Was this written by AI?
Paste a passage below and get a straight, three-way answer — human-written, mostly human with possible AI assistance, or likely AI-generated. Free, no sign-up, no fake precision.
How the check works
Modern AI writing has fingerprints. Language models favour the most probable next word again and again, which leaves text that is unusually smooth and predictable: sentences of similar length, even rhythm, common transitions, few surprises. Human writing is burstier — a short blunt sentence next to a long wandering one, an odd word choice, a half-finished thought. The checker reads a text the way the engine behind HalluSpot, Arbiter, does: it scores the whole passage and its individual sentences for these patterns, then weighs them together into one of three conclusions.
That is also why the answer comes in three bands rather than a single percentage. A score collapses a lot of uncertainty into a number that looks exact but is not. A band tells you what the evidence actually supports — and when the text sits in the middle band, it says so instead of pretending to know.
What it cannot tell you
Detection is pattern-matching, not mind-reading, and it has real limits. Short texts carry too little signal. Heavily edited AI text drifts toward human patterns, and heavily templated human text — lab reports, formal applications, anything written to a strict formula — drifts toward machine ones. Writers working in a second language are flagged more often by detectors generally, because a smaller working vocabulary produces more predictable sentences. None of this makes the tool useless; it makes it one input among several. Anyone using a result to judge a student, an employee, or a freelancer owes them the rest of the evidence too.
Polishing is not ghostwriting
The question behind most checks is not really “did a machine touch this?” but “whose work is this?” Those are different questions. Running your own draft through a grammar tool, or asking an AI to tighten a clumsy paragraph you wrote, leaves the ideas, structure, and voice yours. Handing over a prompt and submitting whatever comes back does not. Most university policies are written around exactly that distinction — assistance with your text is broadly allowed, presenting generated text as your own is misconduct. This checker measures AI involvement; the judgement about what that involvement means is still a human one.
Read further
How to tell if something was written by AI
The visible signs, the invisible ones, and a practical order to check them in — with or without a tool.
Can AI detectors be wrong?
Yes — here is how false positives happen, who they hit hardest, and why a score alone should never decide anything.
AI polishing vs. AI writing
Where universities actually draw the line between allowed assistance and misconduct, in their own policy language.
Frequently asked questions
Can this checker prove a text was written by AI?
No — and no honest checker can. Detection looks at statistical patterns in the writing, not at the writing process itself. A result is evidence to weigh alongside everything else you know about the text and its author, never a verdict on its own.
What is the difference between the three results?
“Human-written” means the text shows no significant patterns of AI-generated writing. “Mostly human — possible AI assistance” means the text reads as human-authored but carries surface patterns consistent with light help such as grammar cleanup or polishing. “Likely AI-generated” means most of the text follows the patterns of machine-generated writing.
Why don't I see an exact percentage?
A single number looks more precise than detection really is. For clearly human texts we show the conclusion without a score, because a low number like “4% AI” invites false confidence in both directions. The full report on HalluSpot includes exact scores with the sentence-level evidence behind them.
I wrote this myself — why was it flagged?
False positives happen. Formulaic genres (lab reports, legal boilerplate, cover letters), very regular writing habits, and writing in a second language can all resemble machine patterns. Read our guide on how detectors get it wrong, and treat any single result as a prompt for a conversation, not an accusation.
Is polishing my own writing with AI the same as having AI write it?
No. Most universities draw exactly this line: basic assistance and polishing of your own draft are usually acceptable, while submitting AI-generated text as your own work is not. Where the line sits varies by institution — our polishing vs. writing guide walks through the common policy wording.
Do you store the text I paste?
The free checker sends your text to the HalluSpot scanning service to produce a result. Don't paste anything confidential. The full service's data handling is described in its privacy policy on halluspot.com.