The Prompt Lab — Contrastive Goal Framing
Learn the contrastive goal framing prompting technique with concrete before/after examples.
Contrastive Goal Framing
The Technique
Contrastive Goal Framing means telling the model not just what you want, but explicitly what you want instead of — pairing your target outcome with the default or mediocre outcome it should actively avoid. This works because language models are trained on enormous amounts of generic output; naming the failure mode you’re trying to escape pulls the generation away from that attractor and toward something genuinely differentiated.
The Naive Prompt
Write a cold email to a VP of Engineering at a mid-size SaaS company
pitching our AI code review tool, Revue.Why It Falls Short
Without a contrast, the model defaults to the median cold email it has seen in training — which means opening with “I hope this finds you well,” burying the value proposition in paragraph two, and closing with a soft ask. The output is technically correct but indistinguishable from the hundred other cold emails that VP deletes before lunch.
The Improved Prompt
Write a cold email to a VP of Engineering at a mid-size SaaS company
pitching our AI code review tool, Revue.
My goal: an email that feels like it was written by someone who understands
engineering culture — skeptical, technical, pressed for time.
What I'm NOT trying to write: a marketing email. Avoid:
- "I hope this finds you well" or any warm-up filler
- Vague claims like "saves time" or "boosts productivity" without numbers
- A soft close like "would love to connect" — instead, give them a specific,
low-friction next step
The email should earn credibility in the first sentence, deliver one concrete
proof point (we cut PR turnaround by 40% at a 200-person fintech), and end
with a single yes/no question.Why It Works
By naming the exact failure modes — filler openers, vague ROI language, weak closes — you give the model an active constraint to push against rather than a vague target to aim at. The contrast between “feels like a marketer wrote it” and “feels like an engineer wrote it” also implicitly shifts tone, vocabulary, and structure without requiring you to specify every stylistic choice manually.
When to Use This
- Creative and persuasive writing where the model’s “average” output is trained toward safely bland — cold outreach, pitch decks, ad copy, executive summaries
- Any task with a well-known, hackneyed default format — press releases, job descriptions, LinkedIn posts — where naming what you don’t want breaks the gravitational pull of cliché
- High-stakes single-shot requests (especially with more expensive calls to Claude Fable 5.1 or GPT-6 Astra) where you can’t afford two or three iterations to coach the model away from a mediocre first draft