AI that coaches, not corrects
Hinge’s first AI-powered coaching feature, and one of the first in the dating industry to be built around responsible AI principles from day one. As the sole content designer, I established the model, the language framework, and the guardrails ensuring the feature amplified every dater’s voice rather than homogenizing it.
My role
I owned the content design end to end—from the framework and principles down to the strings a user would read—collaborating directly with behavioral scientists, engineers, and product.
- Feedback language: wrote all three tiers of guidance copy, calibrated to feel encouraging and specific without being prescriptive or robotic
- Responsible AI principles: co-developed the transparency, authenticity, and equity framework that governed every content and UX decision
- Nomenclature and framing: named the feature, defined the terminology, and established how AI capability would be introduced to users
- UX copy: wrote all onboarding, empty states, error states, and user-facing explanations of how the feature behaves
- 0-to-1 thinking: defined the content model from scratch, including the principle that the AI should never provide replacement language
Three principles shaped every content decision
Transparency
The AI only reacts to what the user has already written; it never invents or extrapolates. Feedback is framed as a friendly nudge, not a verdict, so the user always opts whether to act on it.
Authenticity
The feature is designed to amplify the user's true voice, not standardize it. People write differently: brief, dry, chatty, precise. All of it is valid. Light, encouraging feedback, and no suggested replacement language.
Equity
Backgrounds and identities shape how people communicate: humor, cultural idioms, niche references. The model can't penalize difference. Guidance was tested against a wide range of expression styles, tones, and cultural contexts.
Three levels of guidance. Zero ghostwriting.
Prompt Feedback evaluates each answer and responds with one of three tiers—nudging daters toward authenticity without putting words in their mouth.
Great Answer
Your response is strong—keep it as-is. Positive reinforcement for an answer that shows personality and sparks conversation. In the calendar-invite screen, the AI recognizes the humor and keeps it.
Try a Small Change
A light nudge (e.g., a specific artist name, a concrete detail) makes the answer feel more vivid. Gentle, encouraging, and it preserves the user's voice.
Go a Little Deeper
Too brief or generic. "Asking me out" gets a prompt for specificity, but the feature never rewrites it. Thumbs up and down on every response closes the responsible-AI loop.