Turn a newsletter into posts for three networks
Takes one long email and produces a LinkedIn post, an X thread and a short video script, each written for how that platform actually reads, without three copies of the same paragraph.
Tested on GPT-5 ·
-
Pull out the claims, not the prose
Extraction first. If you skip this the model paraphrases your sentences instead of using your ideas.
Step 1 Read this newsletter and list every distinct claim it makes, as bullet points. A claim is something that could be argued with. Skip transitions, greetings and calls to action. {{newsletter}} -
Pick the one claim worth leading with
Each platform wants a different one, so ask which claims are strongest rather than deciding for it.
Step 2 Here are the claims from an article. Rank them by how likely each is to make a professional reader stop scrolling, and say in one line why the top one works. {{claims}} -
Write for LinkedIn
Explicit constraints beat the word professional, which produces the exact voice everyone is tired of.
Step 3 Write a LinkedIn post built on this claim. Rules: open with the claim, not with context. Under 150 words. No hashtags. No question at the end. One concrete number or example. Plain sentences. Claim: {{claim}} Supporting detail: {{claims}} -
Write the X thread
Ask for the thread as numbered posts with the character count, otherwise you get a blog post split with line breaks.
Step 4 Turn this into an X thread of 5 to 7 posts. Each post must stand alone if quoted. First post states the claim, last post has no call to action. Give me each post numbered, with its character count, under 280 each. Claim: {{claim}} Material: {{claims}} -
Write the video script
Say the length in seconds spoken aloud, not in words. The model converts it more accurately than you expect.
Step 5 Write a 45-second spoken script for a short video on this claim. Conversational, first person, no on-screen text directions, no sign-off. The first sentence has to work with the sound off in captions. Claim: {{claim}} Material: {{claims}}
The mistake this avoids is asking for "social posts" in one go. You get three lightly reworded versions of the same text, all of them sounding like a newsletter. Splitting the extraction from the writing fixes it.
Notes from the author
The claims list is worth keeping - it is reusable for months and takes the model ten seconds. On smaller models split step one per section of the newsletter, otherwise it starts summarising rather than extracting.
Tagged