GPT-written affirmations, minus the "grateful / blessed / manifest" script
Right Day pipes GPT-4o-mini through a tuned system prompt that refuses wellness cliché. Every morning, a new sentence — written for the woman you are today.
Type "give me an affirmation" into ChatGPT and you'll get: "I am worthy. I am enough. I am blessed." Every time. Because that's what the training set contains. Right Day patches the register at the prompt layer — the AI can't produce those lines even if you ask it to.
The prompt we built
Our system prompt (see enSystem in functions/index.js) explicitly forbids: "blessed", "grateful", "universe", "divine", "sacred", "surrender", "manifest", "abundance", "goddess", "trust the process". Instead it seeds the model with anchor lines like:
- "My time is expensive. My silence is a boundary."
- "I am the only person qualified to define my week."
- "My no does not require a reason."
Why prompting matters more than the model
GPT-4o-mini + a mediocre prompt = generic wellness slop. GPT-4o-mini + a 10-month-tuned sovereign prompt = something that reads like it was written by a woman who signs her own contracts. Model choice ≠ product quality. Prompt engineering is the product.
Native Arabic, Russian, English
Each language has its OWN system prompt, written by a native speaker. Arabic morning lines are not English lines with different letters. Russian sovereign register is not the Russian version of "you've got this." Three registers, three voices.