Introducing AI without damaging employee trust starts with explaining what is changing, what remains undecided and how people can influence the work. Employees need practical support, clear information about data use and a credible way to raise concerns. Managers then need to act on what they hear.

Communication cannot guarantee trust. A reassuring launch message will not compensate for unexplained monitoring, unrealistic workload expectations or promises that leaders cannot keep. The aim should be to make decisions understandable and commitments open to scrutiny.

I would treat communication and listening as part of designing the change. The following proposed sequence gives an HR Director and leadership team a starting point, with examples to adapt to decisions they have actually made.

Before the announcement: agree what can honestly be said

Bring the executive sponsor, HR, the operational owner and relevant data specialists together before asking managers to brief their teams. Agree the problem being addressed, the initial scope and which decisions are still open.

Separate the trial from the longer-term strategy. Testing assistance with routine enquiries is a defined activity. Redesigning a service department is a different decision, with different consequences for people. Employees should be able to understand which discussion is taking place.

Prepare managers for questions about jobs and workload. If role changes are being considered, do not hide that possibility behind a general promise that AI only helps people. If no decision has been made, say what remains uncertain and who will decide. Avoid assurances about permanent job security that nobody is authorised to give.

The argument in AI Should Augment People, Not Just Reduce Headcount helps leaders consider a broader purpose for change. That purpose still needs to be reflected in actual resource and staffing decisions.

For a hypothetical enquiry-drafting pilot, a manager could use this opening, provided it accurately describes the agreed plan:

Example manager message: “We are testing whether an approved AI assistant can help prepare routine enquiry responses. People will continue to review replies. We do not yet know whether this will reduce the total workload, because checking and correction take time too. At the end of the pilot, we will review quality, workload and your feedback before deciding the next step.”

Follow the message with the actual review date and accountable manager. An unresolved question needs an owner and a time for an update, even when the eventual answer is not yet available.

Before designing the trial: listen to different experiences

Invite people who perform the work to describe its difficult parts. Include experienced staff, newer colleagues, frequent AI users and people who have chosen not to use it. Ask about exceptions and informal coordination that may be missing from the process diagram.

The UK Government Communications guide The People Factor recommends feedback from both users and non-users, alongside visible action on what they report. It is guidance developed around public-sector generative AI adoption; the communication sequence here is my proposed approach for a company.

Give staff something concrete to examine: a proposed task, an approved example and the intended handover. A general invitation to “share your thoughts on AI” is harder to answer usefully.

Example listening questions: “Which part of this task looks simpler than it really is?” “Where could a plausible answer cause trouble?” “What would you need to check the work properly?” “Which part of the proposed change concerns you most?”

Record the issues without labelling every objection as resistance. A concern may reveal a data gap, an accessibility problem, extra review work or a legitimate disagreement about priorities. Staff may also question an assumption that leaders have not examined.

Explain which aspects participants can change. If the tool has already been selected, be clear about that decision while inviting input on the workflow and support. Do not present a consultation as an open choice when its main outcome has already been fixed.

Before first use: explain data, expectations and support

Show employees what information the tool receives, what records the organisation keeps and who can access them. Explain the intended purpose of those records and the applicable retention arrangements in language people can understand. Confirm the answers with the responsible owners before circulating them.

Pay particular attention to employee assessment. If usage information, generated summaries or other AI outputs may inform decisions about people, explain the proposed use and how it can be questioned. Have the relevant privacy, HR and other specialists examine the arrangement for the actual setting.

I would keep security monitoring, service improvement and individual performance assessment as explicit, separate purposes. A log collected to investigate errors should not quietly become a league table of employee enthusiasm. Do not promise anonymity unless the collection method and reporting genuinely provide it.

Make training part of the working plan. Provide approved material, practice time, a route for basic questions and support for different starting levels. Managers need to know which existing tasks can move while people learn. For marketing departments, the CMO’s preparation plan connects these responsibilities to everyday team management.

A second message can clarify expectations. This example assumes the organisation has made the stated decisions:

Example manager message: “The first practice session uses fictional enquiries, not customer records. We will explain what the tool stores and what our team records before you use it. Bring an example of an output you would reject. Asking for help or challenging a suggestion is part of learning the process.”

State whether participation in each activity is optional or required. An employee following an assigned workflow has not necessarily endorsed every aspect of the change.

During the trial: make reporting useful and safe

Provide more than one route for feedback. Team discussions can reveal shared problems, while a private conversation with a manager or HR may suit concerns about workload, treatment or confidentiality. Use established employee representation channels where available.

Distinguish an urgent operational incident from a suggestion for improvement. People should know how to report a suspected data disclosure or harmful output promptly, as well as where to suggest a clearer instruction. Do not ask them to reproduce confidential details in a general team channel.

The operational owner should confirm receipt, arrange the necessary response and tell the reporter what happens next. Managers should avoid promising an outcome before the issue has been investigated. They can still explain who is responsible and when an update will be provided.

Example listening questions during the trial: “What checking work has been added?” “Have you felt pressure to accept an answer you doubted?” “Was it clear where to raise the problem?” “What happened after you reported it?”

Protect time to discuss difficult examples. If meetings only celebrate impressive outputs, staff have little reason to bring a failure that complicates the story. Recognise careful judgement, useful challenges and corrections alongside successful use.

At the review: show what changed because people spoke

Return to the questions raised at the start. Explain what changed, what remains unresolved and what will not change, with reasons. A simple written update can connect each material concern to an action, owner and next review point.

For example, if the hypothetical pilot exposed a review bottleneck, a truthful follow-up could be:

Example follow-up message: “You told us that checking drafts was interrupting other work. We have reduced the trial volume and allocated review time. We will check whether that resolves the problem before increasing use. Your concern about how longer-term responsibilities might change remains open; HR and the service director will address it at the next briefing.”

Separate adoption measures from questions about trust. Account activation, logins and completed tasks show forms of activity. They do not establish confidence in leadership, agreement with the purpose or consent to additional uses of employee data.

Ask people whether the explanation matches their experience, whether they can challenge an output and whether managers respond to concerns. Offer a suitable private route as well as group discussion. In a small team, avoid reporting combinations of answers that make individuals easy to identify.

Treat silence cautiously. It may mean there is little to add, but it may also mean people are busy, uncertain or reluctant to speak. Seek context before announcing that the team is comfortable with the change.

Make the next commitment specific

Begin with one workflow and one communication cycle that leaders can support properly. Name the sponsor, manager and feedback owner; agree the questions they must answer before staff begin; then schedule the follow-up.

If the organisation cannot explain how employee information will be used or respond to material concerns, resolve those gaps before extending the trial. Trust is something employees judge through their experience. Leaders can make that experience more credible by explaining decisions honestly and doing what they said they would do.