How this AI content detector works
Most AI detectors are a mix of two things: a curated list of phrases that large language models overuse, and a set of structural signals that AI-generated text tends to share. This tool uses both. It runs entirely in your browser — no uploads, no signup — and checks your text against eight separate signals: the AI phrase bank, em-dash density, binary rhetorical stops, triadic rhythm, sentence-length variance (burstiness), contraction ratio, personal voice, and corporate-vocabulary density.
The phrases that scream AI
If a piece of text contains “in today’s fast-paced world”, “let’s dive in”, “it’s not just X, it’s Y”, “unlock your potential”, or “delve into”, it’s almost certainly AI-written. These phrases are disproportionately produced by GPT, Claude, Gemini, and DeepSeek because they appear so often in the training data for marketing copy. Removing them is the single biggest win for your score.
Burstiness: the #1 structural tell
Humans write unevenly. A sentence that’s three words long. Then one that sprawls across 30 words and develops a thought fully before landing. AI tends to produce sentences that cluster tightly around 12–18 words — uniform, tidy, and lifeless. When your sentence-length variance drops, detectors pick it up instantly. Mix in short jabs with longer flows and your score drops.
Using this alongside Content Drifter
We built this detector in the same week we rebuilt our own content engine to pass it. Every post generated by Content Drifter is checked against the same heuristics you see here, then automatically rewritten if it scores too high. That’s why our users’ LinkedIn, X, Threads, Instagram and Facebook posts read like they wrote them — because, effectively, they did.
Generate human-sounding posts freeFAQ
Is this really free?
Yes. No signup, no card, no request limit. The tool runs client-side in your browser, so it costs us nothing to offer it.
Can it detect ChatGPT, Claude, DeepSeek, Gemini?
All four models produce output with overlapping phrase patterns and similar structural habits, and the heuristics here target those shared patterns rather than any specific model.
My text is under 20 words. Why no score?
Short snippets don’t contain enough structural signal to score reliably. We require at least 20 words to avoid false positives.
Does raising sensitivity make it more accurate?
Strict mode catches more borderline AI text but also raises false positives on neutral or formal human writing. Balanced (the default) is what we recommend for most social and marketing copy.