AI for Content Webinar March 12th

This session was all about AI for content, and my main point was that content is much bigger than content generation. We covered trend analysis, content repurposing, video analysis, and content preparation, with a live build connecting Apify to Claude Code along the way. If you produce videos, posts, or case studies and want to get more out of what you already have, this recap is for you.

Content is more than generation (2:58)

When I started sending out emails for this session, I only wanted to present content generation. But content is a lot broader than that, and a lot of people are against generation through generative AI tools anyway. So instead of arguing about it, I tackled everything surrounding it (6:23):

• Trend analysis: making sure you create content that is relevant for your industry right now.
• Content repurposing: turning case studies into carousels, videos into LinkedIn posts, long form into short form. One piece of human-made content, replicated across channels with AI.
• Content analysis: analyzing any video you have ever created, frame by frame, and asking questions like "in which sections is someone talking about physics?" Your content library becomes a set of living assets.
• Content preparation: most of the time we spend on content is not creation, it is preparing assets, uploading, writing descriptions, wiring everything up.

For following along, ChatGPT, Gemini, or Claude covers every quick win (5:22). For the advanced use cases I use Nano Banana Pro inside Gemini, Antigravity, Claude Code, and Apify for extracting real live data from any social platform.

Quick win: deep research for trend analysis (8:43)

Deep research is one of the most underused features in every major AI tool. It gives you a fully blown report on any topic, so if you want the latest YouTube, TikTok, or Instagram trends for your niche, this is where you start. You can even set it up as a recurring task so you get fresh trend data every day.

I triggered the same prompt in ChatGPT, Gemini, and Claude, asking for a report on the latest trends in AI education and AI integration on YouTube. I did not even type it, I dictated it with Wispr Flow. A deep research run normally takes five minutes for me, but for niche topics it can easily go to thirty, and I have had it run up to an hour. That is normal: it is combining the reasoning capabilities of the model with web research, thinking back and forth about where to find good content.

The report I had prepared showed things like "AI upskilling is exploding" and that the AI automation YouTube niche is in a gold rush phase, with analysis of the communities and channels being created, mine included. You can ask it for precise examples of which channels are growing fastest and build a tracking database from that.

Then comes the reshaping (13:47). Claude can turn that report directly into a native PowerPoint presentation you can show your boss or a committee. ChatGPT cannot do that yet to my knowledge. In Gemini, the move is different (14:57): give the report to Nano Banana Pro and ask for an infographic. The current version can build infographics and even change the perspective of images, which opens up huge opportunities for fields like architecture. Neither output is perfect yet, but you get something presentable within five minutes.

Live build: Apify plus Claude Code (17:35)

Deep research is good, but it is not live data, and generative AI on its own cannot give you live data from YouTube, Instagram, or TikTok. That is where Apify comes in. It is a platform of scrapers that pulls fresh data from any major platform, even Google Maps, and it is extremely cheap: around three dollars for a thousand YouTube videos.

I searched for the YouTube scraper, showed how the interface works with either a channel or a search term, and ran a test search on ChatGPT (19:12). But raw data alone does not help you, you have to analyze it. So I connected the scraper to Claude Code (20:47), which is already wired to my Apify account via API, and asked it to fetch one hundred videos about the migration from ChatGPT to Claude. It triggered the actor remotely, ran four consecutive calls, and came back with 108 relevant videos.

From there, one prompt: generate an HTML dashboard with this data and add recommendations based on the trends you collected (25:03). Dashboards take a few minutes because there is real code involved, but compare that with the alternative. Analyzing 86 videos at once is something that would have taken an intern an entire day, and honestly, Excel is just hell. I never want to see a spreadsheet again.

The dashboard showed almost two million total views across the analyzed videos, with migration tutorials as the top category (33:10). A lot of people are moving from ChatGPT to Claude right now, so it is a fantastic moment to position yourself on migration content. When I noticed I could not filter the results, I asked for basic filtering and got it in a fraction of the original build time, because the dashboard already existed. It even produced concrete content recommendations, including ideal video length based on what is actually getting views.

