Free Article Spinner
Rephrase, Rewrite and Paraphrase Sentences, Paragraphs or Whole Articles for FreeHere's a number that explains why this topic won't go away. In the 2026 HEPI survey of 1,054 UK undergraduates, 94% said they use generative AI to help with assessed work. The share who paste AI-written text straight into an assessment went from 3% to 12% in two years. Most students aren't asking whether to use these tools anymore. They're asking where the line is.
Three kinds of software sit in the middle of that question. Plagiarism scanners compare your text with sources that already exist. AI detectors guess whether text looks like it came from a machine. And AI humanizers rewrite text so it sounds more natural, which often lowers its detector score. People keep asking all three questions they were never built to answer.
Report | What it can tell you | What it can't tell you |
|---|---|---|
Similarity scan | Which passages match existing sources | Who wrote the text, or whether AI did |
AI detector score | How much the style resembles machine writing | Who wrote it, and there's no source to check |
Humanizer output | A rewrite that sounds more human | Whether the ideas were yours |
Every row says the same thing. These tools measure text. Misconduct is about conduct, and no score can see that.
Not quite. A similarity scan is honest about what it does: it shows you its matches, and a person can judge them. But text written by an AI usually matches nothing, because those exact sentences don't exist anywhere else. The same goes for made-up citations. A reference a model invented won't match any real page. Turnitin itself describes its similarity score and its AI writing percentage as two separate measures that don't affect each other.
Detectors mostly look at predictability, meaning how likely each word choice is and how evenly the sentences flow. You can disturb that pattern without adding a single original thought, and that's essentially what a humanizer does. Turnitin added bypasser detection in August 2025 (English only) because those tools can nudge machine text toward the human range. One arXiv preprint reported that after a popular humanizer, fewer than 4% of AI-labelled rewrites stayed flagged. It's one study, and results will vary by tool and version, but the direction is clear. A low score isn't proof of authorship.
This one does the most damage. A 2023 study in the journal Patterns ran seven detectors on 91 essays written by non-native English speakers. On average, 61.3% of them were labelled AI-generated, while essays by native-speaking eighth graders were classified almost perfectly. People writing in a second language lean on common words and familiar phrasing, and detectors read that as machine-like. When the researchers reworked the same essays with more idiomatic wording, the false-positive rate dropped by about 49 percentage points. Separately, an experiment on Turnitin's indicator found AI scores anywhere from 0% to 20% in groups that hadn't used generative AI at all, and advised allowing for that margin.
Some are sold that way, with promises of "undetectable" output. Others work more like a style editor. The software can be identical in both cases, which means the software can't be the thing that decides.
Plagiarism means taking someone else's words or ideas. An AI-written essay doesn't take anyone's words, so the old term feels stretched. That's why many institutions now talk about misrepresentation: giving your reader a false picture of the work. A draft can misrepresent four things:
Authorship. Claiming ideas and arguments that a machine came up with.
Process. Hiding help your course asked you to disclose.
Sources. Citing articles you never read, or references a model made up.
Understanding. Handing in polish you couldn't explain or defend.
Intent matters. Forgetting a quotation mark is a mistake. Rewriting machine text again and again until a detector reads zero is a deliberate move to avoid being found out, and most policies treat that kind of concealment as making things worse.
Three questions settle most situations.
Do the rules allow it? Silence isn't permission, and some institutions restrict AI rewording tools even on text you wrote yourself.
Are the ideas yours? If a machine supplied the argument and structure and your course forbids that, rewording the surface doesn't change who wrote it.
Would you be comfortable showing how you made it? If your instructor watched you work, would they call the paper a fair picture of what you can do? If your plan depends on them never finding out, you already know the answer.
Running your own honest writing through a rewriting tool just to lower a score is a gray area. Your authorship is truthful, but the tool may be off limits, the rewrite can drift from what you meant, and the draft history can start to look like concealment.
Vendor pages love to treat detection like a puzzle you can beat. Before you trust one, ask who tested the claim, whether it reports errors in both directions, and which detector versions it used, since those tools update all the time. Most of all, ask whether the promise answers the right question. Text that no detector can flag still misrepresents authorship if the rules say AI can't draft it.
The best protection is a process people can see. Reviewers increasingly look at drafts, revision patterns, pasted content and source trails, and some writing platforms record all of this automatically.
Write somewhere with version history, and don't paste a finished draft into a blank page at the end.
Keep dated outlines, notes and drafts, along with the sources you actually read.
Jot down an AI log: which tool, when, why, and what you did with the output.
Disclose in plain words when asked: "I wrote this myself and used an AI tool to suggest alternative phrasings, which I reviewed and edited."
Ask first if the policy is silent, and save the reply.
If you do get flagged, take a breath. Keep every file, then answer in writing with your drafts and a clear account of how you worked. A score is an estimate, not proof. If your writing style tends to get flagged, say so early.
A detector score should make a reviewer look closer, not reach a verdict. Check that the cited sources exist and say what the paper claims. Read the draft history. Then ask the student to walk you through the argument, because someone who wrote it usually finds that easy. Set clear expectations too: put the AI policy in the syllabus and the assignment brief, ask for outlines or drafts along the way, and make disclosure simple and free of punishment.
Not every piece of evidence deserves equal weight. A detector score, a sudden style change or unusually smooth prose are weak signals that justify a closer look and nothing more. A close paraphrase in the similarity report or references that can't be found are moderate signals. A draft history showing the whole text pasted at once, fabricated citations, or a student who can't explain the core claims are strong ones.
Detection tools will keep changing. What won't change is the expectation that you're honest about who did the work. Check the AI policy for your next assignment, and if it says nothing, send your instructor a quick two-line email today. Then start your draft somewhere that saves its history, and write the first paragraph yourself.