wroiter

AI rewriter that respects your intelligence

AI detectors give you a score. Humanizers give you worse writing. WROITER gives you a prompt that fixes exactly what's wrong.

Run the Diagnostic — it's free Learn how it works

No cost. No account. No limits.

A 3AM ENERGY PROJECT
The problem

You brain-dumped a great idea to your LLM to refine it into something readable, and now it sounds like slop.

You used AI to draft an email reply (yes, you do it too) and now you wonder if it sounds like slop.

You did extensive research, handed it to AI for a rewrite, and now—guess how it sounds.

“Oh, this is AI-generated content—I'm reading all of it!”
— said no one ever

Possible solutions
  • Rewrite manually — if you have the energy and an hour to spare
  • Use an AI detector / humanizer — watch your text get rewritten into something you didn't want
  • Ask your AI to “make it less AI” — it'll drop the most obvious lines and miss everything else
  • Use WROITER — get a revision prompt and full control of what gets revised

WROITER is not a tool to help you bypass AI detectors. It's designed & constantly refined to help you create content that people won't close the door on in 5 seconds.

Under the hood

Why trust WROITER

0

AI in the detector

No LLM calls, no neural network, and no probability guesses. The detector can't hallucinate because there's nothing to hallucinate with — just deterministic pattern matching that runs in your browser.

Tested against real literature, again and again

Moby-Dick, 1984, Pride and Prejudice — we run real human writing through every update and tune until the false positives drop. When the algorithm flags Melville as a machine, we fine-tune it again.

Recent algorithm updates
v1.6.0 Revision engine live — per-instance fixes, not just flags
v1.5.0 Anti-AI-writing detectors. 31 total, 6 families
v1.4.0 Removed the 0–100 score — findings, not a verdict
Full detection engine changelog →

The most comprehensive AI pattern library on the web

Most detectors publish a word list and call it a day. We publish the entire method, the proprietary research — and constantly update the pattern library.

The product

This is what a real revision prompt looks like.

Run WROITER’s built-in slop sample and it returns 32 revisions across three severity tiers. Each one points at a specific phrase, says what to change and why, and flags how mechanical the fix is. Here are four — rendered exactly as the tool shows them, not a cleaned-up mockup.

AI-typical phrase template BANNED_PHRASES high confidence

Findit's important to note

ActionDelete "it's important to note" and start the sentence with the substantive claim that follows. If the sentence has nothing left after the deletion, the sentence had nothing to say.

WhyTemplate phrases defer commitment to a specific claim. Detectors flag them at the lexical level and editors flag them at the reading level; removing them sharpens both at once.

Hedging language HEDGING high confidence

Findit could be argued

ActionDelete the hedge "it could be argued". If the surrounding claim is true, state it directly. If it isn’t, name the specific uncertainty — a source, a counter-example — instead of softening with a vague hedge.

WhyAI text hedges to feel safe rather than to mark real uncertainty. Stacked hedges blur a claim instead of qualifying it.

Sequence-marker scaffolding OVER_SIGNPOST high confidence

FindFirst

ActionRemove "First". The reader can see this is the next paragraph; they don’t need a label that says so. If the order matters, the content carries it.

WhyStacked sequence markers ("First … Second … Finally") signal essay scaffolding that survived into the final draft.

AI-scented vocabulary BANNED_WORDS medium confidence

Findlandscape

ActionReplace "landscape" with a more specific noun that names what’s actually happening. The current word adds register without specifying meaning — pick the concrete word it was standing in for.

WhyMedium confidence: the instruction is mechanical, but choosing the replacement needs you to read the surrounding sentence. That’s the line between “apply directly” and “your call.”

…and 28 more across the medium and low tiers. Every flag carries its own severity and a confidence badge — the badge is the honest part: it tells you which fixes are mechanical and which are your editorial call.

93
Actionable mean before
→
0
Actionable mean after

Latest public revision benchmark: 5 of 24 fresh Polygraf AI drafts generated actionable prompts, and those drafts fell from a mean score of 93 to 0 after one pass. The other 19 drafts produced no prompt and stayed frozen unchanged.

Try it with your own text →
Questions you're already asking
Changelog

What shipped recently

May 29
The revision prompt went live

The diagnostic stopped showing a sample and started doing the work: it points at each flagged phrase, says what to change and why, flags how mechanical the fix is, and hands you a prompt for any AI — or you work through it by hand.

May 28
Every detector learned to write its own fix

28 of 31 patterns now generate per-instance revision instructions, grouped by severity with a confidence level on each — high means the fix is mechanical, medium means it’s your editorial call.

May 28
Recalibrated against a fresh corpus

Re-ran the full calibration harness — books, essays, journalism, Reddit, and current AI samples — and published the numbers and the method, including the three detectors the corpus can’t exercise yet.

May 27
Removed the 0–100 score

WROITER shows a profile of detected patterns now, not a verdict. Every flag points at specific text you can check yourself — auditable in a way a single number never was.

May 27
New detectors for AI’s tells

Added credibility theater (“let me be direct”), telegraphed reveals (“the takeaway:”), and manufactured drama (“buried in the changelog”) — patterns distilled from editing real AI drafts.

Next up: better voice control, richer research support, team workflows, and eventually API access. The method stays public, and user-facing changes go in the changelog.

Read the changelog See calibration log