Advanced demo: video analysis with Gemini (27:34)

Before the webinar I filmed a one-minute video of myself doing math on a whiteboard: a simple operation, a ninth-grade style equation, and a university-level integral. Then I asked Claude Code to analyze the video with Gemini 3.1 Pro, solve the equations, and, for fun, tell me what it thinks of my haircut.

The analysis of a one-minute video took about a minute, and it got everything right: one plus one equals two, the equation solved as x equals minus two thirds, and even the university-level integral, all from handwritten notes on camera. The business case is obvious: you can detect precisely where a given person talks about a given subject inside any video you have ever created, and chunk out pieces of content for repurposing.

My favorite real example: I noticed a security problem in my videos, and I have about ten hours of webinar footage. Instead of watching it all, I had AI analyze every video and tell me exactly where I had shown an API key. It found all of them. I verified. That is mind-blowing.

One caveat I gave in response to a question: once you do this in volume, you need code. At around twenty to fifty videos, chat interfaces stop being the right tool, and you scale through the Gemini API driven by Antigravity or Claude Code.

LinkedIn analytics and a simple CMS (36:14)

I also showed the system that syncs my LinkedIn metrics to Notion. An agent runs at 7:00 in the morning and 7:00 in the evening, fetches every post I have ever produced, and maintains a clean database: 55 pieces of content with impressions, engagement rate, whether it is a lead generation post, and downloaded carousels. I do the same for YouTube, with day-to-day data on views, subscribers per video, and retention. Combine that with the trend system and you can compare your own content against what is currently trending.

This also answers a question from the Q&A: my best performing post ever got seventeen thousand impressions, and I wrote it in two minutes while angry before leaving home. I still have not cracked why it worked, but with everything in one database, I can ask AI what my best posts have in common and what to try again.

Content repurposing: carousels with Antigravity (38:50)

For visuals, I asked Antigravity to create a LinkedIn carousel about the ChatGPT to Claude migration. The interesting part is how it works: it does not generate images at all. It builds web pages in code, screenshots each one, and stitches the PNGs into a single PDF. You can hand it your branding guidelines, or better, have it analyze your website and extract the branding itself. The finished carousel followed my brand, and the whole thing can be automated. Give it a case study and get carousels, Instagram posts, or Facebook content out the other side.

Content preparation: YouTube automation (44:44)

I am extremely lazy with my YouTube channel, so Claude Code handles the whole upload pipeline. I hand it a video and say upload it. The workflow is intelligent: webinars go up as private videos, private sessions stay private, standard videos publish normally, and everything lands in my database. It transcribes, generates the description, pulls the correct affiliate links from my link database so I never ship the plain URL, offers me title options to pick from, and creates my thumbnail. The thumbnail script analyzes my face in the frame, finds the empty space around it, and places tool logos there. Everything is ready within about two minutes.

Q&A highlights (50:37)

Can you share these workflows? Yes. Most of what I showed already exists as skills I created, like the carousel one. The method is simple: do your workflow end to end once, then tell Claude Code to take all of it and turn it into a skill you can reuse. Skills are shareable, so everyone in an organization gets the exact same workflow and the exact same standard, whether it is an intern or the CEO. Quality stops being an issue once you have a good skill.

Can Claude Code edit video? Yes, within limits. Use Gemini to analyze the video, then Claude Code executes the plan: cuts at spoken markers, transitions, intro videos. Extreme editing, no. Blurring, I have not tried, that is still uncharted territory for me.

What about programmatic video? That led to the bonus (55:11): Remotion, which is video created with code, and it is completely free. I showed a video plotting every SpaceX launch on a timeline with parabola trajectories, all generated from code. Because AI is so good at code, it is great at this. It is also part of why my bills keep dropping: with Claude Code or Antigravity at the center connected to pay-per-use services, I no longer pay for tools like Canva.

Wrap-up (1:00:44)

These sessions run every Thursday at the same time. Next week is AI for outbound sales, and I gave a sneak peek of my LinkedIn Prospector: it finds trending AI posts daily, fetches and qualifies them by engagement, and gives me a list to comment on. My posts average one to two thousand impressions and my comments average five hundred, so eight comments a day equals a post in impression impact. It was built with the exact same tools from this session, Apify and Claude Code.

If you want to catch the next session or watch previous ones, the full list is on the webinars page.