A power user applies AI skillfully to their own work. An AI champion helps colleagues develop useful practice too. One person can do both, but personal proficiency alone does not tell you whether they want to coach, can explain their judgment, or have time to help.
If you are choosing champions for a team, look for competent use, patient coaching and sound judgment, then agree what the person is willing and able to take on. Here, champion means a colleague helping peers with everyday work. Some organizations also use the title for senior leaders responsible for an entire adoption program.
Separate personal skill from the peer role
OpenAI Academy's guide to the AI Champion role makes this distinction between personal proficiency and helping teams adopt useful practices. Its program includes both organization-level leaders and people activating practice within teams. The title is broad enough that you need to define the job locally.
Consider the difference through the work you expect someone to do:
| Question | Evidence of a power user | Evidence of a peer champion |
|---|---|---|
Can they do useful work with AI? | Shows a task, its output and the checks they made. | Has enough practical competence to guide the tasks they support. |
Can someone else learn from them? | Explains their own method when asked. | Adapts the explanation and lets a colleague try it. |
What happens when the output is weak? | Recognizes problems and revises or stops their own attempt. | Helps the colleague identify the problem and decide what to do next. |
What happens beyond their expertise? | Finds help for their own task. | Recognizes the boundary and connects the colleague with the right owner. |
What commitment is involved? | Uses AI as part of their own responsibilities. | Agrees time and scope for helping others alongside their ordinary work. |
These are overlapping capabilities. The table helps you examine the additional responsibility, rather than assign permanent types to employees.
In a public Reddit account from February 2026, one Reddit user said their engineering manager had asked them to champion AI use for company efficiency without giving concrete direction. Their role and employer were not stated. The account captures a problem a role agreement should prevent: a person can receive the title without knowing what they are expected to change.
Find candidates beyond the visible enthusiasts
Start with people already doing useful work and people colleagues approach for help. Ask managers and team members for examples of someone who explained a difficult task, made a reusable example or followed through on a question. Invite those people to discuss the role before assuming they want it.
Usage information can help you find experienced users, where your organization legitimately collects it. It cannot show whether they listen well or whether the work they produce is sound. Microsoft's Power Platform champion guidance considers practical expertise alongside leadership, collaboration and willingness to help. That guidance comes from a broader platform-adoption context, rather than a study validating how to select generative-AI coaches.
Mendo co-founder Quentin Amaudry describes a related distinction in a LinkedIn account of deployment reviews. He reports seeing champion lists conflate people with deep workflow knowledge and visible advocates, while quieter users were not asked how they worked. This is a vendor's practitioner account, with no independently verified sample. It is a useful reason to ask for concrete examples before relying on visibility.
Someone with a valuable method may prefer to contribute an example for another colleague to teach. Preserve that option. Making peer coaching the price of being recognized for good work can leave you with a reluctant coach and less access to their expertise.
Observe a small coaching session
A conversation about enthusiasm will only take you so far. Try a short, voluntary practice session with a willing colleague and a task the candidate understands. Explain what you want to observe and invite the candidate to reflect afterward. Use it to inform a role conversation, with room for development; one session is not a validated predictor of future performance.
- Choose a familiar, low-risk task. Agree an approved tool and shareable material. Describe what a satisfactory result requires before opening the tool.
- Ask the colleague to explain the work. Let the candidate establish the colleague's goal, current approach and uncertainty. Watch whether the explanation changes in response.
- Let the colleague make an attempt. The candidate can demonstrate part of the process, then hand control back. A polished demonstration alone does not show whether someone else can proceed.
- Inspect the result together. Ask how they would check it, what would make it unacceptable and when they would stop or seek help. Use an imperfect output if one appears naturally.
- Debrief separately and together. Ask the learner what they could now repeat and what remains unclear. Ask the candidate what they noticed, what they would change and whether they want to do more of this work.
Record evidence and choose the next step
Use the following prompts to discuss what you observed. Keep notes specific to behavior and the task. Do not total them into a score that implies more precision than the exercise provides.
- Practical competence. Could they explain the task, demonstrate an appropriate use and check the output? If their own checks are unreliable, arrange learning and another supported attempt before asking them to teach that task.
- Listening and explanation. Did they ask what the learner needed, adapt their language and leave room for questions? If they repeatedly took over, practise handing the task back before offering a wider coaching role.
