Free · Unlimited · Client-side

Is your text
AI-generated?

Paste any text and instantly see the AI score. Eight pattern checks run in your browser — no server, no login, no limits. See which stylistic patterns the checker finds.

8 pattern checks Zero data sent Works offline
0 words · free · unlimited

False positives

Falsely accused
of using AI?

AI detectors make mistakes — serious ones. A 2025 Yale case became a flashpoint when students faced academic discipline based solely on detector output. None were found responsible.

Research shows 5–20% false positive rates. Non-native speakers are hit hardest — formal academic writing in a second language shares structural patterns with AI output. This tool lists which stylistic patterns it found, so you can see how your text reads; it is a heuristic, not a verdict.

5–20%
False positive rate
Documented in peer-reviewed research
ESL
Most at risk
Second-language writers flagged disproportionately
Yale 2025 precedent
Students faced discipline based solely on AI detector output. No findings against any student. Detector alone is not proof.

8 pattern checks

What the detector
actually measures.

HIGH
Sentence variance
AI produces uniform sentence lengths (CV < 25%). Humans mix short punchy sentences with longer complex ones.
HIGH
AI vocabulary
Words like "furthermore," "delve," "robust," "seamless," "leverage" are drastically overrepresented in LLM output.
HIGH
Transition density
"However," "therefore," "consequently" appearing more than once per 50 words signals a model following an outline.
MEDIUM
Passive voice
AI favors passive constructions at rates higher than most human writers. 40%+ passive sentences is notable.
MEDIUM
No short sentences
Humans write "That was it." or "She left." AI defaults to complete multi-clause constructions every time.
MEDIUM
Em dash overuse
GPT-4 and Claude reach for em dashes as a stylistic default. 3+ in a short text is reliable signal.
MEDIUM
Parallel lists
Numbered or bulleted lists with grammatically parallel items are an AI organizational fingerprint.
LOW
Missing contractions
Humans use contractions constantly. AI defaults to formal full forms unless explicitly instructed otherwise.

Free

Why is this
completely free?

The detection logic runs as JavaScript in your browser. When you click "Check for AI," no data is sent anywhere. No server processing, no API call, no database. The marginal cost of one more user running one more check is exactly $0.00.

Other tools that charge for AI detection are either using server-side models (which have real costs) or charging because they can. We keep the detector free because it is a simple browser-side check. The paid humanizer is a separate tool that uses an LLM to rewrite text.

Want to understand how AI detectors work and where they fall short? The Refrazr blog covers the research — including a 100-essay test of GPTZero accuracy.

$0
per check
∞
unlimited
0 B
data sent

FAQ

Common questions.

How accurate is this AI detector?
The detector checks eight structural signals that LLM output consistently shows: low sentence length variance, AI vocabulary, heavy transition phrase usage, passive voice, absence of short sentences, em dash overuse, parallel list structures, and missing contractions. It's a pattern-based screen, not a trained detection model, so read the score as a signal of how AI-typical your text looks rather than a verdict on how it was written. The analysis runs instantly in your browser — no server model involved.
Can AI detectors give false positives?
Yes. False positive rates of 5–20% are well-documented in academic research. A 2025 Yale case drew national attention after a professor accused students based solely on detector output — none were found responsible. Non-native English speakers are disproportionately flagged because formal academic writing from second-language speakers shares structural patterns with AI output. This checker only lists the stylistic patterns it finds; it is a heuristic, not a verdict on how a text was written.
What AI models can this detect?
The detector is model-agnostic. It looks for structural patterns common to all large language models — ChatGPT, Claude, Gemini, Llama, Mistral, and others. These models share training objectives and produce similar fingerprints: even sentence lengths, formulaic transitions, hedging vocabulary, parallel clauses. The specific model matters less than the patterns it leaves behind.
Is this detector free? Are there limits?
Completely free with no limits. The detection runs 100% client-side in your browser — no server calls, no API keys, no usage tracking. We can offer this because it costs us literally nothing to run. The only paid features are on the humanizer side, where we use an LLM to rewrite your text.
Does the detector store or log my text?
No. The text never leaves your browser. The analysis is pure JavaScript running locally. You can verify this by checking your browser's network tab: zero outbound requests when you click Check for AI.
My text scored high but I wrote it myself. What should I do?
A high score does not mean the text was written by AI. This checker is a pattern-based heuristic that counts surface features such as sentence-length variance, transition words and contractions, and careful human writing — especially formal or second-language writing — can trip it. Treat the result as a hint, not a verdict.

Want a more
natural-sounding draft?

Refrazr rewrites AI-assisted text for rhythm and voice — deep structural rewriting, not synonym swap. This checker is a heuristic, not a verdict. Free tier available.

Try Refrazr free →