I first gave this talk at the Masterminders Conference (One Book, 250+ Pieces of Content) and then I created a longer version and more in detail version for my newsletter subscribers (One Book, 500+ Pieces of Content).
Also, I have some gossip for you, at the Masterminders conference I was named the Top AI Content Specialist of 2026 💅
Something that still surprises me: my book Mastering AI Search for Crypto and Web3 Brands, became the second biggest source of leads in my business, behind referrals, in a matter of months.
From that book came more than hundreds of pieces of content repurposed across every channel I care about.
This is the full breakdown of how the system works, the exact tools behind each step, and the results so far.
Watch the Webinar Recording:
Check the slides:
Why a book, and why it could not be slop
The reason was authority. No book existed on SEO and GEO for crypto and Web3, and that gap was the opportunity. Very few people write books anymore, which is exactly why so few would copy the move. Low competition, high credibility signal. The real question I was answering: does AI recommend me?
The plan was to write it fast with AI. That plan did not survive my own standards. I sell AI bots, skills, and automations that talk like a brand, so shipping a book that read like generic AI output was never an option. Two days became two months.
Here is the belief that carried it, and it is unpopular: bad content is better than no content. Bad content still has a chance to reach a customer, rank, or show up in an AI answer. No content has none. The difference between bad AI content and good AI content is prompting. A brand voice skill and a fact sheet of real data separate the two, and every output in this system used both.
If you want the deeper framework behind this, it is the foundation of my work on AI content systems and agents, and the book itself goes far further: Mastering AI Search for Crypto and Web3 Brands.
The nine-step system for writing a book with AI
I had no system, and without one my work drifts. That is what turned two days into two months. Here is the process that finally worked:
- Gather your source material into one project and bounce ideas back and forth: goals, format, length, audience.
- Build an outline.
- Get other LLMs to review the outline, then lock the final chapter list.
- Create a writing skill that captures your voice, rules, and style, so every chapter sounds like you.
- Write section by section, not all at once.
- Run a humaniser skill to strip the common AI markers.
- Get other LLMs to review again. A second round of AI critique catches what you missed.
- Add a hook or a human line to the open of each section. I built a skill for that too.
- Review everything manually yourself. This step is non-negotiable.
Steps six and nine are what separate this from slop. No AI replaces your own eyes.
The image discovery: NotebookLM
Breaking up 130 pages of text needs visuals. The best ones came from NotebookLM. Upload each chapter, use the slide deck option rather than the infographic one, and cherry-pick the best image per chapter.
Editorial-style images worked far better than dense infographics. They are accurate, visually distinct, and tied to your actual source material rather than generic stock. The same tool also generates quizzes and flashcards if you want extra assets from the same source.
Here’s an image example:
The launch: an asset built to be repurposed
The book was designed as a source, not a destination. The launch ran on several tracks at once:
- Free on my site behind a newsletter signup
- Paid on Amazon Kindle, with a custom GPT and bonuses for implementation. Check it out!
- A PR push for LLM seeding
- Promotion across my own blog, social, and newsletter
- A launch webinar on Luma, with the recording on YouTube
- Pitching conferences, podcasts, and media outlets with the book
LLM seeding is worth defining, because it is the heart of this. It means telling the world what you do and multiplying that message across many independent outlets, so models capture it quickly.
That was a decent job. I was happy with it. Then I did the maths: two months for one book. I needed to squeeze every word out of it.
Why repurposing is a search strategy now
For years the question was whether Google could find you and rank you. The new question is whether AI recommends you. When someone asks ChatGPT, Perplexity, or Gemini for the best provider in a niche, you are either in the answer or you are invisible.
That shift rewards one specific behaviour. LLMs do not trust a single source, they trust patterns. Patterns build confidence. One clear explanation repeated ten times beats a hundred pieces of scattered content. The goal is to get the rest of the internet to say the same thing you say, consistently, independently, and across multiple platforms. Also, this what I explained on my book: “PART 4: Distribution & Reinforcement“.
Repurposing done properly is how you build that pattern without burning out.
The book as a source asset
Two months for one book. I was not doing that again. The book did not need to be a destination, it needed to be a source. So it became the input to a set of repurposing lines, each one multiplying the last, all running through AI skills and automations.
Book to 60 blog posts
Each chapter became a content pillar, and each pillar spawned a cluster of posts. A blog-writer skill handled the writing: custom image prompts, LLM-optimised formatting with headers, FAQs, and schema, the brand voice and fact sheet baked in, and internal linking built on Tom Riley’s Monster internal linking method. Publishing ran through the Claude Chrome extension and the WordPress REST API, one post scheduled per day, hands-off. Sixty posts. Sixty days of daily content from a single source.
