GLTR
A visualizer that reveals AI text by showing how predictable each word is.
GLTR takes a different angle from the percentage-generating detectors. Built by researchers at MIT and Harvard, it visualizes how predictable each word in a passage is by showing the ranking of the actual next word against the model's top predictions. Words the model would have chosen are highlighted green, yellow, or red, so you can literally see whether a text follows the most likely path a language model would take. It is free and open-source, and it is best understood as a teaching and research instrument rather than a pass/fail checker. For educators who want students to grasp why AI text looks the way it does, the visual is more instructive than a single score. The limitation is that it focuses on GPT-2-era text and works best on English, so it is a complement to, not a replacement for, modern detectors.
| Best for | Researchers and educators who want to see AI text, not just score it |
|---|---|
| Pricing | Free / open-source |
| Category | AI Detectors |
Highlights
- Word-level predictability
- Open-source
- Teaching visual