The Percentage That Changes Its Mind
ZeroGPT's appeal is obvious: paste text, get a number, free. The problem is what the number actually is - a statistical guess about whether finished text resembles AI output. Users routinely report the same human-written passage scoring very differently across runs, after trivial edits, or between ZeroGPT and other detectors. That's not a bug unique to ZeroGPT; it's the nature of output-analysis detection. There is no ground truth in the text itself, so every detector is extrapolating from style.
For casual curiosity, that's fine. For anything with stakes - a grade, a byline, a client contract - a number that wobbles between runs is worse than no number, because it creates accusations it cannot adjudicate.
The self-check trap
Writers accused of AI use often run their own work through ZeroGPT hoping for a low score to show as "evidence." It doesn't work: a low score from one detector doesn't outweigh a high score from another, and the accuser can always re-run the roulette. You cannot win a probability argument with another probability.
Proof Doesn't Wobble
ValidDraft sidesteps the entire detection debate. You write inside a tracked editor that records keystroke dynamics, pauses, revisions, and cursor behavior - the irreproducible texture of real composition. The result is a certificate of human authorship with a public verification link and a permanent, tamper-evident record of the session.
Run a certificate through any number of detectors and nothing changes: the certificate doesn't claim your text "looks human." It documents that a human was there, writing it - a claim no after-the-fact detector can make or unmake.
Free vs free - the real difference
Both tools cost nothing to start. ZeroGPT's free output is a guess that expires the moment someone re-runs it. ValidDraft's free output is a permanent certificate with a verification page anyone can check, forever.