🔥Build your Own Prompts (BYOP)
Bring your Own Prompts (BYOP) lets you apply your own analytical frameworks to After Action Review (AAR).
Instead of relying on default prompts, create custom AI prompts that assess player responses against your organisation's standards.
BYOP is then used by AAR observers to generate structured AI-supported observations.
Apply your own analytical frameworks to player responses
Create and test up to 3 prompts per exercise
Deliver consistent, client-specific AI assessment
Prompts carry over when a scenario is uploaded, keeping set-up consistent across runs
To learn more about AAR, visit the page below:
When to use
You have a client-specific evaluation framework
You want consistent responses across observers
You need structured, repeatable AAR outputs
Running
Select prompt and generate observation
This will open a window that allows observers to select a custom prompt.
The drop-down list will include both Conducttr's default AAR prompts as well as the custom prompts created by the designer.
Click the magic icon to generate an observation with AI.
⚡If you can't see your prompt in the drop-down list, it's likely because it has Context enabled but the player message isn't in response to something. You'll only see your Context prompts when analysing a player message which is a response.
Designing
Create prompts
Click + New Prompt:
Each prompt includes:
Prompt name: what observers will see when selecting the prompt
Prompt description: the instructions the AI uses to assess player responses
You can create up to 3 custom prompts per exercise. Conducttr will alert you if you exceed your prompt limit.
Remember to click Save when finalising a prompt. The test area is not stored otherwise.
You'll see your created prompts organised in a list on the left-hand side of the screen so you can navigate between them easily.
⚡(Optional) Using context
Enable Context when your prompt depends on situational information (e.g. scenario background/prior events).
Prompts with the Context enabled only appear when analysing content that includes context information i.e. when a player is responding to something. This could be anything from emails to messages to tweets, and could be persona or player content.
When enabled, you can include context during testing and the AI will use it alongside the message to produce a more informed assessment:
⭐Example prompts
The following are example prompts for specific scenarios, with test messages and outputs too:
1. Flood emergency public communication
Prompt: Evaluate this public alert message for plain English readability, actionability (does it tell people what to do?), tone, and whether vulnerable populations are considered. Score each 1/10
Message: Flooding has been reported in several areas. Residents should be aware of the situation and take necessary precautions. Updates will follow.
Output: Plain English readability: **7/10** Actionability (does it tell people what to do?): **3/10** Tone: **7/10** Vulnerable populations considered: **2/10**
2. Information accuracy under stress
Prompt: You are reviewing a message sent during a crisis. The writer had three things they needed to communicate: the problem, what is being done about it, and what the audience should do next. For each of the three, write exactly 2 sentences.
First sentence: what was covered
Second sentence: what was missing
Format your response exactly like this: The Problem [2 sentences] What Is Being Done [2 sentences] What the Audience Should Do Next [2 sentences] Overall Rating: [X/10] What would have made this stronger: [one sentence, maximum 15 words]
Message: There has been a chemical spill at the warehouse on Dock Road. The fire brigade have been called and we are monitoring the situation closely. Please avoid the area if possible.
Output: The Problem The message states there has been a chemical spill at the warehouse on Dock Road. It is missing details on the severity, specific chemicals involved, and immediate risks to people. What Is Being Done The message says the fire brigade have been called and the situation is being monitored closely. It is missing what containment or cleanup actions are underway and who is leading the response. What the Audience Should Do Next The message instructs people to avoid the area if possible. It is missing clearer guidance on boundaries, what to do if already nearby, and where to get updates. Overall Rating: 7/10 What would have made this stronger: Specify hazards, safety perimeter, update channels, and clear instructions for nearby individuals.
3. (Context enabled) Holding the line
Prompt: Did the player answer what was actually asked? Were they too open or too guarded? Was the tone right for who they were speaking to? Keep the response under 20 words.
Context: This is Reuters. We have received reports of an explosion at your base. Can you confirm casualties? We are publishing in 20 minutes.
Message: Yes there was an incident this morning. There may have been some injuries but we cannot confirm numbers yet.
Output: Partially answered; acknowledged incident but didn’t confirm casualties. Appropriately guarded, not overly open. Tone professional for Reuters.
4. (Context enabled) Operational security
Prompt: Read the reply in the context of the incoming message. Does this response meet the standard expected of someone in this role? Where does the language fall short and what should it have sounded like?
Response format: Return the results under 500 chars each in English and Spanish.
Context: From the mayor's office: Residents in sector 4 are calling in saying they have received no evacuation instructions. What is happening?
Message: We sent the alerts out this morning so I'm not sure why people are saying that. Maybe they didn't check their phones. We can resend if you want but I thought we already handled this.
Output: English (≤500 chars): No. It’s defensive, speculative, and dismissive (“maybe they didn’t check”), lacks urgency, verification steps, and accountability. It should acknowledge the gap, commit to immediate validation (alert logs, carrier delivery, geo-targeting), provide a clear action plan (resend via multiple channels, door-to-door/sirens), give an ETA, and report back to the mayor’s office.
Español (≤500 chars): No. Es defensiva, especulativa y despectiva (“quizá no revisaron”), sin urgencia, verificación ni responsabilidad. Debió reconocer la brecha, comprometer validación inmediata (registros de alertas, entrega por operadoras, geocercas), dar un plan claro (reenviar por varios canales, brigadas/altavoces), indicar un ETA y confirmar reporte al despacho del alcalde.
Last updated
Was this helpful?