From ''Search'' to ''Generation'', from Links to ''Summaries.'' The Internet''s Knowledge Structure is Changing.

In October 2025, the Wikimedia Foundation published findings: since May 2025, global Wikipedia human visitor traffic declined approximately 8% YoY. This isn''t simple usage decline — it signals AI and social media fundamentally changing how the internet consumes information. People no longer "search Wikipedia" — instead AI reads Wikipedia and delivers only summarized answers to users. Bot traffic paradox: Wikimedia detected abnormal traffic surge in spring 2025 (initially seeming like user increase, revealed as AI crawlers and data collection bots); when bot detection logic updated, significant "human visit" traffic was reclassified as AI bot access, revealing the ~8% human decline. The knowledge distribution severance: AI grows on Wikipedia — ChatGPT, Gemini, Claude, Perplexity, Copilot all use Wikipedia data as core training corpus (humanity''s most extensive structured, verifiable, neutral open dataset); but AI absorbs Wikipedia into its model weights without returning "links" to the source to users. The "free riding on public goods" problem: Wikipedia maintained by donations and volunteer editors — a digital public commons; AI companies train freely on this commons but their value extraction is creating a structural problem where "the information source is open but distribution is closed." Wikipedia''s sustainability challenge: if human traffic (and thus donor/contributor motivation) continues declining while AI training demand grows, the incentive structure sustaining Wikipedia''s volunteer editing ecosystem may collapse — creating a tragedy of the commons where AI companies extract value from a resource their use is undermining. Proposed solutions: AI training data compensation frameworks; mandatory attribution in AI-generated answers; Wikipedia API rate limiting for commercial AI training; direct licensing agreements with AI companies.