Does this text look like a robot wrote it? Probably.
Honest disclaimer up front: This is a heuristic toy, not a reliable AI detector. Frontier models have moved well past the obvious tells, and plenty of humans write in patterns this tool flags. Take the score as entertainment, not evidence. We're aware of the irony that an AI helped build this.
Results are based on surface-level heuristics: em-dash frequency, stock transition phrases, sentence-length variance, and a small vocabulary of AI-favoured words. These signals are neither necessary nor sufficient to identify AI text — they just correlate weakly with it. A high score means this text is consistent with AI-generated writing. It does not mean it is AI-generated writing.
Short answer: not reliably, no. The longer answer involves a lot of caveats and ends in the same place.
Early GPT-2-era AI text had fairly consistent tells — unusual em-dash usage, a preference for certain transitional phrases ("it's important to note that", "delve into", "in conclusion"), suspiciously uniform sentence lengths, and a particular kind of hedged corporate blandness that humans rarely produce unironically. Detectors trained on this worked reasonably well against those models.
The current generation of frontier models has been trained on massive amounts of human writing and in many cases explicitly fine-tuned to avoid the obvious tells. The signal-to-noise ratio for surface-level heuristics has dropped significantly. Worse, there's a meaningful false-positive problem: some humans genuinely write in patterns that trigger AI detectors, especially non-native speakers and writers who've been influenced by corporate or academic style guides.
This tool checks for some of the older, weaker signals. It might catch lazy AI output that hasn't been edited. It will definitely miss well-edited AI output. Treat the score accordingly — it's a starting point for your own critical reading, not a verdict.