n8n + n8n: Content Repurposing Agent: 1 Article → 10 Formats in 60 Seconds — n8n Workflow

Your marketing team publishes one solid article a week. That article should become a LinkedIn post, a Twitter thread, a newsletter, five short-video scripts, a Telegram broadcast, and three ad variations. In practice it becomes exactly one thing — the article — because turning it into ten formats by hand takes a copywriter half a day, and half a day per article does not scale. This workflow closes that gap: paste a URL or raw text, and in about 60 seconds you get all ten assets, on-brand, ready to schedule.

The problem: distribution is where content dies

Writing is the expensive part, but under-distribution is the part that quietly kills ROI. A 1,500-word article represents real research and editing effort, yet most of it reaches a single channel and a single audience segment. The people who would have converted from a punchy LinkedIn hook never see it. The prospects who only open email never get the digest version. The audience living on short-form video gets nothing at all.

The manual fix — a person rewriting each piece per channel — fails for three reasons. It is slow (30–60 minutes per format), it is inconsistent (voice drifts across ten rewrites), and it is the first task dropped when the week gets busy. So the backlog grows, published articles rot in the archive, and your best content earns a fraction of the reach it paid for. This is not a creativity problem. It is a throughput problem, and throughput problems are exactly what n8n solves.

The solution: one input, ten channel-native outputs

The Content Repurposing Agent takes a single source — a live URL or pasted article text — and fans it out into ten channel-specific assets in one pass:

• A LinkedIn post with a scroll-stopping first line and a professional CTA
• A Twitter/X thread broken into hook + numbered tweets under the character limit
• A plain-text email for your newsletter, with subject line and preview text
• Five Reels/Shorts/TikTok scripts, each with a hook, beats, and on-screen text cues
• A Telegram summary formatted for a broadcast channel
• Three ad headlines for paid testing

The key idea is that each output is generated with a channel-native prompt — the model is told the format, the audience mindset, and the constraints of the platform, so a LinkedIn post reads like LinkedIn and a Reels script reads like a shot list, not a repackaged blog paragraph. Everything is produced in a single execution, returned as structured JSON, and pushed wherever you want it: a Google Sheet, a Notion database, Slack, or straight into a scheduling tool.

Step-by-step: building it in n8n

The workflow is deliberately linear so it is easy to debug and cheap to run. Here is the node-by-node build.

1. Trigger and input

Start with a Webhook node (method POST) so you can fire the flow from a form, a browser bookmarklet, or another workflow. Accept two fields: source_url and source_text. For a no-server option, swap in an n8n Form Trigger with a URL field and a long-text field — that gives you a hosted paste box with zero front-end work.

2. Fetch and clean the article

Add an IF node: if source_url is present, route to an HTTP Request node that GETs the page HTML. Follow it with an HTML Extract node — target the article, main, or .post-content selector to pull clean body text and drop nav, footer, and sidebar noise. If source_text was pasted instead, skip the fetch and pass it straight through. Merge both branches with a Set node that normalizes everything into a single field, article_body, and truncates to a sane length (aim for ~6,000–8,000 characters) to keep token cost predictable.

3. The repurposing engine

This is the core. Use one Anthropic Chat Model node (or the Basic LLM Chain node with an Anthropic credential) pointed at claude-sonnet-5 — it is the right balance of quality and cost for high-volume rewriting, and its strong instruction-following keeps all ten formats on-brand in a single call. Configure it to return everything in one shot:

• Set a system prompt that carries your brand voice, tone rules, and target audience — this is what keeps voice consistent across all ten outputs.
• In the user prompt, inject {{ $json.article_body }} and explicitly request all ten assets, each wrapped in a labeled key.
• Attach a Structured Output Parser (or enable JSON mode) with a schema defining linkedin, twitter_thread (array), email (object with subject/preview/body), reels (array of 5), telegram, and ad_headlines (array of 3). Forcing structured output is what makes downstream routing reliable — no regex scraping of prose.

Set temperature around 0.7 for natural variety, and max tokens high enough (4,000+) to fit all ten formats without truncation. If you want tighter cost control, split the heavy formats (Reels scripts) into a second parallel LLM node, but a single call is simpler and usually enough.

4. Split and route the outputs

After the model, drop an Item Lists or Code node to break the JSON into individual items keyed by channel. From there, fan out with a Switch node or parallel branches:

• Google Sheets / Notion node — append every asset to a content-ops table for review and scheduling.
• Slack or Telegram node — post the summary to your team channel for a quick approval step.
• Optional Gmail / SMTP node — drop the email draft into your inbox.

Finish with a Respond to Webhook node returning the full JSON payload, so whatever triggered the flow gets all ten assets back immediately.

Benefits: what this actually changes

The obvious win is time — 60 seconds of compute replaces four to six hours of manual rewriting. But the compounding wins matter more:

• Reach multiplies without new writing. The same research now lands on six channels instead of one, hitting audiences that never overlap.
• Voice stays consistent. One system prompt governs all outputs, so you do not get ten slightly different brand voices from ten tired rewrites.
• Testing gets cheap. Three ad headlines and five video hooks per article means you always have variants to test — the raw material for finding what converts.
• It scales linearly. Ten articles a week costs the same per-unit effort as one. The bottleneck moves back to writing quality, which is where you want it.

For a lean team, this is the difference between "we publish content" and "we run a content engine."

Common pitfalls (and how to avoid them)

Garbage extraction in, garbage assets out. The most common failure is a bad HTML Extract selector that grabs cookie banners or related-post lists instead of the article. Test your CSS selector against three or four target sites before trusting it, and always keep the paste-text fallback for pages that block scraping.

Skipping structured output. If you let the model return free-form prose, your routing nodes will break the first time it phrases a header differently. Enforce a JSON schema with the Structured Output Parser and set the workflow to error loudly on parse failure rather than pushing malformed data downstream.

No human gate before publishing. The agent generates drafts, not final truth — it can hallucinate a statistic or overstate a claim. Route outputs to a review table or Slack approval, not straight to auto-publish, at least until you trust the prompt on your content.

Token truncation. Long articles plus ten formats can exceed your max-tokens ceiling, silently cutting off the last format. Set generous limits, truncate the input to a fixed length, and add an IF check that flags any output missing a required key.

Rate limits under load. If you batch-process an archive, add a Loop Over Items node with a small wait between iterations so you don't hit API rate limits mid-run.

Get these five right and the workflow runs unattended for months. Paste, wait 60 seconds, review, schedule — that is the entire loop.

PRONTO PARA USAR

Ja construimos isso pra voce

Nao comece do zero. O Content Repurposing Agent: 1 Article → 10 Formats in 60 Seconds — n8n Workflow e um workflow n8n pronto para instalar — conecta suas ferramentas em minutos, sem codigo.

Instalar por $49.0 →