The Consensus, a site that promotes good technical writing in software infrastructure, has built a detector to flag articles that may read like AI-generated writing. The site features articles in its daily At a Glance section, its weekly newsletter and its third-party feed, and it has found that readers strongly dislike AI-written prose, even when it is technically polished.
The editor, Phil Eaton, explains that the team had occasionally featured articles that drew criticism from commenters for sounding AI-generated, some from companies that publicly advocated responsible AI use. Commercial detectors such as Pangram are accurate but too expensive to run on every article, so the site built a free, rule-based tool instead. It compares writing features before and after 2023 and does not claim an article was written by AI. Because it is cheap and sometimes wrong, even some of the editor's own articles have been flagged.
Articles scoring in the 99th percentile for AI tells are excluded from the homepage At a Glance section, social media, the sparkle icon and the newsletter. Readers can still see them by filtering the feed. The editor acknowledges that the tool could be gamed and says that distinctive, company-specific writing voices would be welcome.
The Consensus is not banning AI-assisted writing. Its goal is to freely promote articles that are thoughtfully written by humans, while leaving readers free to choose what they see.
Source: Hacker News · Summarized by HeadlinesBriefing