How it works
What happens to a sentence between you writing it and Tracely flagging it.
The check, step by step
Step 1
You write
Every few seconds while you write, Tracely reads the draft.
Step 2
Claims are picked out
Each sentence that states a fact is checked on its own.
Step 3
It's judged
A model decides whether it's wrong, thin, or missing a citation.
Step 4
It's looked up
Before flagging a claim for evidence, Tracely searches OpenAlex and the web.
Step 5
You decide
Nothing in your doc changes until you accept a fix.
Example check
- Draft readWhy school should start later
- Claims picked out5 sentences state a fact
- Judged“Most teenagers only need six hours…”
- Looked up“Moving the first bell to 8:30…”
- Nothing read backs itShown as an unverified claim
- Underlines drawnRed, blue, amber and grey
What's kept
- Text you check
- Processed and not stored
- Desktop app results
- Kept 1 to 7 days, then deleted
- Usage counts
- Kept 13 months
- Training
- Never used to train models
OpenAI judges the text under its API terms: it never trains on it, but may keep it up to 30 days to check for abuse.