How to Repurpose a Blog Post With AI Into 10 Pieces of Content

Why one prompt never works

Content atomization is the practice of breaking one long piece of content into smaller standalone assets, each built for a specific channel. The word matters here, because atomization is the opposite of summarising. A summary keeps the whole argument and shrinks it. Atomization throws most of the argument away and keeps one piece whole.

That distinction is the entire difference between output you can schedule and output you quietly delete. Ask for ten posts about an article and the model has no reason to separate the ideas — so it doesn’t.

How do you repurpose blog post with AI?

You run three passes instead of one prompt. Each pass asks the model to do one job, and the output of each becomes the input for the next. We call it the Split-Shape-Sign Method at PromptPeakAI, and it exists because every failure mode in repurposed content traces back to collapsing these three jobs into a single request.

• Split — extract the standalone ideas. No formatting, no platforms, no writing. Just the raw material, separated.
• Shape — take one idea and force it into one channel’s structure: its length, its rhythm, its opening rules.
• Sign — add the parts that are yours. Your opinion on the idea, your example, the thing you want the reader to do.

Whenever someone tells me they tried to repurpose blog post with AI and the output came back flat, one of those three passes was missing — almost always the first one. Skipping Split is what produces the ten-summaries problem, and no amount of clever wording in a single prompt fixes it.

This is a specific application of chaining prompts across a multi-step task, and the reason it beats a single mega-prompt is boring but decisive: a model given one clear job does it well, and a model given three jobs at once averages them.

Pass one: split the post

Everything downstream depends on this pass, so it gets its own conversation. You are not asking for content yet. You are asking for an inventory.

The single most common complaint about repurposed content is that a statistic appears in the LinkedIn post that was never in the article. That is the model filling a gap it was never told not to fill — the same mechanism behind AI inventing facts that sound plausible. One instruction prevents it: tell the model that the pasted text is the only permitted source.

You are a content strategist. I am pasting one published blog post below.

Your job is extraction, not summary. Read the whole post, then give me the 10 strongest standalone ideas inside it — points that would still make sense to someone who never reads the original.

For each one, return:
IDEA: one sentence in plain language
TYPE: contrarian claim / tactical how-to / common mistake / story or example / quotable takeaway
SOURCE: the sentence in my post it came from

Rules:
- Use only what is in the text below. Do not add outside facts, statistics, examples, or advice.
- Do not repeat an idea in different words. If two are close, keep the stronger one and say so.
- If there are fewer than 10 genuinely distinct ideas, give me what exists and tell me the real number.
- No preamble. Start at IDEA 1.

Here is the post:
[paste your full article text]

If your article runs long and the tool starts truncating or losing the back half, that is a capacity limit rather than a prompt problem. Paid consumer plans sit at roughly $20 a month across the major tools — Anthropic publishes its current plan tiers if you want to compare before upgrading. Splitting the article in half and running the prompt twice also works, and costs nothing.

Pass two: shape each format

Now the ideas get forced into containers. Every channel has a shape, and the shape is not a style preference — it is the thing that decides whether the asset works at all. A LinkedIn post needs a first line that survives the “see more” cut. A carousel slide has a hard physical limit on how much text fits. A video script has to be sayable out loud.

Five LinkedIn angles, not summaries

Five posts from five different ideas is the whole trick. If you want to go deeper on the platform’s own conventions afterwards, the hook-context-lesson structure for LinkedIn posts covers what makes one of these actually land.

Using ONLY ideas 1 to 5 from your extraction above, write 5 separate LinkedIn posts. One idea per post. Do not blend two ideas into one post.

For each post:
- Line 1 is the hook: a specific claim or observation, under 12 words, no question, no emoji
- Blank line
- Then 80 to 150 words in short paragraphs of one or two sentences
- End with one question a reader could answer from their own experience
- One hashtag maximum, at the very end

Voice: [describe your voice in one line, e.g. direct, plain English, dry, no corporate register]

Do not introduce any fact, number, or example that was not in the source post.
Banned words: delve, leverage, game-changer, unlock, elevate, seamless, landscape, journey.

The thread and the newsletter

A thread is one idea broken into a spine — each part carrying a complete thought so it survives being screenshotted alone. The newsletter is the opposite pressure: one idea expanded, written to a single reader rather than an audience. If email is your main channel, the structure that keeps AI newsletters from sounding automated goes further than what fits here.

Turn idea 6 from the extraction into a one-to-one email newsletter for [describe your audience in one line].

