Project 04Product
Gen Z Tarot
A tarot app with a voice, built to find out how much of an app’s appeal is tone rather than function.
- Solodesign and build
- Toneas the product
- 3-cardspread
01
At a glance
The whole project in three steps.
- The problem
Every tarot app reads the same, because the model writes in whatever register you let it.
- What I did
Specified the voice first, then built a typographic three-card app around it.
- The result
Shipped. Same mechanics, a different voice, and the output stopped sounding like every other tarot app.
Skills it shows
- Tone of voice
- Prompt design
- Product design
- Interface copy
Tools
- OpenAI
- Gemini AI
02
The challenge
Why the project had to happen.
Tarot apps are a solved problem functionally (draw cards, return text) and almost all of them read identically, because the model writes in the register you let it. That made it a good place to test something I care about professionally: how much of a product’s appeal is the thing it does versus the way it talks.
Read the full context
So the app is deliberately not mystical. The readings are blunt, specific and occasionally unflattering, which is both the joke and the point.
03
How I did it
3 steps. Open any one for the detail.
Write the voice before the app
Specified the tone first and built everything else to serve it.
How
The tone was specified first and everything else built to serve it. A voice added at the end is a system prompt; a voice designed first is a product.
Give the question a shape
Asked what the question is about, so readings come out pointed.
How
An open text box produces generic readings. Asking what the question is ABOUT narrows the output enough to be pointed, which is what makes it land.
Keep the ritual, cut the mysticism
Kept the shuffle, the draw and the three-card spread; cut the illustration.
How
The shuffle, the draw and the three-card spread stay, because the ritual is the pleasure. The illustration and the incense do not: the interface is typographic.
04
See it
Gen Z Tarot
Demo — scripted, no live model
The shuffle, the draw and the spread are genuinely interactive — none of that needed a model. Only the reading text is fixed, keyed by card and by what you asked about.
Choose a subject, then shuffle and draw.
05
What it achieved
A finished, shipped app whose distinguishing feature is entirely editorial rather than technical.
Confirmed the premise: with the same mechanics and a specified voice, the output stops sounding like every other tarot app.
The demo on this page is fully interactive. The shuffle and the spread are real; only the reading text is fixed.
06
What I took from it
A voice added at the end is a system prompt. A voice designed first is a product.
An open text box produces generic readings. Asking what the question is about makes the output pointed.
With the same mechanics and a specified voice, the output stopped sounding like every other tarot app.