Watch: Engadget: Quantitative Linguistic Analysis

Can you analyze eighteen years of online journalism without reading a single word? By using nothing but data compression and basic information theory, we can map the history of the website Engadget. Before analyzing the writing, this math-only approach solved a major data-quality problem. Out of more than thirteen thousand scraped web pages, the compression algorithm instantly separated them into two distinct groups. It revealed that sixty percent of the pages were actually empty product templates rather than real stories. This left five thousand genuine articles, spanning from 2006 to 2023. On this cleaned-up collection, the data reveals a clear editorial arc. Instead of getting steadily longer over time, Engadget's articles hit a long-form peak in the mid-2010s, averaging around ten thousand characters, before retreating to about half that length in recent years. How easily these articles compress closely tracks their length, because longer texts naturally repeat themselves more. But even when we adjust for length, recent articles are still more compressible than older ones, suggesting a subtle drift in writing or website layout. Meanwhile, the basic variety of characters and symbols remained remarkably stable over eighteen years, only showing major volatility during the site's early, formative years. This method cannot tell us why these changes happened, but it proves that basic math can map the lifecycle of a digital publication with absolute precision, all without reading a single word.