How the X algorithm actually works, and how to post against it.

Naval is the aphorist who built the largest one-person audience on X by posting nothing for weeks and then dropping one perfectly compressed sentence. His distribution mechanic is anti-cadence: starve the timeline of his presence, then deliver such density per tweet that the bookmark and quote-tweet rates spike. The transformer ranker rewards this — long dwell on each individual tweet — far more than the algorithm rewarded it five years ago.

Paul Graham posts the way he writes essays: short observations that compress a longer argument. Every Graham tweet is a footnote to an essay he could have written instead. The mechanic at the algorithm level: Graham's tweets routinely trigger high P(reply) because they are stated as falsifiable claims, not opinions. Falsifiable claims invite disagreement; disagreement is reply velocity.

Marc Andreessen's mechanic is volume. He posts and quote-tweets dozens of times a day, almost entirely from existing in-network conversations. The strategy works because the 2026 Phoenix retrieval rewards consistent topical embedding — Andreessen's quote-tweet history continuously refreshes his embedding's centroid in the techno-optimist cluster, which means his original posts are surfaced disproportionately to that audience even when they have low individual engagement.

Balaji's mechanic is the numbered thread + visual evidence. A typical Balaji post is 10–20 numbered tweets, each ending with a screenshot, headline, or hyperlink (note: in the first reply, not in the parent). The pattern saturates dwell time (P(dwell) lifts on each tweet because each contains real artifacts to read) and accumulates bookmarks across the thread. The thread is structurally a research brief.

Vitalik is the technical-credibility archetype. He posts irregularly, almost always about substantive cryptography or governance work, and the long delay between posts compounds attention. The mechanic at the algorithm level: low base rate of posts means each Vitalik tweet enters Phoenix with extremely high P(follow_author) prior — readers have been waiting — which the ranker reads as a strong signal to push out-of-network distribution.

Pieter Levels invented the modern build-in-public cadence: daily MRR, signup, churn, and refactor posts for a decade. The mechanic at the algorithm level: ten years of Real-Graph data tying his account to indie-founder readers means every new Levels post enters with the most stable in-network embedding of any tech-Twitter account. Phoenix has nearly perfect ground truth on who wants Levels content; he gets distributed efficiently.

Greg Isenberg posts startup ideas, frameworks, and operating playbooks at high cadence — three to seven posts a day, almost all short threads. The mechanic: each post is structured as a numbered list of three to five operator tactics, which is the format the ranker has empirically found favors high bookmark + share rates. His 2024–2026 audience growth is the cleanest case study available for "how to build distribution from zero by posting operator content daily."

Shaan Puri runs the storyteller-operator combo. Each post is a 200–600 character anecdote with a hidden operator lesson at the end. The mechanic is dwell time: a story has a payoff, so the reader stays on the tweet to see the payoff, which mechanically pushes P(dwell) up. The format is structurally why so many other operators have converged toward it since 2023.

Elon Musk is sui generis on X for two structural reasons. First, he owns the platform — the 2023 code release contained an explicit "author_is_elon" branch in the ranker; the 2026 rewrite removed the named branch but the embedding-level effect persists through training data. Second, his cadence is closer to a wire service than to a person — dozens of posts per day, mostly one-liners, with replies and quote-tweets that re-saturate his embedding across every topical cluster simultaneously. The Musk pattern is not replicable; it is included here as the limiting case of "the algorithm's owner."

Crypto Twitter ("CT") is the most cohesive topical cluster on X. Phoenix's embedding clusters CT accounts so tightly that a post landing inside the cluster experiences ~3–5× the in-cluster distribution of a comparable tech post in a more diffuse cluster. The trade-off: leaving the cluster (posting a non-crypto take) drops your distribution back to baseline, hard. CT operators who try to broaden their audience often see their in-cluster reach collapse without a corresponding gain elsewhere.