Yes, with limits. An owner can almost always require a standard of work, such as quotes out within a day. Requiring staff to use AI itself changes how the job is done. In the UK that can be a change to contract terms, which needs consultation and usually agreement. In the US most employers have wide discretion, but disability and religious accommodation still apply.
This is a way of thinking, not legal advice. Employment law differs by country and, in the US, by state, and a union agreement or your own contracts can change the answer. Take the details of any change to an employment lawyer or, in the UK, to Acas.
What is the difference between asking for a result and requiring a method?
Asking for a result sets a standard of work, while requiring a method dictates how the work is done, and both lawyers and staff treat the second far more carefully. Most owners who ask whether they can make people use AI have a result in mind: faster replies, fewer errors, quotes out before the competition's. The trouble starts when the result gets translated into an instruction about tools.
I use one line to keep the two apart.
The Method Line is the point where an owner stops asking for a result and starts requiring a way of working. Asking for a result, such as faster quotes or fewer errors, needs no special care. Requiring that a task be done with AI crosses the line, and it should happen only for a named task, with a written reason, after training and a trial, and with a plan for anyone who cannot or will not.
Here are two sentences on either side of it. "Every maintenance request gets a reply within one working day" is a result. "Every maintenance reply must be drafted in the approved AI assistant" is a method. The first leaves the how to the person and survives a change of tools. The second ties the job to a product, and that is where contract, consultation and accommodation questions begin. The Method Line is my suggested way of thinking, not research.
Why is this question coming up now?
Owners are asking because AI use is spreading faster than the plans that should accompany it. In Gallup's US employee data, labeled May 2026 and drawn from its panel of US workers aged 18 and over, only 25% said their organization had communicated a clear plan for integrating AI into current practices. In Great Britain, an Acas survey by YouGov (1,023 employees, fieldwork 27 March to 1 April 2025) found that the biggest single concern about AI at work was job losses, at 26%.
Those two findings come from different countries and different years, so I would not compare the numbers. Read side by side, they describe the situation a requirement lands in: a team that has often not been told the plan, and that worries about its jobs. A rule delivered into that gap tends to produce quiet compliance rather than skill. My own guide to getting employees to use AI covers why mandates alone make resistance quieter, not smaller. This article covers the question underneath it: what you are entitled to ask.
What can a UK employer require?
In the UK, anything that changes an employee's terms generally needs the employee's agreement, and whether a new AI requirement counts as a change to terms depends on the contract and on how much the job changes. Acas's own advice on AI at work, published 15 May 2025, says that where employers expect certain roles to start using AI, this "could mean a change of terms and conditions", and that employers should consult workers and any representatives on its introduction. The word "could" is the honest answer. It turns on what the contract says and how far the day-to-day job moves.
Acas's page on changing contracts, last updated 20 October 2026 and checked on 25 October, sets out the usual order. Plan first and check that a change is really needed, consult the people affected, and seek agreement: "You must both agree to the changes", unless the contract contains a clause that lets the employer change terms. It names the risks of getting it wrong as claims for breach of contract or constructive dismissal, and it tells employers to avoid fire and rehire.
Two dates matter for the next year. GOV.UK's employment rights timeline, last updated 22 October 2026, says the qualifying period for unfair dismissal falls from two years to six months from 1 January 2027, with the compensation cap removed. The same page lists fire and rehire protections for March 2027, while an Acas page updated two days earlier expects January 2027. I would plan for the earlier date. The practical point is the same either way: a requirement you can only enforce by dismissal is getting riskier to impose. Consultation and a trial are the cheaper route.
What can a US employer require?
In most US states an employer has wide discretion over how work is done, so requiring AI for a task is usually lawful, but accommodation duties and unlawful-reason limits still apply. FindLaw's explainer, published 14 September 2026, says workplace AI use is largely up to employers, that in at-will states (which it describes as almost all) an employer can dismiss for almost any reason that is not unlawful or against public policy, and that it found no case law yet on firing someone solely for not using AI.
The limits are the familiar ones. Under the Americans with Disabilities Act, a worker who cannot use a tool because of a disability may be entitled to a reasonable accommodation unless it causes undue hardship, and a sincerely held religious belief can raise a similar request. The EEOC's enforcement guidance, issued in 2002 and still on its site, says an employer need not eliminate an essential function of a job, and that a request for accommodation starts an interactive process between the worker and the employer.
My reading, which is not legal advice: if you decide AI use really is a core part of a role, say so accurately in the job description. If it is not core, an accommodation conversation is simpler. Either way, a person who says "I can't" deserves a conversation before a warning. State law varies, so check yours.
Which of the four asks fits your business?
