Claude says: Every study in this series has run on samples, because public search indexing exposes only part of a WordPress archive and the rest has to be inferred. That constraint is gone. A full export of lukeford.net, generated August 19, 2026, contains 27,795 posts, of which 27,737 are published. What follows is a census.
Two limits before the results. The WordPress archive begins in 2006, so the first nine years of Ford’s (b. 1966) online career, the LukeFord.com era and its immediate successors, sit in static HTML outside this file and cannot be measured here. And word counts throughout exclude everything inside blockquotes, so quoted material, which Ford uses heavily, contributes nothing to any figure below. These are author-written words.
The headline result concerns 2026. Through August 19, in 231 days, Ford published 3,375 posts carrying 11,421,248 author-written words. The twenty years from 2006 through 2025 carry 12,609,922. Seven and a half months of 2026 amount to ninety-one percent of the preceding two decades combined. At the current rate the year closes near eighteen million words, which would be roughly one and a half times everything published in the twenty years before it.
The obvious hostile reading is that a handful of enormous posts carry the total. They do not. The ten longest posts of 2026 run between forty and seventy-nine thousand words, and the top one percent of posts account for eleven percent of the year’s words. The top quarter accounts for seventy percent. The distribution is broad, and the median tells the story better than the mean: the middle post of 2026 runs 1,724 author-written words, with the first quartile at 703 and the third at 3,482. The typical post has become an essay.
That median is where the finding lives, because it disposes of the second hostile reading, which is that Ford has always had prolific years and this is another one. He has. 2016 remains his highest year by post count, with 4,073 published posts, more than 2026 has produced so far. Its median post ran 362 author words. 2008 produced 2,926 posts at a median of 150. The median by year runs 146 in 2024 and 275 in 2025 before reaching 1,724 in 2026, a twelvefold increase in the typical post in two years.
So the earlier peaks and the current one are different phenomena. The old high-volume years were many short items. The current regime is many long ones, and the count and the length rose together only in 2026.
The timing separates into two events, which is the least expected result in the series.
Posts carrying an explicit machine marker, the “ChatGPT says” and “Grok says” and “AI helped me with this post” openings, run between one and four percent of output from 2022 through November 2024. In December 2024 the share jumps to twenty-two percent. In March 2025 it reaches seventy-three percent, which matches the date Ford stated publicly that he had for the first time run an essay through AI for editorial criticism and intended to continue. The archive confirms his account and adds three months of quieter use before it.
Length does not move with adoption. Median post length stays between roughly 100 and 500 author words through August 2025, six months after the markers spike. The climb begins in September 2025 at 1,036, reaches 1,923 in October and 2,084 in November, and never returns. April 2026 averages 7,187 author words per post.
The gap matters for anything anyone wants to claim about causation. Ford had the tool for half a year before his publication threshold changed, so the arrival of the machine does not by itself explain the collapse. Something in the autumn of 2025 turned an editing aid into a production regime, and identifying it is a better question than the one the earlier sampled studies were asking.
One instrument caveat belongs here rather than in a footnote. The machine-marker share falls to twenty-nine percent across 2026 and into single digits in several months, which does not indicate reduced use. It measures per-post disclosure, and Ford’s disclosure is centralized rather than repeated. Both of his curated index pages, Highlights and Bios, carry a standing line stating that most of the work listed has been AI-assisted since about May 2025, linked to his full account of the practice. So the 2026 decline in per-post markers records a change in where the disclosure lives, not a return to unassisted writing.
The dating is worth setting out, because four independent sources give four dates and they form a sequence rather than a contradiction. Per-post machine markers jump in December 2024. Ford states his adoption of AI as an editor of his own prose on March 9, 2025. His own retrospective dates the transformation of his blogging to about May 2025. And the median post length does not move until September 2025. The tool, the stated adoption, the remembered turning point and the behavioral change stand roughly three months apart from each other across a year.
The census also settles a question the essays could previously only raise. Across twenty years, 58 posts sit outside published status: 48 marked private, 7 trashed, 3 drafts. That is 0.21 percent of everything in the archive, scattered thinly, clustering loosely around 2008 and 2015 through 2017, never exceeding eleven in a year.
The earlier work in this series argued that Ford’s weakest brake is the one leaving no artifact, on the reasoning that not publishing produces no traffic, no essay and nothing that can later join the archive. That was an inference. It is now a measurement, and it is stronger than the inference. On the evidence of the export, withdrawal is not a weak brake in Ford’s practice. It is close to an unused one. Two in a thousand posts have ever been taken down or held back.
What all of this establishes is narrow and hard. The threshold between having a thought and shipping a finished public object has collapsed twice in one career. The first collapse came when direct publication removed the editor, the printing press and the wait until morning, and it is visible in the post counts of 2008 and 2015 and 2016. The second came in the autumn of 2025, and it is invisible in the post count and unmistakable in the median. What became cheap the second time was not the act of publishing. It was the labor of producing something that looks finished.
What the census does not establish deserves equal weight. It says nothing about quality, accuracy, or whether any of the eleven million words are worth reading, and it cannot separate the machine from the changing subject matter, the changing reading list or the changing intent of a man who decided in 2026 to write long analytical essays. Word counts are not arguments. A production series can show that the cost of making an intellectual object fell. It cannot show that the objects are good, and the earlier finding in this series, that recent essays are often rich in evidence surrounding an explanation and poor in evidence identifying it, is the one that speaks to that question.
The monthly series, 237 rows running from 2006 to August 2026 with post counts, author-word totals, median lengths and machine-marker counts, is published alongside this piece so that anyone can check the arithmetic or dispute the exclusions.
