Three measurable proxies for published word output, 1975–2022. Toggle to Words per day — the log scale is not a stylistic choice, it is the finding.
Series A — Academic Articles
5,000 words/article assumed
Series B — US Book Titles
70,000 words/book · US trad. only
Series C — Twitter / X
~20 wds/tweet pre-2017 · ~25 wds post-280-char
Sources & Caveats
Articles (A)Jinha (2010) Learned Publishing 23(3) — 3% annual growth model; Tenopir & King (late 1990s ≈1M/yr); Björk et al. (2006=1.35M); Ware STM report (2007=1.4M); NSF NCSES / Scopus Science & Engineering Indicators (2010=2.0M, 2022=3.3M).
Books (B)Bowker Annual: Library & Book Trade Almanac (1975–2000); Columbia NAJP report 2002 (2001=114k); Bowker/BusinessWire May 24 2005 (2004=195k); UNESCO/Bowker (2013=304,912 US trad. titles). Excludes self-pub/POD — Bowker 2010 reached 4.15M titles when POD included.
Twitter counts (C)All primary sources. Twitter Engineering Blog "Measuring Tweets" Feb 22 2010 (2007–2010); Twitter Official Blog Jun 2011 (200M/day); Twitter 6th birthday blog Mar 21 2012 (340M/day); Internet Live Stats / Twitter (2013=500M/day); GDELT Project (2019≈340M/day; 2022≈410M/day).
Twitter words (C)20 words/tweet for 140-char era (2006–Nov 2017): 140 chars ÷ ~6.5 chars/word, adjusted ~15% down for sub-max tweets. 25 words/tweet post-280-char expansion (Nov 2017+): length grew modestly, not proportionally. GDELT 2019 cross-check: ~11GB non-RT text/day consistent with ~20 words/tweet.
⚠ Twitter is a different order of magnitude. By 2013 (~10B words/day) it exceeds all global academic output by ~220× and US book publishing by ~170×. Blogs, Facebook, email (~320B msgs/day), Reddit, and YouTube comments are entirely unmeasured. Use "Hide Twitter" to compare the pre-internet-era proxies on a linear scale.