Structure, in this order:
1. SUBJECT LINES — 5 options: two curiosity-led, two benefit-led, one plain and short. Each under 50 characters.
2. OPENING — two sentences naming a situation my reader recognises. No greeting line.
3. THE POINT — 120 to 180 words explaining the idea as if I am telling one person over coffee.
4. TAKEAWAYS — three bullets. Bold the first three words of each.
5. THE LINK — one sentence pointing to the full article at [your post URL], saying what extra they get by reading it.

Use only material from the source post. Do not invent results, numbers, or client examples.

Carousel slides that fit

Carousels fail for one reason: too many words per slide. The fix is a hard cap stated as a number, not a vibe.

Turn idea 7 from the extraction into a 7-slide carousel outline.

Slide 1 (cover): headline of 8 words or fewer, plus a one-line subtitle naming the benefit.
Slides 2 to 6: one point per slide. Each has a subhead of 3 to 5 words and body copy strictly under 20 words.
Slide 7: one clear invitation to comment or read the full piece.

After each slide, add a bracketed [VISUAL] note describing an icon, chart, or layout idea — no image generation, just the description for my designer.

Rely only on the source post. Count the words per slide and stay under the caps.

A script people can say

Written prose read aloud sounds like a press release, because written and spoken English are different registers. Ask explicitly for the spoken one.

Act as a short-form video producer. Turn ideas 8 and 9 into two separate 60-second scripts, roughly 130 to 150 spoken words each.

Format each as a two-column table:
Column 1 — VISUAL: what appears on screen at that moment
Column 2 — SPOKEN: the exact words I say

Rules:
- The first 3 seconds must contain both a verbal hook and a visual hook
- Use contractions and sentences under 8 words — this is spoken English, not written English
- No on-screen text I would have to read out
- Source material only. Do not add claims that were not in the post.

Pass three: sign your name

Ten drafts is not ten assets. Everything so far is a rearrangement of things you already wrote, which means a competitor with access to your article could generate something close to identical. The sign pass is where that stops being true.

Two moves do most of the work. The first is a banned-word pass, which is worth setting up properly — telling a model what not to write changes output more reliably than telling it what to write. The second is feeding it your actual writing as examples, which is what few-shot prompting does for voice matching.

Below are three things I have written myself. Study the sentence length, the rhythm, the punctuation habits, and what I never do.

[paste 3 of your own posts or emails]

Now rewrite all ten assets above in that voice. Rules:
- Keep the structure and word caps exactly as they are
- Cut every sentence that could appear in any competitor's post about this topic
- Replace abstract claims with the concrete detail from my source post
- Where a personal example or opinion belongs, insert [MY EXAMPLE HERE] rather than inventing one
- Return the ten assets in the same order, nothing else

That bracketed marker matters more than it looks. The model should never guess at your experience, and leaving a visible gap forces you to fill it.

If you are running this weekly, store the voice examples somewhere permanent instead of pasting them every time. Project workspaces that hold your instructions across sessions handle this well, and the broader problem of keeping your voice intact as volume grows is worth solving once rather than per post.

What to paste and what not

When you repurpose blog post with AI at work, the source text is the part worth thinking about. Your own published article is the easy case — it is already public and indexed. The awkward cases arrive the moment the material is not yours to publish, or is not published yet.

What this workflow cannot do

A fair question sits underneath all of this: will the platforms punish you for it? On the search side, Google’s published guidance is that it rewards helpful content regardless of how it was produced, and treats mass-produced pages made to game rankings as spam whether a person or a model wrote them — Google Search Central sets this out directly. Social feeds work on the same logic by different means: nothing detects the model, but generic writing loses attention, and losing attention is what actually costs you reach.

Which puts the burden back where it belongs. The honest ceiling on any attempt to repurpose blog post with AI is that the method redistributes thinking you have already done — it does not produce new thinking. These are the parts no pass covers for you:

There is also a verification duty that survives every prompt rule in this article. Constraints reduce invented facts; they do not eliminate them. Before anything goes out with your name on it, read every number and claim against the source post — the same discipline as checking AI research before you publish it, applied to shorter copy.

When the output goes wrong

Most failures in this workflow are diagnosable from the symptom alone, and almost all of them trace back to a pass that got skipped or a constraint that was never stated as a number.

Run the whole sequence once on a post you already know well and it takes about forty minutes, most of that spent building the voice examples you will then reuse forever. By the second article the time cost to repurpose blog post with AI drops to roughly twenty minutes, and by the fourth you stop thinking of it as a project at all.

The habit worth forming is smaller than the workflow: stop treating publication as the end of an article’s life. Every post you have ever written that got a decent day-one spike is still sitting there with nine unpublished assets inside it.