Verified against ChatGPT · 2026-08-09
Turn one real insight into an X thread that doesn't lose readers after tweet two
Builds a numbered X thread from a single specific insight you already have, structured so each tweet earns the next tap instead of front-loading everything into the hook and coasting on filler after it.
The prompt
Ready to copy — highlighted parts are example details you can swap.
You are structuring an X (Twitter) thread from one specific insight or result I already have — not inventing a topic, just turning something true and specific into a thread that keeps people reading past the first two tweets, which is where most threads lose their audience.
THE INSIGHT OR RESULT
We cut our onboarding email sequence from 9 emails to 3 and activation rate went up, not down.
WHO ALREADY BELIEVES THE OPPOSITE
Most growth advice says more touchpoints in onboarding always improves activation.
PROOF I ACTUALLY HAVE
Activation went from 34% to 41% over 6 weeks after the cut, same cohort size, no other changes shipped that period.
THREAD LENGTH TARGET
6-8 tweets
CLOSING ASK
Ask people to reply with their current onboarding email count so I can compare.
STRUCTURE RULES
Write the first tweet as a claim that is specific enough to be wrong — a vague hook like "here's what nobody tells you about X" gets scrolled past because it promises nothing checkable, while a hook that states the actual contradicted belief and the surprising result creates an open loop a reader wants closed. Every tweet from 2 onward must do exactly one job: advance the argument, add a specific piece of evidence, or address the most obvious objection a skeptical reader would have at that exact point — never restate the hook in different words to pad the count toward a round number. If I don't have enough real evidence to support a claim in the thread, say so explicitly and either cut that beat or mark it as something I need to verify before posting, rather than inventing a statistic or a specific number to make a tweet sound more authoritative than what I actually gave you. Place the strongest piece of proof by tweet 3 or 4, not saved for the end — threads lose the bulk of their remaining readers in the first few taps, so the payoff has to arrive before that drop-off, with the ending reserved for the takeaway and the ask rather than the best evidence. Write the closing tweet as a specific ask tied to the argument just made, not a generic "thoughts?" or "follow for more."
WHAT NOT TO DO
Do not use thread-starter clichés ("A thread 🧵", "Let that sink in", "Save this for later"). Do not write filler tweets whose only content is a transition phrase. Do not exceed the requested tweet count by padding — if the argument is genuinely finished in fewer tweets than the target, say so and give me the shorter version instead.
OUTPUT FORMAT
Numbered tweets (1/, 2/, 3/...), each under 280 characters, with a one-line note after the draft flagging any claim you couldn't verify from what I gave you and where you'd want a real number or link before I post it.Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
X's own engagement curve for threads drops off sharply after the first couple of tweets — most of the audience that will ever see tweet 4 already decided whether to keep reading by tweet 2, which is why the prompt forces the strongest proof point into tweets 3-4 instead of holding it for a big finish nobody reaches. Requiring the hook to state the specific contradicted belief rather than a vague teaser matters mechanically because GPT-5.1, left unconstrained, defaults to curiosity-gap openers ("here's what nobody tells you") that read as generic pattern-matches on viral-thread structure rather than a claim with actual content — a reader can't evaluate whether to care about an empty tease, but a specific contradicted-belief-plus-result hook gives them something to agree or disagree with immediately, which is what drives the tap to expand. The explicit instruction to flag unverifiable claims rather than invent supporting numbers exists because the model's fluent default is to generate a plausible-sounding statistic to make a weak beat feel stronger, and a fabricated number in a thread that gets any traction is the single fastest way to get quote-tweeted with a correction. Capping each tweet to one job (advance, evidence, or objection-handling) prevents the common failure mode of a thread that restates its own hook three different ways to hit a target length — which reads as padding to anyone scrolling and kills the reshare rate long before it kills the read-through rate.
What you get back
1/ We cut our onboarding sequence from 9 emails to 3. Activation went UP, not down. 2/ The assumption we'd never questioned: more touchpoints = more activated users. Turns out past a point, they just teach people to skim. 3/ Same cohort size, 6 weeks, no other changes shipped: 34% -> 41% activation after the cut. [Note: verified against the 6-week cohort numbers given; no claim beyond that was invented.]
Verified against
ChatGPT GPT-5.1 · 2026-08-09
Changelog
- 2026-08-09 — Initial publish, verified against ChatGPT GPT-5.1.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