Most businesses are best served by the second ask and, once they have evidence, the third. The table below is my judgment, not a research finding, and the legal column reflects the sources above as I checked them on 25 October 2026.
| Ask | What you say | Where it sits | Use it when | Main risk |
|---|---|---|---|---|
| 1. Invite | "Try it and tell us what works." | Below the line. No change to the job. | You are starting, or the tool is new to everyone. | Nothing changes, and the keen few pull away (see the Access Grid). |
| 2. Expect a result | "Quotes go out within one working day." | Below the line. A standard of work. | The result is measurable and the people closest to the work agree it is fair. | People use AI quietly with no rules, which is shadow AI. Pair it with a policy. |
| 3. Require for a named task | "First drafts of maintenance replies are written in the approved assistant, then checked by you." | Across the line, narrowly. UK: consult, and ask whether terms change. US: employer discretion, accommodation duties. | You have trained people, run a trial and can show it helps. | Accuracy and accountability if the check is a signature. Accommodation requests. |
| 4. Require for the whole role | "All coordinators must use AI in their work." | Across the line, widely. UK: likely a change of terms. US: job description and accommodation questions. | Rarely. Usually only when the role itself is being redesigned. | The standard is vague, so enforcement is arbitrary. Highest legal and morale risk. |
The verdict: start at row 2. Move to row 3 only for a task you can name and a trial you can show. Treat row 4 as a role redesign, with advice, not as a memo.
What does the Method Line look like in a real business?
Take a hypothetical 24-person lettings and property management firm. Three coordinators have started drafting tenant letters with an AI assistant and finish them faster. The owner wants everyone to do the same and is tempted to announce that AI is now compulsory.
Using the Method Line, she starts below it. She writes the result she wants in a sentence a tenant would notice: every maintenance request gets a clear reply within one working day. She asks the three coordinators to show their work and notes what each one checks before sending. Nobody is required to change anything yet.
Six weeks later she is ready to consider crossing the line, for one task only: first drafts of routine maintenance replies. She writes the reason on one page (replies are slow and uneven, and the three coordinators have shown a way that works) and holds a training session on the approved assistant and what must never be pasted in. She runs a six-week trial with volunteers and uses the Friday Four conversation to hear what went wrong.
Then she plans for the people the rule could hurt. A coordinator who uses a screen reader may find the assistant awkward, so the task has an alternative route. A coordinator who simply does not want to is asked which chore they would give away instead, which is the Chore Rule from my guide on adoption. Only after that does she write the expectation into role descriptions, after consulting the team. In the UK she would take advice on whether this changes terms. In the US she would update the job descriptions and tell people.
Everything she did before announcing a requirement was reversible. That is the point of the order.
What I See When I Train Business Teams
Across the business teams I train, the mandate nearly always arrives before the plan. An owner sees one person save an afternoon and wants everyone doing it by Friday. What follows is rarely open refusal. It is compliance on paper: logins rise and the work stays the same.
Owners tend to ask this as a yes or no question. In my experience it is a sequence, and the order matters more than the answer. Ask for the result. Learn from the people already getting it. Name one task. Train and trial it. Make a plan for anyone who cannot or will not. Then, and only then, decide whether to require it. Measure hours returned, not logins, using the Before Number Rule, so you can tell whether the requirement earned its place.
What should you do this week?
- Write the result you want in one sentence that a customer would notice. If it says "use AI", it is a method, so rewrite it.
- Ask the people already ahead to show you how they work and what they check before anything leaves the building. Do not announce anything.
- Answer four questions for the task you have in mind. Is it named in anyone's job description or contract? Who would it change? Could someone who cannot use AI do it another way? What will you do if the trial fails?
- Choose one task and run a six-week trial with volunteers. Review it in a Friday Four conversation each week.
- If you then want to require it, write a one-page note covering the task, the reason, the training and support, the exceptions and a review date. In the UK, consult before it takes effect and take advice on whether terms change. In the US, update the job description and take advice on accommodation.
- Check that your AI policy says what people may do before it says what they must, and that the training exists before the requirement does. If you have no policy yet, the UK policy template is a place to start.
Where to go next
If you want people to use AI because it helps them, not because a memo said so, that is what my AI training for teams is built for. If you want to know where your business stands before you decide what to ask of anyone, take the free Business AI Readiness Scorecard. And if you are planning an event where owners will face exactly this decision, my business keynotes are built for that room.
Sources and further reading
- Acas, Changing your employees' contracts: considering changes, web guidance, last updated 20 October 2026 (Great Britain; checked 25 October 2026).
- Acas, One third of employers think AI will increase productivity, press release, 15 May 2025 (Great Britain; YouGov survey of 1,015 senior decision-makers). Source of the advice on consultation and changes to terms and conditions.
- Acas, 1 in 4 workers worry that AI will lead to job losses, press release, 28 April 2025 (Great Britain; YouGov survey of 1,023 employees, fieldwork 27 March to 1 April 2025).
- Acas, Employment Rights Act 2025, web guidance, last updated 20 October 2026 (checked 25 October 2026).
- GOV.UK, Plan to Make Work Pay and Employment Rights Act: timeline update, last updated 22 October 2026 (England, Scotland and Wales; checked 25 October 2026).
- Gallup, Indicator: Artificial Intelligence, figures labeled May 2026 (United States; employees aged 18 and over working full or part time for US organizations, Gallup Panel).
- FindLaw, Can your employer make you use AI at work?, published 14 September 2026 (United States; legal information article, not a primary source of law).
- U.S. Equal Employment Opportunity Commission, Enforcement Guidance on Reasonable Accommodation and Undue Hardship under the ADA, issued 17 October 2002 (United States; guidance, not law).
- On this site: How to get employees to use AI, Who is left behind on AI in your team? and How to measure whether AI is working.
Dan Fitzpatrick, The AI Educator, is a Forbes contributor, international keynote speaker and bestselling author who has trained more than 150,000 people across 30+ countries. About Dan
