Designing Role Plays: Best Practices

This guide covers recommended limits and best practices for building effective role play scenarios in Brevity, including character descriptions, conversation length, and session consistency.

Character & Scenario Description Limits

Brevity enforces the following field-level limits. Designing within these keeps your content reliable across all AI providers.

Persona Attributes: Understanding the Limits

Each individual attribute can be up to 1,000 characters, but the total storage for all attributes combined is 5,000 characters (including JSON formatting overhead). There is no fixed limit on the number of attributes — what matters is the combined size. In practice, this means:

• 4–5 detailed attributes (800–1,000 characters each), or
• 8–10 shorter attributes (~400–500 characters each), or
• Any combination that keeps the total under 5,000 characters

If the total exceeds the limit, you'll see a validation error: "Field 'detail' is too large. Validation failed." To resolve this, shorten existing attributes or consolidate related details into fewer entries.

Recommendations

• Persona descriptions work best when focused on professional context, communication style, and behavioral traits — not exhaustive backstory. Aim for 800–1,200 characters for a well-rounded persona.
• Persona attributes are most effective when each one covers a distinct dimension (e.g., business motivation, pain points, communication style, decision-making criteria). Keep individual attributes focused rather than packing everything into one.
• Custom instructions are the most impactful field for controlling AI behavior. This is where conversation flow, tone, objection patterns, and scenario-specific rules should live. Even though 10,000 characters are available, concise and well-structured prompts (3,000–6,000 characters) tend to produce more consistent results than longer ones.
• Avoid redundancy between persona description and custom instructions. The persona description defines who the AI is; custom instructions define how the conversation should unfold.

Conversation Length

Brevity does not enforce a hard limit on conversation turns. The practical ceiling is determined by the context window of the underlying language model (up to 128K tokens depending on configuration).

Because the full conversation history is sent to the model on each turn, longer conversations consume more of the available context window, leaving less capacity for the model to reason and stay consistent.

Session Length Guidance

The platform does not impose a hard time limit on conversations. Sessions of 30 minutes or longer are technically supported without meaningful quality loss. However, session length is one of the most important design decisions an admin can make — not because of technical constraints, but because of how people actually learn.

The most effective practice sessions are intentionally scoped. Think of it like sports training: you don't learn to play basketball by only playing full games. You drill free throws, run specific plays, and scrimmage — each format building a different kind of skill. The same applies here.

Session Types: Drills, Situations, and Scrimmages

Drills are best for building habits. If a rep struggles with a specific transition or freezes on a particular objection, a 5-minute drill that isolates that moment and lets them run it 3–4 times in a row will build confidence faster than any longer format.

Situations are the workhorse format for most training programs. They give enough room for the conversation to develop naturally while keeping the learner focused on a specific challenge. This is where the "drop into the hard part" technique works best (see below).

Scrimmages are valuable for experienced reps who need to practice reading a conversation and adapting in real time. They're less efficient for building any single skill, but they're the closest thing to the real thing.

Designing Situations: Drop Into the Hard Part

For Situation-length sessions, resist the urge to simulate an entire real-world call. Instead, identify the most challenging 10–15 minute segment and design the scenario to drop the learner directly into that moment.

For example, a discovery call scenario doesn't need to start from "Hi, thanks for taking my call." Instead, set the scene in the custom instructions so the conversation begins mid-call, after rapport has already been built:

"You're a VP of Operations at a mid-market logistics company. A sales rep reached out after you downloaded a whitepaper on warehouse automation. You agreed to a discovery call because the topic is relevant, but you're skeptical about switching vendors. You've already exchanged pleasantries and some small talk — the rep is now transitioning into discovery questions. You're open but guarded. You won't volunteer pain points unless the rep asks specific, thoughtful questions."

This gives the learner practice on the part that actually matters — navigating resistance, asking the right questions, uncovering needs — without spending time on the easy opening minutes they've already mastered.

The same pattern works for any long conversation: set the context for what's already happened, define where the conversation picks up, and let the learner practice the segment that needs the most work.

Suggested Lengths by Use Case

Maintaining Consistency in Longer Sessions

System Prompt Design

The custom instructions field (system prompt) is the single most important lever for consistency. It persists across the entire conversation and is not affected by history length.

• Put non-negotiable rules first. The model pays strongest attention to the beginning of the system prompt.
• Be explicit about persona boundaries. Instead of "you are a skeptical buyer," specify the behaviors: "You push back on pricing at least twice before considering a discount. You never agree to a deal in the first conversation."
• Define the conversation arc. Structure instructions as phases (opening → discovery → objection → resolution) so the AI has a roadmap to follow.

Persona Description Design

• Favor behavioral traits over biography. "Interrupts when bored, asks pointed follow-up questions, values data over anecdotes" holds up better over many turns than "Grew up in Chicago, has an MBA from Kellogg, worked at three startups."
• Include communication style cues. Sentence length, formality level, and emotional tone help the AI stay in character even as context fills up.

Scenario Structure

• Match session length to the learning goal. Drills (under 8 min) for muscle memory, Situations (8–15 min) for practicing specific moments, Scrimmages (15–30 min) for end-to-end flow.
• Drop learners into the hard part. Use custom instructions to set the scene mid-conversation so practice time is spent on the skills that matter.
• Use progress milestones. Defining clear milestones gives the AI structural anchors throughout the conversation.
• Design for a clear outcome. Conversations with a defined goal (close the deal, uncover the objection, schedule the follow-up) stay on track better than open-ended ones.

Summary of Recommendations