AI content repurposing.
One URL in, posts worth publishing out.

AI content repurposing is taking one thing you already made and turning it into the several smaller things each platform wants, without rewriting them by hand. A blog post becomes a LinkedIn post, an X thread, an Instagram caption and a Threads reply, each written the way that network actually reads.

The word doing the work is rewriting. Reformatting is chopping the same text into smaller boxes, and it reads like it. An automated content repurposing tool worth using reads the source, works out what the point was, and says that point again in a shape that fits where it is going. Same idea, different sentence.

Two kinds of tool get called the same name, and they do opposite jobs. Most of them are video clippers: you give them a recording, they cut highlights, burn in captions and resize for vertical. This one is the text kind. You give it a link and it writes posts. If what you have is a two-hour webinar and what you want is thirty seconds of it with subtitles, a clipper is the right tool and this is not.

The Repurpose Engine accepts two source types out of the box:

• Blog posts and articles: paste any URL. We follow redirects, strip nav, cookie banners, newsletter pitches and footers, then take the article body, preferring the publisher's own structured articleBody when the page declares one. The prose is what the engine reads, not the furniture around it.
• YouTube videos: paste any watch, Shorts, youtu.be or embed link. We read the transcript when the video has captions, and the full description when it does not, and the source card tells you which of the two you got. A video whose captions and description are both missing is refused rather than written about, because posts built on a title are guesses.

Reading is free. Pasting a link fetches and analyses it before you commit to anything, so you see what came back, and whether we could read the page at all, before a single post is generated against your quota.

One quota: every repurpose run consumes from the same posts-per-month bucket as a regular AI generation, so you never have to track two limits.

You pick the number, and the number is deliberately small. A source holds a handful of genuinely different ideas; past that, a batch is the same post restated in new clothes, and finding the two worth publishing gets harder as the count goes up.

Before anything is written, the engine reads the source into its parts: the argument it makes, the distinct claims, the figures with the sentences that give them meaning, the quotable lines, the steps, and what the piece argues against. Each post is then assigned one angle and built on it alone: the core argument, the contrarian take, the stat, the quote, the process, the mistake. That is why three posts from one article are three different posts rather than three openings on the same paragraph.

Each post is also shaped for where it lands:

• Instagram: visual-first captions with hashtag clusters, hook lines optimized for the first two lines (where the “more” truncation happens).
• X (Twitter): single posts inside the character limit AND threaded multi-post breakdowns when the source justifies it.
• LinkedIn: long-form takeaways with structured paragraphs, framed for the professional-context feed.
• Facebook: conversational, community-leaning, with a natural-language CTA at the end.
• Threads: punchy, reply-bait, shorter than X but wider than LinkedIn.

Your Brand Kit (colors, voice references, vocabulary preserve and ban lists) and your saved content preferences thread through every generation, so the posts read like you wrote them, not like a generic LLM did.

Repurpose isn't a paraphrase pass. The pipeline runs the source content through the same closed-loop generator that powers /api/generate-content, with three context layers stacked into the prompt:

1. Brand context: voice references, sample posts, preserved terms (your product / brand names), forbidden terms (competitor names, off-limits topics).
2. Content strategy: your audience persona, content pillars, voice references, KPI focus, posting cadence (captured once during onboarding, editable from BrandKit and Settings).
3. Closed-loop signal: your top three and bottom two posts of the last 30 days are distilled into a brief that biases the rewrite toward what your audience actually rewards.

Net effect: an article you spent an afternoon writing becomes a week of posts that compound toward your real audience instead of a week of posts that read like everyone else's.

The repurpose problem has two existing solutions, both worse than owning the loop end-to-end:

• Repurpose.io ($35/mo on its Content Marketer plan, checked August 2026) is built around redistributing VIDEO between platforms. It does that job well, but it is decoupled from your scheduling, your brand kit, and your engagement history: you re-paste your voice every time, and nothing it publishes feeds performance back into the next batch.
• Pasting into ChatGPT is free but every prompt forgets your voice, your audience, and what worked yesterday. Each rewrite is a fresh roll of the dice.

Drifter's Repurpose Engine is built into the same workflow as generation, scheduling, and analytics. So the posts that come out of it land on the calendar, ship to your platforms, and start feeding the same closed-loop signal that improves the next batch. One tool, one bill, compounding output.

1. Paste the URL into the Repurpose tab. It reads the page immediately, for free, and shows you the source: what it read, how many words, and how many claims, figures and quotes it found.
2. Pick 1, 3 or 5, and your platforms. The angles the source can support are listed before you commit, and the split across your selected platforms is shown as you change it. A source that only carries two distinct angles will not offer you five.
3. Review the batch. Every post shows the angle it was built on, its voice-match score, and whether it checked out against the source. Copy, edit in place, rewrite a single post without touching the others, or drop one.
4. Approve and schedule.Drop the batch into the calendar at your best posting times, hand it to Auto-Pilot, or one-click publish a single post on the spot.
5. Closed loop kicks in. The same analytics worker that watches the rest of your posts watches these too. Patterns from your repurposed content feed the next generation.

The repurpose route is deliberately scoped. Things that look like features but we explicitly avoid:

• No video clipping. Most tools ranking for “content repurposing” cut long video into Shorts and Reels: OpusClip and its peers do that job well, and we do not do it at all. This engine reads a source and writes posts. If what you want is clips, you want one of those tools; if what you want is a week of written posts from one link, scheduled and measured, that is this.
• No verbatim copying. Every output post is rewritten in your voice. The source content is reference material, not the published text.
• No paywalled content. The extractor honors paywalls and won't bypass them. Paste a URL you already have legitimate access to read.
• No silent attribution drop.For YouTube videos, the generator includes attribution lines in the post by default (you can edit them out in review).
• No invented numbers, and we check. Telling a model not to make things up is an instruction, not a guarantee. Every finished post is compared back against the source, and one containing a figure the source never gave is flagged on the card before you publish it. Same for a post that barely references the source at all.
• No padding to hit a number.If you ask for five and the source honestly carries three, you get three and a note saying why. The alternative is two posts that repeat the other three, which is what you would have to delete anyway.

The URL you paste, the extracted text, and the generated posts are stored only in your account. We don't train any cross-customer model on the source content or the output. We only fetch the public page, video transcript or oEmbed data for the link you give us (nothing behind a login), and we don't retain the source page after extraction returns. What we do cache, for a day, is the structured reading of a source, so pasting the same link twice costs one read rather than two.

Full data-handling detail lives on the security page.