Solving the image problem at scale
This was the hardest-won part. Claude cannot generate images, and I could not extract the NotebookLM images cleanly from the document. Sixty posts needed two to three images each plus a thumbnail, up to 180 images, all branded, watermarked, and under 50KB per file.
Manus was the only tool that could produce branded images at that volume and file size, logo watermark included. They were bulk generated, then bulk uploaded to the WordPress media library, with a structured filename key that auto-matched each image to its correct post on upload. No manual sorting, no mismatched thumbnails. Anyone can generate an image. Very few people have solved the matching problem at scale.
For example:
The plumbing: MCPs and the Chrome extension
None of the lines below work without two capabilities, so they are worth understanding before the rest makes sense.
MCPs, also called connectors.
An MCP works a bit like an API. It lets your LLM pull data from a platform and act on it. If your team runs on Notion or Linear, connect the MCP and you can ask Claude which tickets are blocked, and it answers from live data rather than guessing.
That is the read side. The write side is what matters here. My workflow runs on three: Blotato and Narrareach for scheduling, and HeyGen for video. So when Claude finishes writing something, I do not copy it anywhere. I say publish it, or schedule it for Friday at 8pm, and it happens.
There is an MCP for almost everything now. Connect the tools you already use, and your LLM has the context and the ability to act on it.
The Claude Chrome extension
This one operates your browser. I asked the room how many were using it and got about six hands out of thirty. If you are in the other 77%, start. I use it for data extraction and task automation: go to my WordPress site, take these blog posts, publish them. It fills forms, moves around, and does the things I do not have time for.
The point of both is the same. I do not want an LLM that tells me how to do something. I want one that does it. Life is too short to be scheduling posts by hand. If you are not using LLMs to execute tasks yet, that is the change worth making this week.
The six repurposing lines
Quality did not get sacrificed for volume. Every piece runs through an accuracy gate that checks it against the source, with the book as the ground truth, plus a human checkpoint. Automation handles production, a human handles judgement. That is the line that separates this from the AI content the industry is already learning to distrust.
Line 1: the original YouTube workflow
The automation I already had. It takes a video, pulls the transcript, writes a blog post with title and meta description, publishes it to WordPress, then spins out an X article, LinkedIn article, Medium article, LinkedIn post, a newsletter draft, and messages for the communities I am part of. Seven or eight pieces per video. Half of my top-performing blog posts came from this automation, bringing traffic, leads, and backlinks. The automation I already had, turning one video into seven pieces across platforms.
Here’s the full breakdown of the YouTube repurposing workflow that started all of this if you want to learn more.
Line 2: the book becomes 60 blog posts
Each chapter became a content pillar, each pillar a cluster of posts. A blog-writer skill handled the writing with brand voice and fact sheet built in, LLM-friendly formatting, and internal linking built with Tom Riley’s Monster Internal Linking skill (get it here). Publishing ran through the Claude Chrome extension and the WordPress REST API. One post per day, hands-off, telling Google and the LLMs exactly what the book is about.
Line 3: social posts
The blog RSS feed triggers a LinkedIn post and an X post the moment a post goes live, automated through Make. 120 pieces. Each one ends by pointing back at the book, reinforcing the message again.
Line 4: video
Each post becomes a talking-head video with an AI avatar and motion graphics, scripted, generated, and published across YouTube, LinkedIn, X, and Instagram with captions adapted per platform. Built with a Claude skill, HeyGen MCP, ElevenLabs for voice, and Blotato for scheduling. Here’s an example which even fooled a friend of mine, she thought this was me – not an AI avatar.
This wasn’t my first AI experience with video, if you are curious here’s my full original AI video workflow.
Why video specifically? It has the strongest correlation with AI visibility of any channel, for both Google AI Overviews and ChatGPT. Gemini can read your script, so every keyword you say out loud is usable. If you only do one channel, do video.
Line 5: carousels
Each post becomes a branded carousel for LinkedIn and Instagram, drafted by a skill and produced in Gamma from a locked template, then scheduled through Blotato.
Here you can see a live carousel example: https://www.linkedin.com/posts/victoriaolsina_media-attachment-activity-7480203983924523009-rJK7?utm_source=share&utm_medium=member_desktop&rcm=ACoAAASfgtoB3SA3J94ViRxbpjbkwj_uBglcgjU
Line 6: long-form articles
Each post becomes an article on LinkedIn and Medium, because both are authoritative domains with a real chance of ranking. This one is harder than it looks: LinkedIn has an API for posts but not for articles, and driving the editor with a browser extension gets messy fast. Narrareach MCP solved it, and it handles LinkedIn, Medium, and Substack.
