Umpteenth

Find out what you keep re‑explaining to AI.

Umpteenth reads your ChatGPT and Claude history in your browser, counts the corrections you make again and again, and writes the instructions that stop them.

A sample Signature: about 300 thin vertical lines across a year, one per conversation, with coloured marks where the user corrected the AI.
A sample Signature. Each line is one conversation; colour marks a correction. Yours is drawn from your own export.
  • Voice
  • Length
  • Handoff
  • AI tells
  • Format

I ran it on my own history first.

These are the five patterns it found in my ChatGPT and Claude conversations. Each one now has a rule I paste into my settings instead of typing it again.

  1. 195
    times I re-explained my voice
  2. 101
    times I asked for something shorter
  3. 95
    times I rebuilt project context in a new chat
  4. 62
    times I removed phrasing that sounded like AI
  5. 36
    times I fixed the format

Three steps, about ten minutes once your export arrives.

ChatGPT and Claude both email you an export of your history. Umpteenth reads that file. It never asks for your account.

  1. Request your export

    In ChatGPT or Claude settings, ask for a data export. It arrives by email, and Umpteenth shows you the sample report while you wait.

  2. Drop it in

    Umpteenth reads the file in your browser, draws your Signature and counts ten kinds of repeat correction, each with examples in your own words.

  3. Paste the fixes

    Copy custom instructions, CLAUDE.md rules, memory lines and skills. Each one says how many of your requests it's based on.

Fixes come with their evidence.

Every line points back to the conversations it came from, so you can see why it's there before you use it.

Custom instructionsChatGPT and Claude

Write the way I write: plain words, short paragraphs, the main point first. Don't add a summary at the end.

Based on 195 of your requests

An example fix. Its wording is illustrative; the count is from my own audit.

Your history stays on your computer.

Your chat history

  • Read and analysed in your browser. There's no server to upload it to.
  • No account, and nothing to sign in to.
  • Optional AI rewording runs on a model in your browser, or with your own key. You see exactly what would be sent first.
  • Share cards show counts and your Signature, never message text.
  • Open source under the MIT licence, so anyone can check.

Your email address

  • Used for the launch email and the updates you tick. Nothing else.
  • Stored by the site's form provider, Netlify, which also records the IP address and browser of each sign-up. The app never sees any of it, and it's never linked to your history.
  • Never sold or shared.
  • Every email has a one-click unsubscribe. To have your address deleted, reply to any email.

Questions

Which AI tools does it work with?

ChatGPT and Claude exports at launch. Perplexity, Gemini, and Claude Code or Codex session logs are planned once their export formats are confirmed.

Does it cost anything?

No. The web app is free and open source. If you add your own API key for rewording, your provider bills you for that directly.

Will it change my ChatGPT or Claude settings?

No. Umpteenth writes the fixes and you choose which ones to paste in. It can't reach your accounts.

What's a Signature?

A picture of your year with AI: one thin line per conversation, with colour where you had to correct it. You can share it as an image. It contains counts only, never what you wrote.

See your own Signature in December.

Join the waitlist and you'll get one email when it's ready, plus the updates you picked.

Join the waitlist