Career Navigation

AI Roleplay for Skill Practice

7 min read

AI Roleplay for Skill Practice

Dimension: Pursuit · Type: Stage

A practice routine that uses AI personas to rehearse high-stakes conversations (interviews, stakeholder pushbacks, performance discussions) before they happen for real. Builds confidence through repetition in private, with immediate structured feedback.

Introduced by Tom Frohner (LinkedIn) at the AI for Your Career, Practical Tools and Prompts session of the UN Inter-Agency Career Week 2026, on 6 May 2026. Extended in the Mastering Job Interviews session on 8 May 2026 by Tamara Roura (OCHA), who added a five-prompt interview-prep toolkit suitable for any general-purpose LLM.

The framework

The routine closes the gap between knowing about a skill and being able to perform it. Mistakes happen in private, repetitions compound, and the feedback loop after each round produces the actual learning.

When to use it

  • Before a job interview, especially competency-based interviews where examples are scored.
  • Before a difficult stakeholder conversation: a partner who has missed deadlines, a manager who is not engaged with your career, a peer with whom there is conflict.
  • Before a performance discussion, your own or one you are giving.
  • As a regular practice before any conversation where the cost of fumbling matters and you would otherwise have no chance to rehearse.

What you need

A specific scenario you want to rehearse, defined enough that the AI can build a credible counterpart. Knowledge of the persona you will be talking with: their likely concerns, their style, the constraints they are operating under. 30 to 45 minutes per roleplay session. Access to LinkedIn Learning’s AI Roleplay (free for all LinkedIn users, available in English, German, and French at time of writing) or any general-purpose AI assistant configured as a roleplay partner.

Steps

  1. Define the scenario in one sentence. “Practise addressing a stakeholder who is repeatedly missing deadlines on a cross-functional project, while preserving the working relationship.” Specific enough that the AI can construct a coherent counterpart.
  2. Define the persona’s characteristics. Experienced or junior. Supportive or sceptical. Time-pressed or relaxed. Defensive about a specific issue or open to discussion. The more specific the persona, the more useful the rehearsal.
  3. Set a success rubric. What does a strong handling of this conversation look like? In LinkedIn AI Roleplay, the system suggests one (describe the impact, give specific behaviour-focused feedback, explore the root cause, co-create a recovery plan, reinforce the relationship). With other AI tools, draft the rubric yourself before you start.
  4. Run the conversation. Respond in your own words. No script, no multiple choice. The AI takes the first turn, you respond, it pushes back if appropriate, you adjust. The key is that mistakes happen in private.
  5. Get the feedback. At the end, ask the AI for a structured strength-and-improvement analysis against the rubric. Identify two or three concrete behaviours to adjust on the next round.
  6. Run again. The compounding value is in repetition. Twenty rounds of interview practice, or five rounds of a difficult stakeholder conversation, produce real confidence rather than theoretical preparation.

Worked example

A programme officer has a competency-based interview in two weeks for an internal P-3 role. She rehearses using AI roleplay.

  • Scenario. Competency-based interview for a P-3 programme management role. Five questions covering planning and organising, partnership management, results-based management, accountability, and adaptability.
  • Persona. A panel chair, experienced UN hiring manager, polite but probing. Pushes back when an answer lacks specificity. Asks a follow-up like “what would you have done differently?” after each main answer.
  • Rubric. STAR-structured answers (R-CAR for written; STAR is the spoken-equivalent: Situation, Task, Action, Result). Concrete examples drawn from her BASIC bank. No filler, no generic claims.
  • Round 1. She runs through five questions. The AI pushes back twice on weak specificity. The feedback identifies that her partnership-management answer used the word “stakeholders” five times without naming any specific person, agency, or working relationship.
  • Round 2. Same scenario, sharper answers. The AI flags that her adaptability example covered the situation but not the result.
  • Round 3 to 6. Iterating, with each round addressing the prior round’s feedback.
  • Round 10. Steady. The answers feel rehearsed without sounding scripted.

By the day of the actual interview, she has had ten rounds of practice. The mistakes that would have happened in front of the panel happened in private.

The five-prompt interview-prep toolkit

Tamara Roura’s prescription was to treat AI as a multi-purpose prep partner and to run each use as a separate, focused prompt rather than asking AI to “help me prepare”. The five uses:

  1. Interview simulator. Paste the JD and ask the model to run a roleplay with three to four competency questions drawn from the post. This is the live-rehearsal use covered in detail above.
  2. Organisation researcher. Ask the model to summarise the entity’s mandate, current strategic priorities, recent major decisions, and likely operational challenges, with citations or links you can verify. The output feeds the Why You test.
  3. Question generator. Ask the model to produce 8 to 10 likely interview questions from the JD, then practise answering each one out loud. This complements Duties-Driven Interview Prep, which generates questions from the duty bullets specifically.
  4. STAR+L story editor. Paste a draft answer (Situation, Task, Action, Result, Lesson) and ask the model: is the action section roughly 70%? Is the example calibrated to the seniority of the post? Are there gaps? Is the lesson generalisable? This is a content-quality check, not a writing assistant.
  5. Motivation pitch editor. Paste your 60 to 90 second pitch and ask: does it feel authentic, is it adapted to the entity, are the four beats present (who you are, how you got here, why this role, why this entity)?

A sixth use, sometimes treated as a variant of the STAR+L editor: ask the model to help you reframe a real weakness constructively, in a way that demonstrates self-awareness without damaging the candidacy.

The operational rules: run each prompt separately and rehearse the output out loud with a human (peer, mentor, coach) before the day. AI is a preparation partner, not a substitute for human practice.

Pitfalls

  • Defining the scenario too vaguely. “Practise an interview” produces generic rehearsal. A scenario that names the role, the agency, and the likely panel composition produces useful rehearsal.
  • Defining the persona as flat. A persona without specific traits is a flat rehearsal partner. A persona with two or three concrete characteristics (sceptical of AI use, time-pressed, defensive about a specific budget topic) produces realistic exchanges.
  • One round and done. The science of deliberate practice is in the repetitions. The first round is calibration; the value is in rounds three through ten.
  • Using AI roleplay as the only preparation. Roleplay rehearses the delivery; it does not replace the substantive work (knowing the role, the agency, the relevant policy context). Use it as the final layer, not the only layer.
  • Skipping the feedback loop. The rehearsal alone is half the value. The structured feedback after each round is the other half.
  • Treating AI feedback as the truth. AI feedback is a strong second opinion, not a recruiter’s view. Pair with one round of human practice (a peer, a mentor) before the real conversation.

When not to use it

When the conversation is so context-heavy that no AI can credibly simulate it (a confidential restructuring discussion, a sensitive whistleblowing report, a delicate cross-cultural negotiation with specific historical context). In those cases, find a human rehearsal partner.

When the persona involves real, identifiable individuals whose names or details you would have to share with the AI. Do not put confidential information about real colleagues into public AI tools.

A note on the source

Tom Frohner cited the science of deliberate practice: 70% of skills are learned through doing, not watching, and AI roleplay enables that at scale, in private, with immediate feedback. Tamara Roura was explicit that AI prep does not replace rehearsing out loud with a human before the day.

How I use it

Personal note pending. Davide to fill.

Notes compiled by Davide Piga. Last updated 2026-05-09.

Graph View

Backlinks