AI Detector and Humanizer

1. 1

   Paste your text

   At least 40 words. Longer passages give steadier results.

2. 2

   Check for AI

   The text goes to a third-party AI detector. You see its score, its name and the sentences it flags.

3. 3

   Humanize if needed

   Press Humanize this text to rewrite it with our engine, then check again.

• Predictable wording. AI models choose likely words. When almost every next word is easy to guess, the score goes up. This is often called perplexity.
• Even rhythm. People mix short and long sentences. AI drafts tend to keep them similar from start to finish, which detectors measure as burstiness.
• Stock phrasing. Transitions such as moreover and in conclusion, and set phrases that models use far more than people do.
• Patterns learned from examples. Modern detectors are trained classifiers. They compare your text with large sets of human and AI writing rather than following a fixed list of rules.

For a longer explanation with examples, read our guide How AI Detectors Work.

• 70% and above. Likely AI-written. The detector sees the even rhythm and predictable word choices typical of AI text.
• 35% to 69%. Mixed or unclear. Often AI text that was edited, or human text with a formal, regular style.
• Below 35%. Likely human-written.

A score is a probability, not proof. Detectors make mistakes, especially on short texts, very formal writing and writing by non-native speakers. Never treat one score as the final word on whether someone used AI.

Each detector is trained on different examples and updated on its own schedule, so the same paragraph can score 10% on one and 80% on another. That is why we always show which detector produced the score. If a result matters, check with more than one tool and look at the writing itself.

• Before you submit. If you used AI to help edit a paper, a check shows which sentences still read as machine-made, so you can rewrite them yourself.
• Editors and clients. A first signal on freelance or guest articles. Talk to the writer before deciding anything; never reject work on a score alone.
• Teachers. A score can open a conversation about how an essay was written, next to drafts, notes and what you know of the student.
• A second opinion. If a text was flagged by another tool, see whether a different detector agrees.

1. 1

   Start with the highlights

   A few sentences often carry most of the score. Read the red ones first.

2. 2

   Rewrite them

   Put those sentences in your own words, or press Humanize this text to rewrite the whole passage.

3. 3

   Add what only you know

   An example from your class or work, a figure from a source you read, your own view. This changes a text more than any rewording.

4. 4

   Check once more, then stop

   Chasing a lower number with tiny edits wastes time. Detectors vary, and the writing itself matters more than the score.

• Short texts. Below a few paragraphs, scores swing a lot. We ask for at least 40 words, and longer is better.
• Formal and formulaic writing. Abstracts, lab reports and legal text follow fixed patterns, so human writing in these genres can look machine-made.
• Mixed and edited text. A text that is part AI and part human often lands in the middle. Sentence highlights show which part looks machine-made.
• Languages other than English. Detectors are trained mostly on English and are less reliable in other languages.

• Non-native English writers are flagged most. A 2023 Stanford study found that seven popular detectors labeled more than half of essays by non-native English speakers as AI-generated. In the authors’ words,

  “these detectors consistently misclassify non-native English writing samples as AI-generated, whereas native writing samples are accurately identified”

  Source: Liang et al., Patterns, 2023