Today I shared my repurposing journey with the babesnet.xyz community and this is the recap of the session. If your couldn’t make it, here’s a recording of the same webinar. Everything you see here was written with my repurposing system live excluding this paragraph and the slides section. The video was inserted manually after.
Two months to write one book. That was meant to be a two-day job, and I wasn’t doing it again. So I stopped treating the finished book as the end point and started treating it as raw material. What came out the other side is a system that turns one source asset into six separate content engines, most of them running without me touching a keyboard.
Here’s the system, why it works, and what I’d change if I were rebuilding it today.
Check the slides:
See the presentation via the link or the microsite.
Or down below, I have given you a lot of options:
Why I Wrote a Book First
No book existed on SEO and GEO for crypto and Web3. That gap was the whole opportunity. Few people read books any more, which sounds like a problem until you realise it means very few competitors will bother writing one. Low competition, high credibility signal, and a permanent asset sitting behind everything else I publish.
The plan was simple in theory: write it fast with AI, ship it, and use the book to kickstart a repurposing machine. Reality got in the way, though: I cared too much to ship AI slop, so two days became two months. I had never written a book before and had no system, and without a system my attention drifts and momentum dies. That’s the part that eventually forced me to build one.
Bad Content Beats No Content, But Prompting Decides the Rest
Bad content still has a chance to reach customers, rank, and show up in LLM answers. No content has no chance at all. That’s not an argument for shipping slop, it’s a reminder that the fear of imperfect output shouldn’t be the thing that stops you publishing.
The actual difference between bad AI content and good AI content is prompting. A brand voice skill and a proper knowledge base or factsheet solve most of the quality problem on their own. Every asset in this system runs against a Claude skill and a fact base, not a bare prompt, which is the difference between output that sounds like me and output that sounds like a language model trying to sound like me.
Manual review still isn’t optional. No AI replaces your own eyes on the final draft, however good the skill is.
How to Write a Book With AI Without the Slop
The nine-step process I landed on:
- Drop source docs into a Claude project and bounce goals, format, length, and audience back and forth before writing a word.
- Build an outline, get other LLMs to review it, then lock the final chapter list.
- Build a custom writing skill that captures voice, rules, and style, so every chapter sounds consistent.
- Write chapter by chapter, not all at once, to keep momentum and quality even.
- Run a humaniser pass to strip common AI markers from the text.
- Get a second round of AI critique to catch what the first pass missed.
- Add a hook or a human line to open each section.
- Manually review everything yourself. Non-negotiable.
- Make time. There’s never enough of it, so you make it.
Turning One Book Into Six Repurposing Lines
The book didn’t need to be a destination. It needed to be a source, so it became the input to six distinct lines rather than a single asset I’d finish and move on from.
Line 1: NotebookLM images. Upload each chapter to NotebookLM, generate the slide deck option, and cherry-pick one image per chapter that’s accurate and tied to the source material. Fast, and it avoids paying for another design tool.
Line 2: 60 blog posts. Each chapter became a content pillar, and each pillar spawned a cluster of posts written by a blog-writer skill with LLM-ready formatting and a monster internal linking structure. A separate publishing skill pushes one post a day through the WordPress REST API, with the Claude Chrome extension as a fallback when the API can’t do something. One post scheduled per day, hands-off, for sixty days straight. Half of my top ten blog posts by traffic were written with this exact system.
Line 3: social posts. A blog RSS feed automated through Make turns every published post into a LinkedIn and X post, no manual step involved. 120 pieces of content, completed and running.
Line 4: video. Each post becomes a ninety-second talking-head reel: scripted, generated with an AI avatar plus motion graphics, and published to YouTube, LinkedIn, X, and Instagram with adapted captions per platform. The stack is a Claude skill plus the HeyGen MCP for generation, ElevenLabs for voice, and the Blotato MCP for publishing and scheduling. 240 pieces, live and running.
Here’s the video we created during the session about the content of the session (very meta, I know!):
Line 5: carousels. Each post becomes a branded carousel for LinkedIn and Instagram through a Claude skill plus the Gamma MCP for generation and Blotato for scheduling. 120 carousels.
Line 6: long-form social articles. Each post becomes an article on LinkedIn and Medium through a Claude skill plus the Narrareach MCP, since neither platform offers a proper publishing API. 120 articles.
One book. Six repurposing lines. A system where I don’t need to think about what’s next, because the pipeline already knows.
The Tools Holding the System Together
Two things make this possible that didn’t exist a couple of years ago: the Claude Chrome extension and MCP connectors.
