Wikipedia Seems Pretty Worried About AI (nymag.com)

Wikipedia’s community team discovered that an apparent surge in traffic around May–June 2025 wasn’t a wave of new human readers but sophisticated bots. After updating bot-detection logic and reclassifying March–August data, they found much of the unusual activity came from automated agents built to evade detection and scrape Wikipedia for training or summarization. Once those requests were removed from “human” counts, Wikipedia’s human pageviews show an approximate 8% decline versus the same months in 2024, a meaningful drop for a volunteer-funded site that relies on visits to attract editors and donors. The episode highlights a broader AI-era tension: large language models and AI-powered search are heavily consuming curated web resources like Wikipedia and returning compact summaries that users accept instead of clicking through. Technically, the story underscores two problems — evasive scraping bots that defeat traditional detection and the downstream effect of model outputs reducing referral traffic and contributor incentives. Implications for AI/ML include urgent needs for better traffic attribution and bot-detection, clearer data-provenance and licensing practices (or API partnerships), and engineering fixes like rate limits, watermarks, or access controls to balance model training needs with the survival of public knowledge infrastructure.