Scenarios

A scenario is a branching AI roleplay. The learner holds a real conversation with an AI character - a frustrated customer, a job candidate, a patient - and is scored on how well they handle the exchange, not on picking a right answer. This page walks a course creator through setting one up, and shows what the learner sees.

Accessible to: Trainers, Workspace Owners, Admins (authoring) · Learners (playing a scenario)

The idea in one minute

• You give the AI the facts and character it should play (its "grounding").
• You write a short brief: who the learner plays, who the AI plays, and what the learner is trying to achieve.
• You set the grading criteria the AI marks the conversation against.
• The learner plays it out in chat (or by voice), ends the scenario, and gets a debrief with their score, per-criterion feedback, and a transcript - with the option to restart and try again.

The Scenario Editor is organised as a short, ordered strip of steps across the top - Ground the AI, Write the learner brief, Set the objectives, Set scoring criteria - so you can work through it top to bottom. Start with grounding: everything else can be drafted from it.

Step 1 - Ground the AI

Open a scenario from a course's Content tab (add a Scenario block, or click an existing one) and go to the Grounding (AI only) tab.

Grounding is the material the AI roleplays from. Without it, the AI improvises and can invent details. With it, the AI stays in character and weaves your facts into the conversation.

1. Attach a source - an image, video, YouTube link, a web page URL, or a PDF - then click Draft from this source. Aimee reads the source and drafts the grounding for you.
2. Or click Course content to pull text straight from this course's own slides. Pick the modules and slides in the pop-up - slides with no text are shown but greyed out - and their text and voice narration are added to the grounding below, ready to review and edit.
3. Or click Knowledge base to pull from your workspace's own documents (see below).
4. Review and edit the drafted text. This is the AI's private script - it shapes how the character behaves.

Grounding from your knowledge base

If your workspace has the Knowledge Base turned on, a Knowledge base button appears alongside the other sources. Use it when the facts the roleplay needs already live in a document you've uploaded - a policy, a procedure, a product sheet.

You can pull from it two ways:

• Use a whole document - pick a document and its full text is added, in its original order. Best when the roleplay follows a procedure, where the order of the steps is most of the meaning. Very long documents are refused rather than cut short, with a message telling you to search instead - a half-added procedure is worse than none.
• Search for what this roleplay needs - type what the AI needs to know ("our refund window for annual plans") and the most relevant passages are added.

Either way the text is added to the grounding below with a note of which document and section each passage came from, so you can see what you're working with. Only documents that have finished processing are offered.

Note

What gets added is a copy, not a link. You can edit it freely, and it won't change underneath you if the original document is later updated or removed. If the source document does change and you want the scenario to match, pull it in again.

Note

Grounding is normally used only to brief the AI - learners don't see it. The one exception is a YouTube video (see Showing a video to the learner below).

Switching to a new topic

If you swap in a different source and click Draft from this source again, Aimee asks whether to also clear the learner brief, scoring criteria, title, and voice narration - since those were likely written for the old topic. Choose Replace + clear other fields when you're starting a genuinely new scenario, or Replace context only if you just want to refresh the grounding and update the rest yourself.

Step 2 - Write the learner brief

Go to the Learner brief tab. Here you set who plays whom:

Field	What it is
Learner role	Who the learner plays (e.g. "Customer Service Representative")
AI role	The character the AI plays (e.g. "Dissatisfied Customer")

If you have already added grounding, click Draft from this source and Aimee fills these fields in for you from it - you can then adjust them. Writing the brief by hand works too; the fields simply make sure the roleplay has a clear shape.

Role portraits

Each role needs a portrait - a photo of the person the learner is talking to, and one for the learner's own role. The portrait is what the learner sees during the roleplay, swapping to whoever is currently speaking. For each role you can:

• Upload an image of your own,
• Choose from your workspace avatar library (portraits are kept under an avatars folder so they're easy to reuse across scenarios), or
• Generate one with AI - Aimee creates a forward-facing photorealistic portrait that fits the role and scenario. Each generation produces a different person; press Regenerate if you want another.

The AI character's voice matches the chosen AI-role portrait (there's a female / male toggle if you want to override it).

Step 3 - Set the objectives

Go to the Objectives tab and list what the learner should try to achieve, one goal per line. Objectives are drafted for you along with the brief when you draft from a source, and you can edit them here.

Step 4 - Set scoring criteria

Go to the Scoring tab. Grading criteria belong to this scenario - each scenario carries its own set, so two scenarios with different goals are marked differently.

Every scenario needs at least one criterion. A scenario with no criteria cannot be published - the publish check flags it and links you straight to this tab. Without criteria there is nothing for the AI to score the conversation against.

For how criteria, weights, and guidance work in detail, see Scoring & Rubrics.

Checking your work - Preview

Click Preview in the scenario editor header at any point to see the block exactly as a learner meets it: the title, the cover image, the brief built from your roles and objectives, the "Play your part" instructions, and the grounding video if you've chosen to show it.

The preview picks up what's on screen right now, so you don't have to save first - type a role or an objective, click Preview, and the brief updates. Your grounding notes and your scoring criteria never appear; those are for you and the AI only.

The roleplay chat itself doesn't run in the preview. Click Edit to go back to authoring, or use Preview in the Course Editor to walk the whole course as a learner.

Showing a video to the learner

When you attach a YouTube video as grounding, a checkbox lets you decide whether the learner sees it:

• Shown (the default): the learner can watch the video before the roleplay - useful for "watch this, then handle the situation" designs.
• Hidden: the video is used only to brief the AI, and the learner never sees it.

The editor spells out the consequence next to the checkbox as you toggle it, so it's clear whether learners will see the video.

What the learner experiences

When a learner reaches a scenario in a course, the AI opens the conversation in character and the learner replies in role. The chat invites them to reply in character rather than ask questions. The speaking role's portrait is shown as the conversation goes back and forth.

Learners can type their replies or press Start Scenario to talk out loud by voice. Switching between text and voice restarts the scenario, so they pick one and stay with it.

When they've handled the situation, they end the scenario with End scenario & get your feedback. Aimee confirms, scores that conversation, and opens the scenario debrief.

The scenario debrief

Each scenario has its own debrief, shown as soon as the learner ends it - so in a course with several scenarios, each one gives its own feedback. It shows:

• The score for that conversation.
• Per-criterion feedback against the scenario's grading criteria.
• A transcript of that scenario, with the AI's lines labelled by the character it played (its AI role) and the learner's lines as You.

From the debrief the learner can Restart the scenario to try again (their latest attempt is the one that counts) or Continue to the rest of the course. A scenario has to be ended at least once before the learner can move past it - the Next control stays disabled until then, so scenarios can't be skipped.

Note

If a scenario has no grading criteria set, its session is shown as un-graded (a dash, not 0%) rather than penalising the learner for a missing setup step - another reason to add criteria before publishing.

Tips

Tip

Work the tabs in order - ground first, then let Aimee draft the brief, then set criteria. It's faster than writing everything by hand, and the drafted brief usually only needs light edits.

Tip

Read the grounding back as if you were the AI. If a fact the learner should encounter isn't in there, the AI won't reliably bring it up.

Related

• Courses - building a course and adding scenario blocks
• Scoring & Rubrics - grading criteria, weights, and score review