AI Content Signal Checker
Paste your text to see the stylistic patterns that tend to show up in AI-generated writing — sentence uniformity, filler phrases, low variety. Use it to improve your own drafts, not to accuse anyone.
This is not a reliable AI detector, and no such thing exists. It reports style signals, never a verdict. Even OpenAI’s own detector was correct on just 26% of AI text and falsely flagged 9% of human writing before being shut down. Do not use any score here — or from any tool — to make a decision about a real person. Your text is analysed locally and never uploaded.
Your text
Paste at least a few paragraphs — short samples are meaningless.
Signal report
A style profile — read it as “worth a second look,” not “caught.”
The honest facts about AI detection
Before you trust any AI-detection score — including this one — here’s what the research actually says.
The people who built the AI couldn’t detect the AI
OpenAI released a text classifier in January 2023 and shut it down six months later, on 20 July 2023, “due to its low rate of accuracy.” Its own published numbers: it correctly caught only 26% of AI text and wrongly flagged 9% of human text. As of early 2026 it has never been revived.
If the company that made the model can’t reliably detect the model’s output, be very skeptical of any tool that claims it can.
False positives fall hardest on non-native English writers
A Stanford study ran seven detectors over essays written by non-native English speakers. The average false-positive rate was 61.3%, and at least one detector flagged 97.8% of them as AI. The cause is baked into how these tools work: they reward “surprising” word choices, so simpler, more predictable vocabulary — common in second-language writing — gets misread as machine-made.
Detectors have also flagged the U.S. Constitution and passages of Shakespeare as AI-written.
How these tools actually work (perplexity & burstiness)
Text detectors don’t “understand” anything. They measure statistical patterns — mainly perplexity (how predictable each word is) and burstiness (how much sentence length and rhythm vary). AI text tends to be smooth and predictable; human text tends to be lumpier. This tool uses the same family of signals, which is exactly why it can’t be authoritative.
And the patterns are easy to break: paraphrasing, swapping in a few unusual words, or running text through a “humanizer” drops most scores sharply. It’s an arms race detection is losing.
Why serious institutions won’t rely on them
UCLA and other universities declined to switch on Turnitin’s AI detector over accuracy and false-positive concerns. Academic and legal-research guides now advise that detector output must not be the sole basis for an accusation, and should always be paired with human judgement and conversation.
The practical takeaway for publishers: a “score” is a starting point for editing, never proof of authorship.
Does AI content hurt SEO? (the more useful question)
Search engines don’t penalise content simply for being AI-assisted. What gets penalised is low-quality, unhelpful, mass-produced content — regardless of who or what wrote it. With roughly three-quarters of new web pages now containing some AI-generated text, “was this written by AI?” is a far less useful question than “is this genuinely helpful, accurate, and original?”
That’s the real job of a tool like this: not to hunt AI, but to catch the generic, filler-heavy writing that reads as low-effort — and fix it.