- Judgment and boundaries. Did they recognize uncertainty and know whom to ask? An unresolved access or policy question should reach its owner, rather than receive an improvised answer.
- Learner independence. Could the colleague explain or repeat the next step? If not, identify what support was missing. One learner's difficulty is a prompt to investigate, not a verdict on the candidate.
- Willingness and capacity. Does the candidate want the role, and can their manager make space for it? If either is absent, agree a smaller contribution, revisit later or accept a decline.
Avoid making agreement with every AI proposal a selection criterion. The Cabinet Office's People Factor guidance describes learning from interviews that some non-adopters did not see relevance to their roles. Its advice draws on the Government Communications Assist rollout and recommends hearing from users and non-users. It does not establish which champions are most effective, but it supports looking beyond enthusiastic feedback.
A candidate who asks whether a task is worth changing may help colleagues make a better decision. Look at how they investigate and respond to evidence, including when the evidence supports stopping.
Agree the role before announcing it
Leave the selection conversation with a short agreement the candidate and manager both understand:
- People and tasks supported. Name the team, approved tools and kinds of work within scope.
- Help offered. Choose concrete activities, such as a practice session, reviewing an example or collecting recurring questions.
- Time and priority. Agree when this work happens and what ordinary work will move to make room. Use a workload diary and manager agreement to estimate the commitment.
- Referral routes. Name the owners for access, data-use rules, specialist quality checks and workload decisions.
- A review point. Decide when to discuss whether the role remains useful and manageable, including how to change or leave it.
Microsoft's Copilot feedback guide for champions emphasizes hearing from power users and skeptics, discussing specific tasks and escalating recurring problems. A champion needs a route for those findings to reach someone who can act. Collecting concerns without a response path adds work without resolving the underlying issue.
Connect the role to the team's regular review of AI-assisted work. The manager can address priorities and unresolved decisions; the champion can bring examples and help colleagues practise. For the wider support decisions around skills, working conditions and concerns, use the people and change guide.
Your next step is to invite one potential candidate to a role conversation and a small coaching exercise. Bring an actual task and a clear account of the support you can offer them.
Once individual commitments are agreed, use the cross-department launch guide to map coverage and rehearse how colleagues will find help.
Questions about power users and AI champions
Does an AI champion need to be a technical expert?
An AI champion needs enough practical competence to guide the tasks they support and recognize when to ask for help. A marketing colleague coaching someone to turn launch-meeting notes into an action list does not need to build AI systems. They do need to show how to check the assigned people, deadlines and dependencies against the notes.
Technical or high-stakes work needs the relevant expertise and a clear route to specialist review. Define that boundary in the champion's role agreement, including who handles access, data-use and quality questions. Enthusiasm is not a substitute for reliable checking or permission to advise beyond that boundary.
Can one person be both a power user and a champion?
Yes. A skilled AI user may also be an effective peer coach, and their own work can supply useful teaching examples. The extra responsibility is helping someone else understand and repeat the method, rather than completing the task for them. Personal usage alone does not establish that ability.
Use a small, voluntary coaching session to discuss how the person listens, explains and hands control back to the learner. Then confirm that they want the role and have agreed time for it. One session can inform that conversation; it is not a validated test of future champion performance.
Can someone skeptical about AI be a good champion?
They can, if they are willing to investigate suitable uses and help colleagues evaluate results fairly. For example, a marketing coach who questions an AI-generated launch action list can help a colleague compare it with the meeting notes and spot an incorrectly assigned product-claim check. The useful behavior is investigating the output, not agreeing with every proposal to use AI.
Look for openness to evidence, respectful explanations and willingness to acknowledge when an approach works or should stop. Use the behavioral evidence prompts to discuss those observations. Skepticism alone establishes neither competence nor inability to support colleagues.
What if a strong user does not want to become a champion?
Accept the decision. Skill with AI does not create an obligation to run peer coaching, and someone may prefer to concentrate on their existing responsibilities. Ask whether they would welcome a smaller contribution, without making recognition for their work conditional on accepting it.
A marketing coordinator could, for example, share an approved, checked launch action list with an interested coach and explain how they verified its owners against the meeting notes. Agree the time and scope even for that smaller contribution, and accept a decline. The candidate-finding guidance explains why useful methods and willingness to coach should be considered separately.
