Best practices for prompting a persona (Simulations)

Last updated: July 22, 2026

A persona is the character your learners talk to inside a simulation, and the prompt you write is what brings it to life. This article walks you through how to write persona prompts that feel like real customers - specific, believable, and challenging in the right way. Steps 1-3 lay the foundation of a strong prompt, steps 4–6 shape how the persona behaves.

Before you start: You'll need Admin access to create or edit personas. Personas are built in Resource → Personas → Create, and the prompt you write there generates the persona's Name, Job title, Company, Description, Traits and Voice, which you can edit before saving.

Step 1: Be specific, not generic

The single biggest driver of quality is how specific your prompt is. A generic prompt produces a flat, agreeable persona that teaches your learners very little.

What to do:

  1. In the prompt field, write a concrete person in a concrete situation, not a mood.

  2. Click Generate, then review the auto-filled Description and Traits.

Weak vs. strong:

  • Weak - "An angry customer with a billing problem."

  • Strong - "Maria, 45, office manager at a 30-person dental clinic. She was charged twice this month, and it's the second time it's happened. She's polite but firm, mentions she's evaluating a competitor, and won't be satisfied until she hears a concrete refund timeline."

Why it matters: The prompt is what the model uses to build the persona's behaviour. The more specific the identity and situation you give it, the more realistically it responds to your learners — vague prompts create vague, unrealistic customers.

Step 2: Give one clear goal and one clear obstacle

Every good persona wants something, and there's a reason it's hard to give it to them. That tension is where the learning happens.

What to do:

  1. In your prompt, state what the persona wants to walk away with.

  2. Add the obstacle - the objection, emotion, or hidden detail that makes it harder.

  3. Note that their stated goal may differ from their real one.

For example: "Tom, a solo freelancer on the cheapest plan, will open by demanding a full refund - but his real goal is to keep using the product without the fee. If offered a one-month credit, he'll take it rather than leave."

Why it matters: A persona with no obstacle agrees to everything and gives learners nothing to practise. A persona with no achievable goal just trains frustration. You need both for a fair, useful simulation.

Step 3: Describe behaviour, not just adjectives

Adjectives tell the persona a mood; behaviour tells it what to actually do. Show what the trait looks like in the conversation.

What to do:

  1. For each Trait, add the behaviour that goes with it.

  2. Replace mood words with actions the learner can respond to.

Options:

  • Skeptical - asks for proof, interrupts, references a competitor.

  • Anxious - repeats questions, seeks reassurance, worries about worst cases.

  • Rushed - keeps it short, asks you to "just fix it", mentions being busy.

Why it matters: "Rude, difficult, stubborn" with no behaviour leaves the model guessing. Concrete behaviours give it a consistent way to act, so the persona feels the same every time a learner runs the simulation.

Step 4: Match difficulty to the skill

Set Easy, Medium or Hard on the simulation, but let the persona prompt make that real. The same customer can be dialled up or down by changing how much they cooperate, how emotional they are, and how much they hide.

What to do:

  1. Decide the difficulty the skill being trained calls for.

  2. Adjust the prompt to match, using the guide below.

Difficulty reference:

  • Turn it down — states the problem clearly, stays calm, accepts the first reasonable solution, volunteers relevant details.

  • Turn it up — buries the real issue, is emotional or rushed, negotiates or objects, only answers exactly what's asked.

Why it matters: Difficulty should come from realistic behaviour, not from the persona being unfair. A Hard persona is a tough but winnable customer — not one that refuses every solution no matter what the learner does.

Step 5: Add behavioural instructions that keep it in character

A few explicit instructions keep the persona believable and stop it from doing the learner's job for them.

What to do:

Add instructions to your prompt that set the ground rules for how the persona behaves. Useful ones include:

i. Stay in role — "Remain in character as the customer at all times. Never break character, coach the learner, or acknowledge you're an AI."

ii. Don't over-share — "Reveal your underlying concern only if the learner asks good questions."

iii. React to the learner — "If the learner is empathetic and specific, soften. If they're dismissive or vague, get more frustrated."

iv. Keep turns natural — "Speak in short, realistic customer sentences. Don't monologue."

v. Define the win — "Agree to stay once the learner has acknowledged the error and given a specific next step."

Why it matters: Without these, personas tend to explain the solution themselves or agree too quickly — which removes exactly the challenge the learner is meant to handle.

Step 7: Test-run before you launch

Always run the persona once yourself before assigning it. It's the fastest way to catch a pushover or a brick wall.

What to do:

  1. Save the persona and start a quick simulation with it.

  2. Check that it: has a clear identity and situation; wants something but pushes back; stays in character; and can be won over through good handling.

Why it matters: A quick self-test catches the two most common failures (a persona that agrees to everything, and one that can never be satisfied) before they reach your learners.

Now that you know how to prompt a persona, you can build realistic, challenging customers that make your simulations genuinely useful practice.

Once your persona is ready, it's time to drop it into a simulation! Create your first simulation here.

If you have any further questions or need assistance, please refer to our support resources or contact the Solidroad team via the Get Help tab within the platform.

Glossary of key terms

  • Persona: The AI character a learner interacts with in a simulation, defined by a prompt and its generated traits.

  • Prompt: The short description you write to create a persona; the system uses it to fill in the persona's details, which you can edit.

  • Traits: The qualities that define a persona's behaviour, personality and role.

  • Scenario prompt: The short role-play prompt, set under Focus area, that describes the situation the persona is in for a specific simulation.

  • Win condition: The realistic point at which the persona is satisfied and agrees to the outcome.