On the morning of March 17, 2014, a magnitude-4.4 earthquake struck near Los Angeles. The Los Angeles Times published a short breaking-news item about it within minutes - and the first draft of that item was written not by a reporter but by a program called Quakebot. Quakebot was the work of Ken Schwencke, then a journalist and programmer at the Times, who had built it to watch for alerts from the US Geological Survey.
Schwencke described the system in plain terms. The USGS, he said, has a “wonderful data notification service which sends out emails” whenever a quake occurs. When an alert arrives for an event above a set magnitude, Quakebot extracts the relevant data, drops it into a pre-written template, and files a draft into the Times’ content-management system. A human editor then reviews the draft and decides whether to publish - the algorithm did not push stories live on its own. Schwencke framed the tool as a way to “provide a baseline starting point” so the newsroom could get the basic facts out quickly, not as a replacement for reporters.
Quakebot became one of the most cited early examples of automated journalism. It paired a reliable structured data feed (USGS) with a fixed template and natural-language filling, the same recipe that newsrooms would soon apply to corporate earnings, sports results, and election returns. The key design choice - machine drafts, human approval - anticipated how most newsrooms would later try to govern AI-assisted writing.
Why business readers should care: Quakebot shows the durable pattern for safely automating routine content - automate the data-to-draft step where inputs are structured and reliable, but keep a human in the approval loop. That division of labor, not full autonomy, is what made early automated journalism work.