Millions are turning to AI in their worst moments.
It's failing them. Help fix it.

Help build the open dataset for emotionally intelligent AI.

How it works

1. 1.

   Pick a scenario.

   A specific hard message someone needs to write — a voicemail, script, or letter.

2. 2.

   Write your version.

   Short is fine. Your voice, not a model's.

3. 3.

   Get scored.

   Claude applies the scenario's own rubric and returns a 0–100 score plus a per-criterion breakdown.

4. 4.

   See where you stand.

   Compared to four frontier LLMs and every human contributor so far.

5. 5.

   Optionally contribute.

   Your response joins the public corpus, anonymously.

Why it matters

Most instruction-tuning data teaches AI to be helpful, cheerful, and polite. The hardest messages people ever write — eulogies, apologies, last words — are none of those things. They're restrained, specific, and often painful.

A corpus of graded human responses to those exact moments is what's missing. This is that corpus.