To hire an AI operator, choose one repeatable workflow and watch each candidate work on it before you interview them. Judge how they document the steps, check the output and hand the work over, not which AI tools they can name.
If you want a shortlist that has already been tested on real agency tasks, start a Direct Hire brief or explore Managed Teams. Compare both on the pricing page.
Key facts
- UK workplace advisory body Acas says it is a good idea to ask each applicant the same questions where possible. Updated 30 June 2026; checked 10 November 2026.
- Acas describes a scoring system as one fair way to compare applicants, and says the range should be wide enough to tell candidates apart. Updated 30 June 2026; checked 10 November 2026.
- On unpaid work trials, Acas says the time should be reasonable, for example 2 hours or a shift, and not usually longer than one day. It adds that the minimum wage must be paid if the trial is not genuinely for recruitment, for example when the tasks have value to the employer beyond testing skills. Updated 30 June 2026; checked 10 November 2026.
- Shopify's Help Center says you can create and assign roles with the permissions each user needs to complete their tasks. Checked 10 November 2026.
- Acas guidance addresses UK recruitment. Use its practical ideas as a starting point and check the rules for where you and the candidate are based.
What is an AI operator?
In this guide, an AI operator is a person who does recurring agency or ecommerce work with AI tools and is accountable for checking what those tools produce. It is a working definition, not a regulated job title, so write down what your role owns. Our piece on AI agents versus AI-trained operators explains why a person who checks the output matters.
Which workflow should the role own?
Pick one workflow that happens every week, has clear inputs and already has someone who can review the result. Good candidates include a weekly performance summary, first-draft replies for a support inbox, product description drafts, or a pre-send checklist for email campaigns. These are examples, not promises of what any tool will produce.
For the chosen workflow, write down four things before you recruit:
- The inputs the person receives and where they come from.
- The output you expect and who reviews it.
- The decisions that stay with a named person, such as publishing, refunds, budget changes or anything a customer will read.
- How you will tell whether the workflow is working, such as how often a reviewer accepts the output unchanged.
How do you run the work sample?
Give every candidate the same short brief built from fictional or redacted data, and ask for three things: the output, a short note describing how the workflow runs, and a handoff note. Tell candidates in advance how long it takes, which tools they may use and how it will be scored.
A good brief reads like a real ticket. For example: "Here is a fictional week of store orders and support tickets. Draft the weekly summary for the account lead and list anything you could not verify." Let candidates use any AI tool they prefer, then judge the finished work and their explanation of it.
Keep the sample short and never use live client work. Acas notes that a trial can become paid work if the tasks have value to the employer beyond testing skills, so fictional data protects both sides. See the Acas facts above and confirm the rules that apply to you.
How do you score an AI operator candidate?
Score each candidate against criteria you set before you see any work, using a range wide enough to separate them. Record the evidence for each score, not a general impression, and have a second reviewer score the same sample independently.
Workflow documentation
- Evidence: The note lists inputs, steps, tools and the expected output, so a new colleague could run it on day three.
Verification
- Evidence: They checked figures, names and claims against the source and say what they could not verify. A first draft accepted without checks is a weak result.
Human checkpoint
- Evidence: They name the point where a person reviews the work before anything goes to a client or customer.
Judgement on gaps
- Evidence: They flag missing or conflicting information and say what they would ask, rather than guessing.
Handoff
- Evidence: The status, open questions and next action are clear enough for the next person to continue.
Agree what each score means before you start. The example is a starting point, not a formula for every role.
What should you ask in the interview?
Ask the same core questions of each candidate, then follow up on their answers.
- "Walk me through a workflow you documented. What were the inputs, and where did a person check the result?"
- "Tell me about an AI output you rejected. What was wrong and how did you catch it?"
- "What would you ask me before starting this workflow?"
- "How would you show me, after a month, that the workflow is working?"
- "What should this role not be allowed to do without approval?"
Listen for specific steps and examples. Answers that name tools but skip the checks, or that treat every AI output as ready to send, are a warning sign.
What access should an AI operator have?
Give the person an individual login with only the permissions their tasks need, and agree who approves anything public or sensitive. Platforms such as Shopify let you assign roles with specific permissions, so use them instead of sharing an owner login. Review a sample of the work weekly at first, then widen access as the results justify it.
For wider guidance on handoffs, reviews and permissions, see offshore team management.
How does Staff Stack test AI operators?
Staff Stack's StackCertified assessment is our own internal test, not a third-party credential. It includes a role-specific practical test, AI workflow proof from prompt to output to edits to final deliverable, a communication simulation and a scorecard you can see before you interview. Read more on how AI-trained staff are assessed or browse the available roles.
You can use the same approach for any candidate: ask for the work, the workflow note and the handoff, and score them the same way.
How do Direct Hire and Managed Teams compare?
Compare both options against the same role, schedule and support needs. With Direct Hire, discuss the placement and the responsibilities your agency will take on. With Managed Teams, discuss the ongoing support and how responsibilities are shared. Ask Staff Stack to confirm the terms, employment, payroll, management and review arrangements for your role before you choose, and compare them on the pricing page.
Frequently asked questions
What does an AI operator do?
In this guide, an AI operator does recurring agency or ecommerce work with AI tools and is responsible for checking the result. Define the workflow, the inputs and the reviewer for your own role rather than relying on the job title.
How long should an AI operator work sample be?
Keep it short enough to see how the person works without becoming free labour, such as 30 minutes. Use fictional or redacted data, give every candidate the same brief and score against criteria set in advance.
Should candidates be paid for a work sample?
Acas, the UK workplace advisory body, says an unpaid work trial should be for a reasonable time, for example 2 hours or a shift, and should not usually last longer than one day. It also says the minimum wage must be paid if the trial is not genuinely for recruitment. Rules differ by country, so check those that apply to you.
What access should a new AI operator get?
Give an individual login with the permissions the tasks need and no more. Keep refunds, budget changes, publishing and customer messages with a named person until you have reviewed a sample of the work.
Should I choose Direct Hire or Managed Teams?
Compare the responsibilities and support your agency wants, then ask Staff Stack to confirm the terms of each arrangement for your role. Review pricing before you decide.
Ready to describe the role? Start with Direct Hire or explore Managed Teams.