Urgent.News

What's breaking now, across thousands of outlets.

AI

Do AI Humanizers Actually Work? What Detector Tests Show

Originally published on AI Journal Now . Updated: October 29, 2026 AI humanizers only partly work. They often beat weaker free detectors like ZeroGPT, but strong detectors such as GPTZero, Originality.ai, Winston AI and Copyleaks still catch most humanized text. In our own test, one free humanizer lowered GPTZero’s score from 100% to 29% AI, another left it at 100%, and a carefully prompted AI…

AI humanizers are not foolproof. In tests, they often defy weaker free detectors such as ZeroGPT, but more robust detectors like GPTZero, Originality.ai, Winston AI and Copyleaks still catch most of the humanized text. A free humanizer's rewrite may make the text sound better to a person, yet detectors continue to label it as machine-made.

The most important result is that a humanizer can improve text readability without lowering its AI score. Independent tests reveal that humanizers are generally useful against weak detectors but usually not against strong ones. They often lose detail and create a flat reading experience. The better strategy is to edit the draft using personal examples, data and opinions.

We ran our own test using a 124-word AI-generated paragraph. We ran this paragraph through GPTZero's free web app and got a 100% AI score. After running it through the Clever AI Humanizer and TextToHuman (both free), the score dropped to 29% AI and 100% AI respectively. The TextToHuman version was the most readable and readable. In our test, TextToHuman's own built-in checker rated its own output as "1% human".

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at dev.to →

More in AI

Run Claude Code and Codex side by side. Keep track of both.

One terminal has Claude Code working on a feature. Another has Codex investigating a failing test. You switch back to your editor, and a few minutes later you need to work out where both tasks stand.

  • Two terminals run different tasks: Claude Code and Codex
  • Dashboard ccdeck simplifies monitoring both sessions
  • Provides waiting queue, prompt, tokens, and timing inspection

More from Tuesday 17 November →