Phrasly's AI Humanizer rewrites AI-assisted text so it reads naturally. Here's how to get the most natural results, and how to check them before you use them.
1. Use Phrasly Ultra
Phrasly Ultra is our strongest humanizer, included with every paid plan and the 3 Day Access trial. Paid and trial accounts start on Ultra in the AI Humanizer. If you switched to Balanced, pick Phrasly Ultra again from the model menu at the top right of the humanizer. Free accounts use the standard Balanced model.
2. Pick the right mode
Ultra's slider sets how much your text changes. Here is each mode with the bypass rate from Phrasly's own tests (not results for any one detector, and your text may score differently):
Gentle (92.5%): minimal edits, closest to your wording.
Light (94.8%): a light polish that keeps most of your writing.
Balanced (96.1%): natural writing with strong bypass. This is the default and the one we recommend.
Strong (97.1%): a heavier rewrite for text that's still flagged.
Max (98.8%): the strongest rewrite, for tough detectors. Reread it to make sure it still says what you mean.
Start with Balanced. Move to Strong or Max only if a detector still flags the result, and move to Light or Gentle if you want to keep more of your original wording.
3. Give it clean, complete text
Paste the full passage you want rewritten, at least 30 words, rather than single sentences. Each run handles up to 5,000 words and 30,000 characters on a paid plan (800 words on a free account or the trial), so split longer documents into parts.
4. Fine-tune the result
Use the draft arrows under the result to get more drafts of the same text, up to 5 in total including the first. Extra drafts don't use words.
Click any sentence of 8 or more words to see alternatives and swap one in.
On Ultra, the Text settings button (the sliders icon next to the mode slider) has optional settings: more commas, longer or shorter sentences, and cleaner grammar. They nudge the writing; cleaner grammar can lower bypass slightly.
5. Check before you use it
Click Check for AI under the humanized text to run it through the Phrasly AI Detector. The Third-Party AI Scores in the results predict how detectors like Pangram, GPTZero, Turnitin, Originality, ZeroGPT and Copyleaks would likely judge the text. They're predictions, not scans by those services, so if one specific detector matters to you, check the text with that detector too.
Good to know
Detectors differ and change over time, and results vary by text. No tool, Phrasly included, can guarantee a particular score.
Phrasly condemns using its tools for academic dishonesty or cheating, and doesn't support using them to get around educational AI detection systems when they're used to prevent academic misconduct. Always follow your institution's rules on AI use.
Rewrites can occasionally introduce errors, so reread your text before you use it.
See how Ultra compares with other humanizers on our benchmarks page, or read more about it at phrasly.ai/ultra.