PassLab

The submission-ready humanizer

If PassLab passes it,
submit it.

Other humanizers stop when the score drops. PassLab stops when the paper is ready.

Lower AI risk without trading away meaning, evidence, citations, or readable English - and without starting another repair-and-retest loop.

Make it submission-ready Download the evidence
No added claimsNo missing evidenceNo broken citationsNo second rewrite
PassLab release receipt Ready to submit

Delivery PL-0421

Passed.

Risk directions checked
10 local forecasts
Priorities selected
3 targets
Sentence intervention
2 rewritten / 9 unchanged
Meaning and claim directionPreserved
Facts, names, and numbersLocked
Sources and citationsUnchanged
Grammar and whole-document fitCleared
Every released rewrite cleared all four gates.Representative product state
See the difference

The humanizer loop

A lower AI score is not a finished paper.

Anyone who has used a typical humanizer knows the cycle: the number falls, the writing falls apart, and fixing the writing sends the number straight back up.

01

Score drops

The dashboard looks better.

02

The writing breaks

Meaning shifts. Facts disappear. Sentences stop sounding right.

03

You repair it

You put the argument, evidence, and grammar back.

04

The score rebounds

Now you are back where you started.

If you care even slightly about quality, a detector-friendly rewrite you still have to repair is not a usable result.

The PassLab delivery contract

One pass closes the loop.

PassLab does not hand the unfinished work back to you. Detection, targeted rewriting, protection, and release happen in one controlled path.

  1. 01DetectSee 10 local risk directions together.
  2. 02TargetChoose up to three risks that matter.
  3. 03ProtectReject meaning, fact, citation, grammar, or document regressions.
  4. 04ReleaseDeliver the rewrite only when it clears the original.
Every PassLab rewrite is submission-ready - or the original stays.

What happened after the score dropped

The score dropped.
The paper still failed.

We took the apparent wins and checked the part ordinary benchmarks leave out: did the argument, evidence, citations, grammar, and cross-detector improvement actually survive?

BypassAI / 20 same-source tests

8 -> 0

Eight score wins. Zero usable deliveries.

Eight outputs dropped by at least 10 points. After meaning, quality, and release review, not one qualified.

StealthWriter / 64 matched pairs

21.14 -> 3.33

Big on its own score. Small across other judges.

The average home-score drop was 21.14 points. Across PassLab's broader equal-family view, it compressed to 3.33.

Grammarly / 327 controlled reruns

33 changed

The test method changed the answer.

A corrected waiting and extraction protocol changed 33 competitor records. The older results had to be discarded.

PassLab / release rule

4 gates

A drop is not a win until the paper survives.

Meaning, facts and citations, grammar, and whole-document fit must clear before a rewrite can replace the original.
01AI risk goes down
+
02The paper stays intact
=
PassLabQualified delivery

This is the standard behind “If PassLab passes it, submit it.”

See every experiment, date, method, and limitation

Local scores are PassLab estimates, not official detector results. The current evidence supports these findings, not a completed world ranking.

Download the evidence brief

The claim is only as strong as the audit behind it.

The full brief records what we ran, when we ran it, how competing systems were configured, what counted as a qualified result, and what remains unknown.

Download PDF - Evidence Brief v0.1
PassLab / Evidence 01August 2026

Why PassLab measures deliverability - not only detector movement.

  1. 01Experiment inventory and dates
  2. 02Competitor run protocol
  3. 03Protection-adjusted result definitions
  4. 04Negative evidence and limitations
METHODS
INSIDE

Humanize once. Submit once.

Bring the draft.
Leave with the delivery.

Your first check is free. One-time credit packs start at $3.99. No subscription.

Your text stays yours

Local estimates. Never official detector results.