How Jackson asks
3,668 questions across 10,099 host turns — 26% of his turns ask one, and the median question turn runs 33 words: long setups, then the question. Question forms that recur across the show:
Host signature phrases
Phrases Jackson uses at least 8× more often than his guests, per word spoken.
“What’s your relationship with…”
The signature Dialectic question — asked 22 times across 14 episodes. The objects, verbatim; each opens the exact section where it was asked:
Homework
Jackson quotes a guest's own words back at them (“you wrote…”, “you've said…”) in 8% of his question turns overall. Where he brought the most homework:
Name-drops
People, places, and things mentioned in the transcripts themselves — unlike the References tab, which only sees what the show notes linked. Click a name to search the transcripts for it; an episode chip opens that episode.
Showing 50 of 80 recurring names.
Notes on methodology
Distinctive phrases (here and on each episode page) are 1–3 word n-grams whose rate in one episode is far above the rest of the corpus, scored with weighted log-odds and an informative Dirichlet prior so genuinely characteristic phrases beat merely frequent ones. Grams made of stopwords, transcription artifacts, and the episode's own speaker names are excluded; a phrase inside a stronger longer phrase (anna → anna karenina) is dropped.
Host analysis counts a question as any host sentence ending in "?". Recurring question forms are the first 4–6 words of a question, kept when they open questions in 8+ episodes. "Homework" counts question turns matching quote-back patterns ("you wrote", "you've said", "your essay" …) — a conservative regex, so the true rate is higher. Relationship-question objects are extracted verbatim from "relationship with/to …", lightly trimmed, and each chip links to the transcript section where that ask happened. Signature phrases are 2–3 word grams Jackson uses at least 8× more often than guests per word spoken, with a minimum of 25 uses.
The name-drop list uses a capitalized-span heuristic over the original-case transcript: runs of two or more capitalized words anywhere, single capitalized words only mid-sentence, requiring recurrence across episodes and (for single words) that the word rarely appears lowercase. Each episode's own speakers are excluded, and a small hand-curated alias list merges obvious duplicates (New York City → New York, LA → Los Angeles, Elon → Elon Musk); "The New Yorker" is counted only when "the" precedes it, so the bare demonym doesn't inflate the magazine. It is a heuristic, not a named-entity model — expect a few oddities.