We will argue against the obvious use of our own tool here. For authors writing fiction, an AI humanizer is usually the wrong thing to run on your prose, and the reason is structural rather than a matter of quality.
A rewrite works by replacing unusual choices with less unusual ones. Voice is unusual choices. So the mechanism that lowers a detector score is the same mechanism that sands off the thing a reader came for. There are four places this genuinely helps an author, and your manuscript is not one of them.
Which is a description of what voice is made of.
Detectors read how predictable each next word is. A humanizer lowers a score partly by substituting less predictable wording, and that sounds like it should help a stylist. It does the opposite, because the substitution has no concept of why your word was the right one.
Your character does not "walk quickly across the room"; she has a specific way of crossing a room that took you three drafts to find. A rewrite sees an adverb-verb pairing and offers an alternative with no knowledge that the oddness was load-bearing. What comes back is usually more ordinary and slightly wrong, and it reads as translated.
Rhythm you chose. A deliberately long sentence followed by two words is a decision. A humanizer redistributes sentence lengths as a goal in itself and will undo it.
Dialogue voice. Characters are differentiated by how they speak: syntax, register, the words they would not use. A rewrite normalises toward the same middle register, which flattens two characters into one.
Deliberate repetition. Repeating a phrase for effect looks exactly like the pattern a rewrite is built to break up.
If you want the detector reading on your own prose out of curiosity, run the detector and skip the rewrite. That is free, three scans a day at up to 1,000 words, and it costs your manuscript nothing.
All of them outside the manuscript, and all of them real.
1. Blurbs and back-cover copy. This is marketing writing with a house register, not your voice, and AI drafts of it are notoriously full of "a gripping tale of love and loss". Cutting that filler is exactly what a rewrite is good at.
2. Synopses. A synopsis is a summary, and summary prose is naturally flat and even, which is both unpleasant to read and the shape detectors flag. Varying the rhythm genuinely improves it. Watch for moved plot details, which matters more here than in most writing: a synopsis that misstates your ending is worse than a dull one.
3. Query and submission letters. If you drafted one with AI, agents read hundreds of these and the register is recognisable. Worth de-filleting. The comparable titles and the specifics about your book are the part no tool supplies.
4. Newsletters, posts and author-site copy. Volume writing around the book where voice matters less and filler accumulates fast. This is the one place an AI humanizer earns its place in an author's workflow week to week.
A rewrite can invent. In our testing one pass returned "Roughly a year into it" when nothing in the source mentioned a timescale. In a synopsis or a query letter, a fabricated detail about your own book is a bad thing to send an agent. Read the highlighted changes rather than the prose.
On settings: Light varies rhythm and cuts filler and leaves your vocabulary alone, which is the only one worth considering anywhere near creative work. Balanced reshapes phrasing broadly. Maximum replaces most of the wording, is a paid mode, and on anything voice-driven it is the wrong choice.
Stated narrowly, because we are not in a position to characterise the industry.
Submission terms increasingly include a declaration about AI use, and some platforms require disclosure. What any particular publisher, agent or platform requires is a question for their submission guidelines, not for a vendor page, and they differ. We are not going to summarise an industry position we have not surveyed.
What we can say about the tooling, ours included: a humanizer score is not evidence of how text was produced, and nor is a detector score. Detectors disagree with each other on identical passages and they false-positive on human writing, which for novelists has a particular edge, since literary prose is often stylistically unusual in ways that confuse these instruments in both directions. The evidence on false positives is here.
And the practical point that outlasts all of it: keep your drafts. A manuscript with a visible history across months of versions is the strongest answer to a question about authorship that exists, far stronger than any score. If you have been accused, do not rewrite anything. Gather the evidence.
Generally no, and we will say that about our own humanizer. A rewrite lowers a detector score partly by replacing unusual word choices with less unusual ones, and voice is made of unusual choices. The mechanism that moves the score is the same one that sands off what a reader came for. Rhythm you chose deliberately gets redistributed, dialogue normalises toward one middle register, and repetition for effect looks exactly like the pattern a rewrite exists to break up.
Four places, all outside the manuscript. Blurbs and back-cover copy, which is marketing register rather than your voice and where AI drafts are dense with filler. Synopses, which are summary prose and therefore naturally flat. Query and submission letters, where the AI register is recognisable to agents who read hundreds. And newsletters, posts and author-site copy, where volume is high and voice matters less.
It will flatten it. Characters are differentiated by syntax, register and the words they would not use, and a rewrite normalises toward a common middle register with no concept of who is speaking. Two distinctly voiced characters come back sounding like one person. If you want to run anything near creative work, use the lightest setting, work on a copy, and read what changed line by line rather than accepting the output.
Submission terms increasingly include declarations about AI use and some platforms require disclosure, but what any specific publisher, agent or platform does is in their submission guidelines rather than something we can tell you, and they differ. What we can say about the tooling is that a detector score is not evidence of how text was produced: detectors disagree with each other on identical passages and they false-positive on human writing, which has a particular edge for literary prose that is stylistically unusual by design.
Yes, and for an author that is usually the sensible half. The detector runs free with no account at three scans a day up to 1,000 words, so you can get a reading on a chapter out of curiosity without a rewrite touching your prose. Treat the number as one instrument rather than a verdict, especially if your style is deliberately unusual.
Your drafts. A manuscript with a visible version history across months, with notes and dated files and the chapter you restructured twice, is a far stronger answer than any detector score, because it shows the work accumulating rather than asserting a probability about the finished text. The corollary matters: if you are already being questioned, do not rewrite, because that replaces the record that would have defended you.
Four questions hiding in one, answered separately.
Read the answers →Why unusual prose confuses these instruments both ways.
Read the evidence →What a draft history evidences that no score can.
See the steps →Three scans a day up to 1,000 words, no rewrite, no account.
Check a passage →Three rewrites a day, no account. Work on a copy, and read the highlighted changes.