Long-term published trends in readership, questions, coding and crawler rules. Source definitions and coverage can change over time. Launch markers provide context; these observational series do not isolate the effects of AI.
50.7%
of latest reported Wikimedia views classified as bots
Sep 2026 · declared crawlers plus undeclared automation
−15%
human Wikimedia reads vs three years earlier
Sep 2026 against Sep 2023
−99%
Stack Overflow questions per month since ChatGPT
1,143 in Sep 2026 vs 109,188 in Nov 2022
78.1%
Google's share of searches
Sep 2026 · 83.8% in Nov 2022 · StatCounter, quoted
Every request to every Wikimedia project, per month, split by the Foundation's own classifier: humans, declared crawlers (Googlebot, GPTBot and the like, by user agent), and undeclared automation (traffic that behaves like a bot while claiming to be a browser; classified since 2020). Automation categories include conventional bots as well as possible AI traffic, so these series cannot isolate AI scraping.
Views per month across all Wikimedia projects, all access methods. Rung: quoted source (Wikimedia Analytics).
Questions asked on Stack Overflow each month, counted from the public API. Changes in posting can have many causes; this series does not show where people sought answers instead.
Counts as the API reports them today, so months lose deleted questions over time. Rung: quoted source.
Coding: pull requests on GitHub, all and by agents
The GH Archive census counts every pull request opened on GitHub per month; agent pull requests are the ones opened by a known coding-agent account or on an agent branch prefix. The years before 2025 are the human baseline; the AI co-author trailers in commit messages during that time are on the census page.
Pale bars: months where GH Archive captured only part of GitHub's feed. Rung: bot account and branch fingerprint. Earliest month counted so far: Jan 2022.
Crawling: who the web blocks, search engines against AI
Share of sampled sites whose robots.txt fully blocks each crawler, per Common Crawl crawl. Googlebot and Bingbot have been blockable for two decades and almost never are; the AI crawlers went from nonexistent to the most-blocked identities on the web within three years.
Nothing recorded yet. The sensor is live; data appears here as it arrives.
StatCounter's monthly search-engine market share, worldwide, from browser-side sampling on its member sites. Quoted as published; it measures searches on search engines, so questions that moved to assistants simply disappear from it.