Here’s a Linkedin article example: https://www.linkedin.com/pulse/four-layers-generative-engine-optimisation-why-your-seo-olsina-2myhe/?trackingId=uBirPFQ4Tb2nCRFzmzZW2A%3D%3D
Medium article: https://medium.com/@volsina/brand-mentions-seo-why-ai-cites-your-content-but-recommends-your-competitor-ad08873554d9
The results so far
Three months versus three months, as of 17 June 2026, from Google Analytics:
- AI search is now my second biggest source of leads, behind referrals only. A channel that barely existed for me a year ago. How do I know? My calendly intake form asks everyone who books a call how they found me. That self-reported attribution puts AI search second.
- Organic search sessions up 37%.
- Organic social sessions up 344%.
- Organic video sessions up 44%, and the AI-generated videos are the fastest-growing content on the channel.
And the part the whole thing was built for most LLMs: ChatGPT, Grok, or Google AI Overviews, recommend me as SEO consultant, and I am generally in the answer, with my site cited as a source and the videos showing up too.
ChatGPT recommends Victoria Olsina as Web3 SEO consultant:
Grok recommends Victoria Olsina as Web3 SEO consultant:
Google AI overviews recommends Victoria Olsina as Web3 SEO consultant:
Repurposing the repurposers
Here is the part I did not plan. The book project did not just generate content, it generated a better system than the one I started with. Every new skill, MCP, workflow, image prompt, and orchestration pattern built during the project was fed back into my original YouTube repurposing automation, rebuilt in Claude, and repackaged as a single installable plugin that runs inside Claude Cowork. Not a pile of scripts. One deployable thing.
That is the real takeaway. One book. Multiple repurposing lines. Compounding results. A system where I no longer have to think about what comes next, which, for a brain like mine, is the entire point.
What the systems actually costs to run
The most common question, so here is the itemised answer:
- Claude: $20 – You could probably run it with the free version but will hit the limits soon
- HeyGen: $29 – Free up to 3 videos per month
- Blotato: $29 – Free trial
- Narrareach: $19 – Free trial
- Gamma: Free up to 10 cards – I used the $12 version
Roughly $100-110 per month for the whole operation.
You can go leaner. HeyGen’s free tier gives you three videos a month, which is fine if video is not your focus. Blotato and Narrareach overlap, so you can run Narrareach alone and schedule everything through it (and the analytics dashboard is really good).
For comparison, an editor costs at least $300 per video, plus the back and forth and the delays. $100 a month for this is not expensive.
If you want a social media scheduler that is 100% free you can use the self hosted version of Postiz, but you need your own VPS (I use Hostinger for this, OpenClaw, Hermes, etc) and the set up is a bit tricky for non technical people like me – I just gave up before actually getting anywhere.
Want this running in your business?
You can build this yourself with the stack above, or I can build it for you. Helping Web3 and AI brands turn one serious asset into a multi-channel engine is exactly what I do.
Session Feedback: 5/5
I don’t want to brag but people said they found it funny and insightful.
Frequently Asked Questions
How do you know AI search is your second biggest lead source?
Calendly asks everyone who books a call how they found me. That self-reported attribution puts AI search second. The Google Analytics figures cover sessions, which is a separate measure.
What does the whole system cost to run?
Around $100 per month: roughly $20 Claude, $29 HeyGen, $20 Blotato, $20 Narrareach, plus Gamma. You can run it leaner using HeyGen’s free tier and choosing either Narrareach or Blotato rather than both. A single outsourced video edit costs at least $300.
How do you make the AI avatar and the voice?
You train HeyGen with a three-minute video of yourself. It produces an avatar with an ID, and the skill references that ID every time. ElevenLabs handles the voice clone, and it is better than HeyGen’s own voice, particularly for non-standard accents. You do not have to appear at all: HeyGen has stock avatars, or you can use animation only.
Is this just mass-produced AI content?
No. Every output runs on a brand voice skill and a fact sheet of real data, and videos and carousels get read and approved by me before they publish. Volume without fidelity is just noise.
Do I need to be technical to use this approach?
No. The strategy stands on its own and the tooling is explained in plain terms. If you are technical there is enough here to go and build, and if you are not, there is enough to brief someone who can.
What tools does the system use?
The core is Claude, with skills for writing, images, video, and publishing. NotebookLM for editorial images, Manus for branded watermarked images, HeyGen and ElevenLabs for video and voice, Make and n8n for automation, Gamma and Blotato for carousels, Narrareach for article scheduling, and the WordPress REST API for publishing.
Do I need to be technical to use this approach?
No. The strategy stands on its own, and the tooling is explained in plain terms. If you are technical there is enough to go and build, and if you are not, there is enough to brief someone who can, or to bring me in.
Some links above are affiliate or referral links. If you sign up through them I may earn a small commission at no extra cost to you. I only link tools I actually used to build this system.
Let’s build your repurposing system →
