GLTR's strength lies in its understanding of the differences between human and machine writing. While machines follow predictable patterns, human writing can be unpredictable; therefore GLTR uses GPT-2 to predict which words would likely be used in each position from any input, then color-codes these predictions based on rank (green for top 10 words, yellow for 100th, red for 1000th etc). Words outside these rankings are marked purple; thus creating a secret code which shows its true nature by visual representation of whether or not its likely to have been generated machine generated.

GLTR also generates three histograms that provide aggregate information across an entire text, such as how many words in each color-coded category appear in it; probability ratio between top predicted word and subsequent predicted word; and an "entropy distribution histogram", showing uncertainty level associated with predictions made.

However, even the most powerful entities have their limits; GLTR cannot always detect fake text accurately and is best used in individual cases rather than large-scale abuse detection. Still, in a world where human and machine-generated texts become increasingly indistinct, GLTR remains an invaluable ally: helping us see through disguises while reminding us that AI text generation may be powerful yet it remains critical to identify true creators from fraudulent ones.