AI content repurposing agent: one draft, one variant per platform, published

A repurposing agent takes one draft, writes one variant per platform, then publishes. It has to: Threads caps a post at 500 characters, a YouTube title at 100, and Instagram refuses PNG outright. Here is the pipeline, the limits that force the rewrite, and what the tools cost. Everything checked on October 9, 2026.

Repurposing is not cross-posting

Cross-posting sends one payload to every account. I wrote the tooling side of that in post to all social media at once. Repurposing inserts one step before the send: rewrite per destination.

The reason is not taste, it is arithmetic. The same text cannot fit everywhere:

Threads figures from Meta's Threads posting docs, Instagram from Meta's content publishing guide, both read today. The per-platform media rules are in the PlugKit publishing guide.

Rate limits bite too: Instagram allows 100 API-published posts per rolling 24 hours, Threads 250 published posts per 24 hours, a carousel counting as one in both cases. Repurposing one blog post into six platform variants is nowhere near those ceilings, which is why this is a quality problem, not a quota problem. The full table lives on social media API rate limits.

What an AI content repurposing agent actually does

Four steps, and only the second one needs a model:

1. Read the source. A blog post, a transcript, a long LinkedIn draft. Plain text in.
2. Write one variant per platform. Different length, different opening line, different call to action. The agent needs the limits above in its instructions, otherwise it writes 900 characters for Threads and the publish fails at the API, not at the model.
3. Show you the variants. Six drafts in one message, before anything leaves the machine.
4. Publish, one call. Each target carries its own text.

Step 4 is the part people hand-roll for weeks. The API takes per-target overrides directly: use the platforms array instead of accountIds, and each entry sets its own customContent and customMedia. Exactly one of the two fields is required, so you pick identical or per-platform at the call site.

curl -X POST https://api.plugkit.co/v1/posts \
  -H "Authorization: Bearer sk_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "profileId": "YOUR_PROFILE_ID",
    "content": "fallback text",
    "platforms": [
      { "accountId": "ACCT_LINKEDIN", "customContent": "The long version, 1200 characters, three paragraphs." },
      { "accountId": "ACCT_THREADS",  "customContent": "The 400-character version with the hook first." },
      { "accountId": "ACCT_X",        "customContent": "The one-line version plus the link." }
    ],
    "publishNow": true
  }'

One request, three different texts, three platforms. The response returns a target per account with its own status and, once live, its platformPostUrl, so a partial failure tells you which variant did not land.

Repurpose content ai: the three families of tool

They are not competing, they solve different thirds of the job. Prices below are from each vendor's own pricing page, read on October 9, 2026.

Read that last row honestly: PlugKit does not repurpose anything. It is the publishing layer the agent calls once it has written the variants. If your bottleneck is turning a 40-minute podcast into nine vertical clips, OpusClip does something PlugKit never will. If your bottleneck is that nobody is holding the pen, a scheduler will not fix it either.

Buffer is the better pick if you want a human in a composer with AI suggestions, and it is cheaper below about twenty channels. The $5 per channel maths turns against it there, where a flat $97 for 30 wins, and Buffer has no MCP-style interface for an agent to drive it end to end.

Content repurposing tool vs agent: the difference that matters

A tool waits for you. An agent runs without you. That sounds like marketing until you look at what each one needs:

The failure mode of the agent version is worth naming: a vague brief produces six variants of the same bland paragraph, which is exactly the outcome repurposing was supposed to avoid. The fix is boring. Give it your actual constraints (limits, banned phrases, the hook style you use) in writing, and read the variants before approving. I keep a human approval on step 3 and I am not planning to remove it.

AI repurpose blog to social: the shortest working setup

What I run, and the smallest version of it that works:

• One markdown file as the source of truth.
• An agent with a social MCP server connected, so it can list accounts, draft and publish in one session. The PlugKit MCP server exposes 103 tools and three workflows, one of which (publish) shows the targets and the final text before it sends. Verified today.
• A brief that includes the character limits from the table above, and two examples of my own writing.
• Approval before publishing, every time.

Whether you give the agent MCP or just the REST API is a real decision with a real cost, and I wrote it up separately in MCP or REST API for your agent. Short version for this use case: MCP for the drafting session where you talk to it, plain HTTP for the nightly job that ships an approved queue. For what the model handles well and what I will not delegate, see Claude as your social media manager.

What it costs

PlugKit is $49 a month for 5 connected accounts, cancel anytime, or $97 a month for 30, every tool included. Seven days free, no card, with one connected account during the trial. Add your model provider's tokens for the rewriting step, and X API calls if X is one of your targets: those are passed through at X's own rate, $0.015 to publish a post, and $0.200 if the post contains a URL. Billing figures from docs.plugkit.co, checked on October 9, 2026.

FAQ

What is AI content repurposing?

Taking one piece of content and rewriting it for each destination instead of copying it. A 1,200-word blog post becomes a LinkedIn post, a 400-character Threads post, a one-line X post with the link and a Reel script. The rewriting is the work. Publishing the result is one API call.

Is repurposing the same as cross-posting?

No. Cross-posting sends an identical payload everywhere, which is one request and zero thinking. Repurposing writes a variant per platform first. The practical giveaway is the API call: cross-posting passes a list of account ids, repurposing passes a per-target customContent.

What is the best content repurposing tool?

It depends which third of the job you are stuck on. OpusClip for cutting video into clips, from $15 a month. Repurpose.io for automatic routing between platforms, from $35. Buffer for a composer with AI suggestions, $0 for 3 channels then $5 per channel. An agent plus a publishing API when you want the drafting and the shipping to happen without you in the loop.

Can AI repurpose a blog post to social media automatically?

Yes, and the publishing half is solved. The honest caveat is the drafting half: with a thin brief the variants read as paraphrases of each other, which performs worse than one good post. Give the model your constraints and examples, and keep an approval step before it publishes.

Does the same post on every platform really perform worse?

The part I can prove is mechanical, not statistical: the same text often cannot be published everywhere. Threads truncates at 500 characters, a YouTube title at 100, Instagram stories take no caption at all. Anyone quoting you a precise engagement lift from repurposing is quoting their own sample, not yours.

AB

Antoine BlancoFounder of PlugKit. Runs his own Instagram, YouTube and X accounts through the API and MCP server described here.