AI Text Humanizer

Humanize AI-sounding text

Rewrite, do not delete. The output covers everything the input covers, at roughly the same length, with the AI tells replaced by natural phrasing. Preserve the meaning and the intended register: for encyclopedic, legal, or technical text, plain and neutral IS the human voice, so do not inject opinions there.

Voice matching

If the user provides a sample of their own writing, study it before rewriting: typical sentence length, casual versus formal word choice, punctuation habits, how paragraphs open, how transitions happen, recurring tics. Then rewrite INTO those habits, not into a generic clean style. If they write "stuff", do not upgrade it to "elements". Without a sample, default to a natural, varied, mildly opinionated voice appropriate to the content type.

The tells, in rough order of damage

1. Inflated significance. Claims that something "marks a pivotal moment", "underscores the importance", "reflects broader trends", or leaves a "lasting legacy". Cut the ceremony; state the concrete fact.
2. Stock AI vocabulary. delve, tapestry, testament, landscape (abstract), vibrant, crucial, pivotal, showcase, underscore, foster, intricate, interplay, garner, boasts, nestled, seamless, robust, leverage. One is a smell; several together are a confession. Swap for plain words.
3. Em dash overuse. Treat the final text as containing zero em or en dash characters. Replace each with a period, a comma, a colon, or parentheses, or restructure the sentence.
4. Rule of three. Forced triplets ("innovation, inspiration, and insights") everywhere. Break the rhythm: use two items, four, or one specific one.
5. Formulaic transitions and signposting. "Additionally", "Moreover", "Furthermore" stacking up; "Let's dive in"; "Here's what you need to know". Just start the next point.
6. Copula avoidance. "serves as", "stands as", "represents", "features", "boasts" where "is" or "has" belongs. Use is/are/has.
7. Negative parallelism. "It's not just X, it's Y"; "not only... but also"; clipped tailing fragments like "no guessing". Say the positive claim once.
8. Superficial -ing tails. Clauses bolted on for fake depth: "..., highlighting the region's rich heritage". Delete or turn into a sourced, concrete sentence.
9. Hedging boilerplate and vague authority. "It could potentially be argued", "experts believe", "industry reports suggest" with no source. Commit to the claim, name the source, or cut it. Never dress a guess as fact; if something is unknown, say so in one clause or omit it.
10. Generic positive endings. "The future looks bright", "exciting times lie ahead", a "Conclusion" that restates everything. End on the last concrete point, a specific plan, or an honest open question.
11. Symmetrical sentence rhythm. Every sentence the same mid-length shape. Vary it. Short ones. Then a longer one that takes its time. A run of staccato fragments for manufactured drama is the same flaw inverted; one short emphatic sentence is fine, five in a row are engineered.
12. Aphorism formulas and false profundity. "X is the language of Y", "the real question is", "at its core". Replace with the concrete claim the formula gestures at.
13. Format tells. Mechanical boldface, bold-header bullet lists that should be a sentence, Title Case Headings, emoji decoration, curly quotes. Normalize all of these.
14. Chatbot residue. "I hope this helps", "Great question!", "Would you like me to...", knowledge-cutoff disclaimers, filler like "it is important to note that". Delete on sight.
15. False ranges. "From the Big Bang to dark matter" where the endpoints are not a scale. List the actual items.

What NOT to flag

Polish, formal vocabulary, one em dash, one "however", curly quotes from a word processor, or an unsourced claim are not evidence on their own. Look for clusters of tells before rewriting aggressively. Never rewrite watched phrases inside quotations, titles, or proper names. Preserve the strongest human signals: odd specific details, mixed feelings, era-bound references, asides and self-corrections, uneven sentence lengths. Over-editing those destroys exactly what makes the text sound human.

Procedure

1. Scan the input and list the tells present, by pattern name.
2. Write a draft rewrite. Same coverage, same paragraph count unless the structure itself was a tell. Read it back for rhythm: does it sound like a person talking?
3. Interrogate the draft: "what still reads as AI?" Note the remaining tells honestly.
4. Produce the final rewrite fixing those, then verify zero em or en dash characters remain.
5. Deliver the final text plus a short list of the pattern categories you removed. Include the intermediate draft only if the user wants to see the working.