The X Playbook
Algorithm-aware · bilingual archive
§ Patterns

Eight structures the ranker rewards.

Each pattern is paired with a worked example, the bad-version-for-contrast, and a one-paragraph explanation of which Heavy Ranker probabilities it pushes.

hook
#scroll-breaker-hook

The scroll-breaker first line

Three to nine words that make a reader's thumb stop, before they have decided why.

Every other lever in this playbook depends on the reader not scrolling past. The first line of a tweet has roughly 0.4 seconds to win. The best openings are short, low-context, and committed to a position. Long openings hedge; short openings provoke. A reader's brain treats a hedged opening as filler and a committed opening as data.

Example
Bad   →  In today's fast-paced world, we have been thinking a lot about why so many product launches fail and what we can learn from them …
Good  →  Most product launches die in the first hour. Here's why:
Why it works

The Heavy Ranker has a dwell-time-over-2s prediction that depends almost entirely on first-line legibility. A short committed opening also improves predicted reply rate, because committed claims invite agreement or disagreement; hedged claims invite nothing.

structure
#one-claim-per-tweet

One claim per tweet

Density beats depth. Each tweet should make one point and stop.

X is not a blog. A tweet with three half-arguments performs worse than a tweet with one whole argument, even at identical character count. The Heavy Ranker scores per-tweet, not per-thread; readers reply to specific claims, not paragraphs. Compress until one idea owns the entire space, then move the rest to subsequent tweets or replies.

Example
Bad   →  AI is changing how startups raise capital, and the new wave of solo founders is using AI tools to scale faster than ever, plus VCs are adjusting their portfolio construction to match.
Good  →  The first solo founder to hit $10M ARR with no employees lands this year. Bookmark this tweet.
Why it works

Specific concrete claims drive bookmarks (high weight, post-2023 signal) and replies (the heaviest weight). Vague aggregations drive neither.

structure
#specificity-bias

Numbers, names, and proper nouns

Concrete particulars travel; abstractions stay home.

Compare "some banks failed last year" to "Silicon Valley Bank, First Republic, and Signature failed in 90 days in 2023." Same fact, very different reach. Specific numbers and proper nouns make a tweet quotable: the reader feels they are getting information rather than commentary. The ranker reads quotability as predicted-bookmark and predicted-reply lift.

Example
Bad   →  Crypto is back.
Good  →  BTC at $103k, ETH at $4.2k, USDT in 38 countries' sovereign reserves. 'Crypto is back' is now an understatement.
Why it works

Specificity is hard to fake; the reader's brain rewards it with attention. Specific posts get bookmarked and quote-tweeted at far higher rates than the same claim stated abstractly.

structure
#reply-bait-self

Self-reply within 15 minutes

The post that lives long enough to win is the one whose author shows up first.

The Heavy Ranker decides whether to keep distributing your post based largely on engagement velocity in the first 15 minutes. The single best move you can make in that window is to reply to your own tweet with a substantive follow-up — a stat, a counter-example, a screenshot, a question. The reply creates a fresh engagement event that triggers the ranker to re-score the parent. Many high-performing posters drop the self-reply within 4–8 minutes, not later.

Example
Main tweet  →  Most product launches die in the first hour.
T+6m self-reply  →  Specifically: only 11% of Product Hunt launches above #5 spot last 48 hours of engagement past the day-of. Here's the dataset:
Why it works

First-15-minute velocity is the single largest variable in whether a tweet escapes its initial follower graph and reaches the out-of-network candidate pool. A timed self-reply is the cheapest way to manufacture velocity legitimately.

thread
#thread-as-cliffhanger

Thread as cliffhanger, not summary

Each tweet should make the next one inevitable to click on.

The mistake amateur threaders make is treating thread tweets as section headings of an essay. The mistake works against the algorithm because each tweet is scored independently. Each tweet in a thread needs its own hook, its own dwell-time payoff, its own quotability. The thread structure that works: each tweet ends one sentence before the next tweet's most interesting word.

Example
Bad   →  T1: Three things every founder should know. ↓
Good  →  T1: A founder told me last week he wasted 18 months on a deck nobody read. The thing he did instead, in tweet 3, almost made me close the laptop. ↓
Why it works

Tweets 2 through N in a thread are themselves served as out-of-network candidates to other users; if any individual tweet in the thread fails the per-tweet scorer, the whole thread's distribution stalls. Optimize every tweet, not just T1.

format
#image-as-thumb-stopper

Images as thumb-stoppers

An attached image is, on average, a 30–60% lift in impression-to-engagement ratio.

The Heavy Ranker does not directly add weight for an attached image, but media-bearing tweets win on the upstream variables it does score: dwell time (the image arrests the scroll), reply rate (images invite quote-replies and screenshot replies), and bookmarks (visual artifacts are bookmark-bait). The image does not need to be related to the tweet — a screenshot of a tasteful quote, a chart, even a black square with one bolded sentence outperforms text-only.

Example
Pure text  →  Most product launches die in the first hour. (~5% engagement rate ceiling)
With chart →  Same text + a screenshot of a Product Hunt engagement-decay curve. (~9–11% engagement rate ceiling)
Why it works

Dwell time over 2 seconds is now a ranker input. Images mechanically increase it. The result is a fully legitimate ranker lift, not a trick.

viral
#build-in-public-cadence

The build-in-public cadence

Daily small posts beat occasional polished ones — but only if the small posts have stakes.

Pieter Levels, Greg Isenberg, and the broader indie-founder cohort built audiences by posting a metric per day — MRR, signup count, refactor ship, churn rate — for years. The cadence works because the Heavy Ranker rewards consistency (in-network engagement compounds for posters whose followers come back) and because each metric tweet is, structurally, a stake-bearing claim that invites reply ("how are you growing so fast?" / "that churn rate is too high"). The cadence fails if the metric is fake. Lifetime detection of fakery is near-immediate; recoverability is near-zero.

Example
Monday:    'Day 412 of building. MRR $42,800. +3.2% w/w. Two churned this week, both because a competitor cut price 40%. Will not match.'
Tuesday:   'New onboarding step shaved 18% off Day-2 drop-off. Screenshot:'
Wednesday: 'Lost a candidate to a YC startup. They offered $260k base. We offered $190k + 1.5%. Founder pay-off math:'
Why it works

Real-Graph (your actual interaction graph) and SimClusters favor consistent posters who attract repeat engagement from the same set of accounts. Three months of daily numbers is more distribution than one viral thread.