The Prompt Lab — Temperature Framing

Temperature Framing

The Technique

Temperature Framing means explicitly signaling the creative latitude you want — not by setting a model parameter, but by describing the expected output’s relationship to convention directly in the prompt. Models like GPT-5.5 and Claude Sonnet 5 are calibrated to infer register and risk-tolerance from context; telling them where on the spectrum from “safe and expected” to “surprising and unconventional” you want to land produces outputs that actually hit the target without multiple rounds of correction.

The Naive Prompt

Write a subject line for a re-engagement email campaign targeting 
users who haven't logged into our project management tool in 90 days.

Why It Falls Short

Without latitude guidance, the model defaults to the statistical center of “good email subject lines” — competent, inoffensive, and completely forgettable. You’ll get something like “We miss you — come back and see what’s new”, which is exactly what every other SaaS company sends and will get buried. The model has no basis for knowing whether you want safe or bold.

The Improved Prompt

Write a subject line for a re-engagement email campaign targeting 
users who haven't logged into our project management tool in 90 days.

Creative latitude: Push toward unexpected — avoid the clichéd 
"we miss you" framing entirely. The line should feel slightly 
surprising or even a little odd, but still clearly relevant 
to someone who's fallen behind on work. Imagine the range from 
"corporate safe" (1) to "a copywriter who just won a Clio Award" 
(10) — aim for a 7 or 8. Provide five options across that range 
so I can see the gradient.

Why It Works

Explicitly naming the creative spectrum — and anchoring it with a vivid reference point like “Clio Award-winning copywriter” — gives the model a calibrated target rather than a vague instruction to “be creative.” Requesting five options across a gradient forces the model to generate meaningful variation instead of five paraphrases of the same idea, so you see what’s actually possible at different risk levels.

When to Use This

• Marketing and creative copy tasks where “make it better” produces mediocre iteration but you don’t want to fully constrain the output with examples — Temperature Framing lets you steer without over-specifying.
• Brainstorming phases where you need to map the solution space quickly; asking for options across a named spectrum is faster than re-prompting with different constraints each time.
• Anytime a model is playing it too safe — particularly with GPT-5.5 Instant or Claude Haiku 4.5 in default API configurations, which tend toward conservative outputs; this technique explicitly licenses the model to range further without requiring a system prompt change.