t keep |
| Will my targets go up? | What the saved time is for, and how targets will be reviewed | Silence, which reads as yes |
| Who checks it, and who is blamed? | A person approves anything that matters; reporting a mistake is welcome | Implying the tool is always right |
| What data does it see? Is it watching me? | The exact sources it reads, and whether its logs are used to assess people | “Only what it needs”, without saying what that is |
| Do I have to use it? | What is expected in the pilot, and how feedback shapes it | Calling it optional if it isn’t |

Two of these need a decision before anyone can answer them. Decide with HR and counsel whether system logs may be used to assess individuals, and say so. Agree what happens to targets before the pilot starts. The answers only hold if they match [how the system is built](https://veridive.com/approach/): people approve what matters, and every step is logged.

## How do you talk about jobs honestly?

Separate tasks from jobs. In a well-designed workflow, AI changes tasks: less copying, searching and first drafting, more checking, exceptions and conversations. Say that plainly, with the tasks named.

Where a role will change significantly, or a piece of work will need fewer people, say so early and plan it with the people affected. Involve HR from the start, and employee representatives where your organization has them, and ask counsel whether consultation or notice obligations apply before anything is announced. Describe the options on the table, such as new responsibilities, retraining or moves to other teams, and when decisions will be made.

> People fill silence with the worst case.

Never promise what you can’t keep. “No one will lose their job because of this pilot” is a promise about one scope. Make it only if it is true, and say what it covers.

## How do you involve the people whose work changes?

Give them real influence, not a feedback form. They know the awkward cases, so they choose examples for the evaluation set and write the answers they would accept. They help design the review step, test it during the pilot and flag what feels wrong, and some become champions for their colleagues. A system people helped shape is one they are more likely to trust, which is where [adoption](https://veridive.com/insights/ai-adoption-needs-a-new-practice/) starts.

## What should follow the first announcement?

A rhythm, not a one-off. During the pilot, a short weekly note from the owner answers three questions: what changed, what people reported and what was fixed. After launch it can move to monthly. Report the uncomfortable parts too: the cases the system gets wrong, the team’s idea that was turned down and why, the pilot that stopped.

Keep a page of the questions people actually ask, with answers, and adjust the training when they reveal a gap. Broader [AI literacy](https://veridive.com/insights/ai-literacy-for-employees/) fits here, once the first team has real examples to share.

## Answers before announcements

Write down your answers to the five manager questions before drafting the message. If an answer is “we don’t know yet”, decide it first or say so honestly. For the leadership view, see the [questions leadership teams ask](https://veridive.com/insights/questions-before-your-first-ai-project/) before a first project; talking change through with teams is part of our [AI enablement](https://veridive.com/services/ai-enablement/) work.

This note is general information, not legal advice.

## Frequently asked questions

### How should you announce AI to employees?

Early, specifically and before rumors do it for you. Name the workflow and team, what the system will prepare and what it won’t do, which decisions stay with people, what data it sees, what the time saved is for, what training comes and who to ask. Brief managers first, so they can answer the questions they hear the same day.

### How do you talk to employees about AI and their jobs?

Honestly, and without promises you can’t keep. Separate the tasks that change from the jobs that change, and say what is known for the current scope. Where roles will change significantly, say so early and plan the transition with the people affected, involving HR and employee representatives where relevant, and ask counsel whether any consultation obligations apply.