When an employee says they do not want to use AI, establish what they are being asked to do, hear the specific objection and agree how the work will proceed. That may mean support, a different method, a pause while a risk is resolved or a decision from someone with the authority to make it. The conversation should leave the employee with an answer or a named person who owes them one.
For a line manager in a larger organization, the difficult part is often the unresolved expectation. You may be encouraging experimentation while the employee hears a new job requirement. You may also be hearing a concern that a demonstration cannot settle. Keep those questions visible before deciding what to ask of the person.
Identify the disputed work and the actual requirement
Start with the task, rather than asking whether the employee is “for” or “against” AI. Drafting a customer response, recording a child's development and generating a marketing image raise different questions. An objection to one use need not describe the person's position on every tool.
Establish these details together:
- The requested work: What input would enter which approved tool, and what output would someone use?
- The required result: What must be accurate, complete or checked before that output can be used?
- The expectation: Is this an optional experiment, a suggested method or an explicitly established part of the work?
- The authority: Who can confirm the requirement, approve the inputs or decide on another method?
Do not turn a suggestion into a requirement during the conversation. If your organization has not answered the expectation question, take it to the responsible leader. The employee AI acceptable-use policy guide explains the rules employees need about permitted tools, data and uses.
In its UK workplace AI guidance, Acas recommends clear policies and consultation with employees and their representatives. It also advises checking which tools and data are permitted and reviewing outputs. An instruction to use AI needs that practical context.
Hear the objection without making agreement the price of being heard
Ask what the employee is concerned about in this particular use. Listen far enough to distinguish a problem they have encountered from a prediction, a missing rule or an ethical position. Ask what would help answer the concern, while allowing that their position may remain unchanged.
A March 2026 Reddit discussion makes that distinction concrete. A Reddit user assigned to create Copilot training videos for an organization of more than 1,000 employees worried that encouraging use could support staffing cuts. In a follow-up, they separated their training work from decisions executives might make. The account describes a fear, not a verified layoff outcome. Another tool demonstration would not answer the staffing question.
In the same thread, a self-described childcare worker who generally supported AI objected to using it for documentation intended to convey genuine observations of children to families. The useful detail is the boundary around the task. Enthusiasm for AI elsewhere did not make this use acceptable to them.
Use the person's stated reason in your notes. “Concerned about putting customer records into this account” is something a data owner can investigate. “Negative about change” loses the question you need answered. If you also suspect a skills or access problem, use the guide to distinguishing capability, reluctance and opportunity rather than assuming the objection explains everything.
An ethical disagreement can remain even when utility is acknowledged. In a September 2026 account on X, the creator motionpunk recognized that AI tools could help their 3D work but objected to replacing other people's work. That is a stated personal position, not evidence of an employer's outcome. It illustrates why explaining a benefit may leave the underlying objection unresolved.
Choose a response that addresses the question
The next step should follow from what is unresolved. Avoid prescribing training to every person who declines, or treating every concern as permission to abandon an established responsibility.
| What remains unresolved | A proportionate next step | What to check before moving on |
|---|---|---|
Whether the customer data may enter the tool | Keep the existing permitted process while the data owner decides | The approved inputs, account and handling rules |
Whether the output meets the task's standard | Examine the problem with the person; change or stop this use if it cannot meet the standard | The original record, error and checking effort |
Whether the employee can perform the approved task | Arrange relevant practice and an accessible way to do the work | The specific step needing support, including review |
Whether another method can deliver the required result | Compare the permitted methods fairly | Quality, effort and any genuine process requirement |
Whether the proposed use conflicts with an ethical concern | Record the concern accurately and seek an accountable organizational response | The decision, its reasons and any available alternative |
Whether AI use is actually required | Ask the responsible leader and HR to clarify the expectation | The role, policy, local obligations and support arrangements |
A consultation should change what the organization knows, even when it does not produce agreement. Acas describes consultation as talking and listening about changes, considering views and exploring solutions. It distinguishes consultation from collective bargaining; listening does not automatically give either participant authority to settle every question.
