The RACE Method
When you're writing custom prompts, lean on standard AI prompting best practices — and remember RACE: Role, Actions, Context, and Execution.
ROLE
Giving the AI a role can seriously improve how it approaches a task. The AI will step into the role you give it and ask follow-up questions from that perspective.
A few examples:
"You’re a market research professional seeking to uncover viewers’ interpretation of branded content…”
"You're a video editor looking for feedback on your latest cut of a branded advertisement..."
"You're a writer for a production company..."
ACTIONS
Once you've assigned a role, tell the AI what you want it to do. AI can't read your mind (yet), so if you want it to do something — or avoid doing something — you need to spell it out.
Example:
“When a viewer leaves a comment with a positive sentiment OR submits a slider rating higher than 7, ask a series of follow-up questions via a guided conversation to dive deeper into what they are enjoying about the content”.
Now the AI has clear directions for exactly when to act and what to do.
CONTEXT
Lumiere automatically gives the AI a few pieces of context:
The question or persistent reaction type
The comment or answer submitted by the viewer
The conversation goal
Adding your own context on top of that helps the model understand the conversation better. A quick mention of the video topic or background info can go a long way.
"Viewers are watching a video featuring the brand [ ]..."
Heads up: if you mention a brand name, the AI might bring it up in conversations. If you'd rather it stay behind the scenes, just say so:
"Viewers are watching a video featuring the brand [ ]. Without mentioning the brand name to viewers…"
EXECUTION
The last step is execution — telling the AI how to format its output. Want it to ask questions a certain way, or hit specific topics? Go for it. The only real limit here is your imagination.
Additional Best Practices
Beyond the RACE method, a few other things worth keeping in mind:
Be clear and specific. Clearly state your goal or question. Vague prompts lead to vague responses.
Use complete sentences. Full sentences help the model understand the structure of your prompt and respond more coherently.
Iterate and experiment. Don't be afraid to test different prompts. Use Lumiere's Preview mode to refine before you send anything to your audience.
Try prompt examples. Sometimes the best way to show the AI what you want is to give it a sample conversation to model.
Keep prompts neutral. Avoid prompts that could lead to biased or inappropriate responses. Biased or offensive cues can produce results you don't want.