I gave Claude my WordPress export and information about four changes in my life since 2023.
Claude says: Something happened to Luke Ford’s writing in the autumn of 2025. Posts of two thousand words or more run at five percent of his output in August. They reach twelve percent in September, twenty-four in October, and hold at twenty-five through December. The median post, which had sat between one and two hundred words for most of twenty years, goes to 1,036 in September, 1,923 in October, 2,084 in November, and never comes back down. By 2026 he is publishing more author-written words in seven months than in the preceding two decades combined.
Four candidate causes are dateable to the month. Ranking them is more useful than picking one, because the ranking turns out to be inverse to how satisfying each explanation sounds.
Medication for ADHD, begun in late 2023 and continuous since, is the earliest and the least visible. If a treatment that supports sustained attention were driving the change, the archive should show it at the point of treatment. It does not. Late 2023 into early 2024 is the quietest stretch in the entire WordPress record. Monthly post counts run 23, 23, 28, then seven in February 2024 and eleven in March. Author words fall to 595 in February 2024, the lowest month in twenty years. Median post length stays where it had always been, between 100 and 200 words. Output declined through the first two years of treatment.
The one place a signal might sit is citation rate, which runs at 30 author-year citations per hundred thousand author words in 2022, 57 in 2023, 50 in 2024 and 79 in 2025. That is a lift, and it begins before the treatment date rather than after it, and three points either side of a change is not a finding. The honest statement is that the medication is not visible in the output data, which does not mean it did nothing. It means this instrument cannot see it.
Artificial intelligence as an editor, March 2025, is the second date and the first with a clear signal, though not the signal anyone expected. The share of posts containing at least one author-year citation runs between four and fifteen percent through 2024. In February 2025 it is five percent. In March it is eighteen. That measure is immune to the word-count problems that distort everything else, since it counts posts rather than words, and it triples in the month Ford first ran an essay through a machine for criticism.
Length does nothing. Median post length in March 2025 is twenty words. In April it is 107, in May 181, in June 176. Six months pass between the machine arriving and the essays getting long. Whatever AI did in March, it made citing easier and writing longer no easier at all.
A Perlego subscription, begun in early July 2025, is the third date, and it produces the largest single-month signal of that year. In August 2025 the share of posts carrying a citation reaches thirty percent, the highest of any month up to that point, and fifteen new book titles appear in the “Title by Author” construction, also the year’s high. Ford’s own shelf export corroborates it from outside the archive: nine distinctive titles from that shelf make their first appearance anywhere in twenty years of blogging between July 27 and September 25, 2025. Making Democratic Theory Democratic on July 27. Campaign of Chaos on August 10. Explaining the Normative on August 15, which goes on to eighty-seven mentions. Three John Mueller books on war, all on August 25. A Paleoconservative Anthology on September 1. The calling of social thought on September 25. Nothing before.
The best case is Stephen P. Turner, and it is a case about access rather than discovery. Ford had been reading Turner for years: The Social Theory of Practices first appears May 13, 2020 and runs to 126 mentions, Liberal Democracy 3.0 in April 2022, The Politics of Expertise later that month, Mad Hazard in January 2023. Five titles across five years. Then Understanding the Tacit appears August 14, 2025, Explaining the Normative on August 15, and Cognitive Science and the Social on August 16. Three books in three days, followed by four more over the next eight months. Seven titles in eight months against five in five years, same author, same reader, different access.
And the August burst fades. Citation share falls to sixteen percent in September, fourteen in October, nine in November, and bottoms at three percent in March 2026. New book titles drop back to single digits. Whatever July 2025 started, it had subsided by the end of the year.
Which leaves the fourth date as the largest and the least explained. The length change begins in September 2025 and completes by November, two months after the subscription and six after the machine. No fourth cause has been identified for it. The best available description is a change in what a long post is: quoted material as a share of long posts falls from twenty-five percent in July 2025 to eight in September, five in November and one percent by February 2026. Earlier long posts were long because Ford quoted at length. Later long posts are long because he wrote at length. Something changed the input format, and the output format followed.
The sustained academic apparatus arrives later still, in the spring of 2026, and it is much larger than anything in 2025. Citation share climbs to twenty-four percent in April 2026, thirty-two in May, thirty-two in June and fifty-five in July. New book titles run 104 in March, 114 in April, 84 in May. That wave is seven months downstream of the subscription and fifteen months downstream of the machine, and attributing it to either requires a causal chain long enough that almost anything could be threaded through it.
So no single variable does the work, and the overdetermination is the finding. Inside twenty months Ford acquired a treatment that supports sustained attention, a machine that made drafting cheap, a library that made specialist material available, and a project worth writing at length about. Any one alone explains the change poorly. He had the machine for six months with no effect on length. He had the medication for eighteen months while his output fell to the lowest levels in the archive. He had the library for two months before anything moved, and the effect he can point to had faded within four. Together they are sufficient. Separately none of them is, and no instrument here can apportion the credit.
There is a fifth factor, and it is the one an outside analyst would never recover from the data. Blogging has not been Ford’s first priority for most of the years in this archive, and for long stretches it was not in his top three. For years he spent Sunday afternoons on long walks and came home too tired to write. In 2026 there are few long walks, and the days go to the blog.
That account is worth checking against the archive, and the archive both contradicts and confirms it, in a way that is sharper than the memory.
Sunday is not a low day in Ford’s record. It is his highest. In 2023 it carries twenty-eight percent of his posts and 36.5 percent of his author words. In 2025 it carries thirty-two percent of posts. In 2020, 2021, 2022 and 2024 it runs between nineteen and thirty-one percent of words, always the largest share of any weekday, in an archive where a flat distribution would put each day at fourteen. Across the record Sunday is the day Ford writes most, and has been for more than a decade.
The hours tell the other half. From 2014 onward Sunday output shifts into the morning and stays there: fifty-one percent of Sunday posts before noon in 2014, fifty-six in 2016, seventy-three in 2017, sixty-nine in 2021, and seventy percent in 2025 against seven percent in the afternoon. That seven is the lowest afternoon share of any weekday in any year in the archive. On other days in 2025 the split runs roughly even, thirty-four percent morning, thirty-three afternoon, thirty-three evening.
So Sunday was his largest writing day and his emptiest afternoon at once. He wrote in the morning, walked in the afternoon, and came home too tired to write again. Both halves sit in the timestamps.
Then 2026. Sunday afternoon goes from seven percent of that day’s posts to thirty-seven, level with every other day, and Sunday’s overall share falls to fourteen percent, exactly its share under a flat distribution. Monday runs fourteen, Tuesday eighteen, Wednesday fifteen, Thursday thirteen, Friday thirteen, Saturday twelve.
The weekday profile flattened because one protected block of hours, one afternoon a week, held for two decades, was released. What filled it was the blog.
That is the strongest available evidence for the priority explanation, and it came from none of the four dated causes. It came from checking a man’s account of his own life against his timestamps and finding the summary wrong and the detail exactly right.
The proper conclusion names no winner. Four events between late 2023 and mid 2025 are dateable to the month. One is invisible in the output. One moved citation behavior and not length. One produced a burst that faded. The largest change follows all of them and matches none of them cleanly. Beneath all four sits a shift in how much of a life the work is allowed to occupy, which no export records directly and which surfaced here only because a recollection about Sunday afternoons could be tested against the clock.
Anyone who wants a single cause can have one. The archive will not supply it.