There is a reason to take that process seriously, without promising a result. The OECD's 2023 workplace AI surveys found an association between consultation and more favorable reported outcomes. The finance and manufacturing surveys were conducted in 2022, asked about new technologies and could not establish causality. They support investigating employee experience, not claiming that one conversation guarantees adoption.
Leave with a work arrangement that can be reviewed
Suppose Jamie Lee manages customer service at a regional furniture retailer. Adviser Claire Bennett is asked to use an approved AI tool to draft replies from complaint notes. She objects to putting identifiable customer details into it because she has not been told which records are permitted.
Jamie and Claire agree that the reply must reflect the complaint accurately, preserve the promised remedy and be checked before sending. That standard already applies to replies written using the existing template. Jamie does not ask Claire to upload a live complaint to prove the tool is safe.
The conversation can end with a short record like this:
| Record | Jamie and Claire's agreement |
|---|---|
Task and standard | Reply to a furniture complaint accurately, including the remedy already promised |
Concern and evidence | Claire has no confirmed rule permitting these identifiable notes in this AI account |
Work for now | Continue the existing permitted reply process and its normal review |
Question and decision owner | The data owner must confirm which complaint information can enter the approved account |
Next update and check | Jamie will return with an answer on Friday; review the permitted method and a checked reply together |
Friday is an update commitment, not a promise that approval will arrive. If the decision is still open, Jamie should explain the delay, confirm the interim process and agree the next update. Once the rule is clear, they can discuss a suitable practice example and the actual work expectation. The purpose is to resolve the missing condition before judging Claire's response.
Use only the information needed to explain the issue. A concern record does not need copies of confidential complaints or personal details unrelated to the decision.
Follow through when disagreement remains
A manager can listen carefully and still reach an unresolved disagreement. Preserve what the employee said, explain which decision you can make and take the remaining question to the right owner. Where an employment or accommodation question arises, involve HR and qualified local advice through your organization's process before drawing conclusions about consequences.
Do not promise that every AI use is optional. Equally, a broad adoption target does not answer whether a particular input or task is appropriate. Explain the actual requirement and its reasons, the available support and any alternative that has been considered. Give the employee a clear route to raise an unresolved concern.
For continuing work, the weekly manager review can track the agreed action and examine a real output. Keep the personal disagreement in an appropriate conversation rather than making the employee defend it in front of the team.
The people and change guide connects this agreement to the support and working conditions employees need.
Questions and answers
Does every employee have to use AI?
There is no single workplace rule that makes every AI tool compulsory for every employee. The answer depends on the actual role, task, employer policy and applicable local obligations. A manager should establish whether the disputed use is optional or required, who set that expectation and what support or alternative is available.
Do not infer a requirement from a general ambition to increase adoption. Start with the task and authority questions, and involve HR when the expectation has employment implications.
How should a manager respond to an ethical objection to AI?
Ask the employee to explain the concern in relation to the proposed work, record it accurately and identify who can give an organizational response. The employee may acknowledge the tool's usefulness while remaining concerned about its effects. A demonstration can clarify a misunderstanding, but it cannot settle every ethical disagreement.
Agree how the work will proceed while the question is considered, including any permitted alternative. The response table helps separate an ethical objection from a data, quality or support question without dismissing any of them.
Should an employee who refuses AI receive more training?
More training is useful when the approved task includes a specific skill the employee needs to practise. It will not resolve an unapproved input, an unsuitable output or a decision about how the organization will use the work. Ask which part of the task is difficult and which part the employee objects to before choosing support.
Where ability and working conditions may both matter, use the guide to investigating an AI adoption barrier. Arrange practice that addresses the identified step rather than repeating a general introduction.
What if the employee still declines after the concern has been discussed?
Clarify what remains disputed and whether the required work can be completed through a permitted alternative. Explain the decision and support you can offer, and refer an unresolved employment question to HR and qualified local advice. A conversation, policy headline or usage target alone does not establish the appropriate consequence.
Retain the agreed interim arrangement, decision owner and next update. The work-agreement example shows what to record while a question remains open.