The Claude Chrome extension lets Claude operate a real browser, which matters most when there’s no MCP or API for a tool. I use it for replying to comments on X and LinkedIn, building GA4 and Looker Studio dashboards, setting up Google Tag Manager, troubleshooting WordPress and Elementor when the REST API hits a wall, and running the same bulk task across a list of properties. Left unattended, it’s repaired websites, renewed SSL certificates, added cron jobs, and found and cleaned a virus, entirely on its own.
MCPs connect Claude to real, live data and real distribution instead of guessing or manual copy-paste. The connectors I actually use, in order: Google Search Console, DataForSEO (which replaced Semrush), HeyGen, Blotato, VidIQ, the WordPress REST API, and Lovable. Chrome extension plus MCPs together mean Claude can execute the task rather than just describe what I should go and do myself.
Reinforcement and Distribution Are the Real GEO Lever
LLMs don’t trust one source. They trust patterns, and patterns come from confidence. One clear explanation repeated ten times across independent, trustworthy platforms beats a hundred pieces of scattered content that all say something slightly different. The goal isn’t more content, it’s getting the rest of the internet to say the same thing you say, consistently and independently, across multiple platforms.
Ahrefs data backs this up from a different angle: YouTube mentions have a 0.737 correlation with AI visibility, higher than backlinks, domain rating, or page count, and the pattern holds for both Google AI and ChatGPT. This is also why the repurposing lines feed each other. A post that ranks in Google, gets cited by ChatGPT, and shows up as a LinkedIn article and a YouTube video is the same explanation reinforced four separate ways, not four separate pieces competing for attention.
The Results So Far
Three months against the prior three months: AI search became the number one source of leads, organic sessions rose 37%, organic social sessions rose 344%, and organic video sessions rose 44%. Content engagement is up 72% and impressions are up 30% over the prior ninety days. Ask any LLM to recommend a Web3 SEO consultant now and this system, not a single piece of content, is what shows up.
What’s Next
The repurposing system is being fed back into itself: every skill, MCP, workflow, image prompt, and orchestration pattern built during this project is going into a single Claude Plugin running inside Claude Cowork. One deployable thing, rather than a collection of scripts I have to remember how to run.
Frequently Asked Questions
What is AI content repurposing?
AI content repurposing is the process of taking one source asset, such as a book, blog post, or video, and using AI tools and skills to automatically produce multiple content formats from it: blog posts, social posts, video, carousels, and long-form articles, without rewriting each one from scratch.
How many pieces of content can one source asset actually produce?
This system took one book and produced sixty blog posts, which then fed five further lines: 120 social posts, 240 video pieces across four platforms, 120 carousels, and 120 long-form articles. The multiplier comes from treating each output as a new input, not from any single tool.
Does AI-generated content rank if nobody edits it?
Rarely, and it shouldn’t be trusted to. A brand voice skill and a proper knowledge base fix most of the quality gap, but manual review of every piece before publishing is still non-negotiable. Bad content still outperforms no content, but that’s an argument for shipping consistently, not for skipping review.
What tools do you need to build a system like this?
The core requirements are a model with skills or custom instructions for brand voice, MCP connectors into your publishing tools (a scheduler, a CMS, a video generator), and a browser automation layer such as the Claude Chrome extension for anything that doesn’t have an API yet. None of it needs to be built at once; each repurposing line was added as a separate module.
Want to see this book, the book behind it, or book a call about building the same system for your team? Find everything here.
I built this repurposing system to run my own content, and now I build the same thing for other Web3 and AI teams through my AI marketing services. If you want a system that turns one asset into a full distribution engine instead of a single post, book a strategy call.
Feedback & Testimonials for my AI Content Repurposing Webinar
“Hi Victoria, joined your babes- session yesterday. It was super inspiring- Thanks for this! Hanna”
Original (Spanish): “Hola Victoria! Estuve en el workshop que diste recén. Me pareció impresionante. Estoy aprendiendo IA y como mi área es el diseño de productos y sistemas de branding, esto no lo tenía. Me gustaría conectar con vos. ¡Que tengas un lindo día!”
English translation: “Hi Victoria! I was at the workshop you just gave. I found it impressive. I’m learning AI, and since my field is product design and branding systems, I didn’t have this. I’d like to connect with you. Have a lovely day!”
Tuti Nicola, Social & Growth Specialist LATAM, on LinkedIn: “The class was amazing! Thank you so much for sharing your processes and the magic of repurposing and automating content. I loved meeting you, and you absolutely rock!”
“my mother in law has 8 pages of notes”, shared after the session.
Grateful for those who listen to me speaking for 90 mins about automating content with AI.
Thank you for facilitating @babesnetxyz . https://t.co/1nu1tNI728
— Victoria: Web3 SEO •AEO • GEO • AI Content Systems (@victoria_olsina) August 28, 2026
