AI for Marketing Webinar February 5th

This session was a practical hour on implementing AI across the marketing funnel, following the AI for Sales webinar from two weeks earlier. The structure was the one I use every week: quick wins you can apply immediately, a live build of a real system, demos of how I actually work day to day, and a Q&A. It is aimed at marketers who want results from AI tools without pretending marketing is only about posting funny images on Instagram.

What is marketing in 2025? (4:30)

Before touching any tool, I asked AI to generate a mind map of what marketing looks like in 2025 (5:04). The point of the exercise: marketing is very, very multidisciplinary now. Influencer marketing, affiliate marketing, email marketing, content marketing, search engine marketing, all different disciplines under one word. It is impossible to cover each one in an hour, so I stuck to the fundamentals that apply across all of them and picked four use cases (6:20): content repurposing, research and trends, analytics and tracking, and lead magnets and assets.

One deliberate omission (7:00): I skipped AI content generation, especially video generation. People are fine with AI memes, but as soon as generated content tries to mimic human behavior, audiences push back. It is a touchy subject, and I believe you should master repurposing your own real content first.

For the quick wins you could follow along with ChatGPT, Perplexity, Gemini, or Claude (2:09). Gemini is by far the best pick for marketers right now because of Nano Banana Pro. The advanced part of the session used Nano Banana Pro, Antigravity, Claude Code, and Apify (2:54).

Quick win 1: content repurposing with Nano Banana Pro (9:01)

Nano Banana Pro is Google's image generation model, and its superpower for marketers is infographics. I showed a demo prompt first: an infographic of all the political regimes in France starting in the 1500s, generated correctly from two lines of instruction (10:03).

Then the marketing version (10:34). Industrial and brick and mortar companies sit on huge libraries of case studies, success stories, and white papers that never get reused, and they are great content. I pasted one of the success stories from my own agency website and asked for an infographic highlighting the pains and solutions. The result pulled the client's wasted hours, the challenge, the solutions we implemented, and the impact straight from the case study, ready for social media, presentations, or a catalog you send to prospects.

Two refinements matter. First, quality: Nano Banana and Nano Banana Pro are two completely different things, and the difference is day and night (13:31). My live run fell back to the non-Pro model because I had burned my Pro credits preparing the webinar, which is exactly why the first output looked ugly. Second, branding (14:24): give it a screenshot of your website or your brand guidelines document and it adheres to your colors and style.

On prompting, a question came in about sharing my prompts, and my honest answer is that prompting is kind of outdated. These were two-line prompts. The limitation is not prompting skill, it is knowing what the tool can do, and you can use AI to improve your prompts anyway.

For heavier usage I recommended Higgsfield (15:03), a platform with access to image and video generation models in one place. On the yearly plan it is 17 dollars per month, with the next tier around 24 dollars per month, and 600 credits equals 300 Nano Banana Pro images (16:09). The 24 dollar tier gives 1,200 credits plus unlimited Nano Banana Pro at 2K resolution, and you avoid the Gemini watermark and sizing limits. Nano Banana Pro also generates equations and educational infographics, which is going to make personalization in education phenomenal (17:46).

Quick win 2: research and trends with deep research (19:31)

If you have ever sat in front of a camera not knowing what to say, this one is for you. Trends change too fast to track manually, so I use the deep research feature that now exists in every major AI tool.

Before running anything, privacy settings (20:47). In ChatGPT, go to settings, then data controls, and turn off the option to improve the model for everyone, because that means they keep your data. In Claude it is under privacy. Mine is actually on, because Anthropic is the only company I am comfortable keeping my data, but that is a personal choice. In Gemini you cannot disable it. Perplexity has the option too, though they keep moving it somewhere hard to find. As the saying goes, if it is free, you are the product.

Then the demo (22:41): the same trend research prompt launched simultaneously in ChatGPT, Claude, Gemini, and Perplexity. ChatGPT asks clarifying follow-up questions about niche, format, and platform before it starts. The four reports come back complementary, and that is the trick: feed all four outputs into one AI and ask it to merge them into the best possible report. The sample output showed what type of content is trending, high-performing keywords people actually search on YouTube, and which tools creators care about right now, like Veo 3, which is itself a content opportunity (26:19).

Live build: Apify's YouTube scraper plus Claude Code (27:03)

Deep research gives you synthesized data. For raw, live data at scale, I use Apify (27:53), a platform of scrapers, robots that extract information from the web. Apify calls them actors. You can use it on Instagram to find influencers and creators posting under a given hashtag, then filter by follower count (29:24). You can combine systems: one that finds the influencers and competitors, another that pulls all their content.

My context: I started my YouTube channel about a month before this webinar, and I realized I had no idea what type of video I should make next. So we built the answer live.

The YouTube scraper costs 5 dollars per 1,000 videos (30:00). You choose how many videos, pass URLs or search terms, and optionally download transcripts. A quick UI run on the search term ChatGPT returned the latest videos, including one sitting at 4.5 million views with 82 thousand likes. But ten videos is something I can check with my own eyes. I wanted more.

So I connected the actor to Claude Code (33:04), the AI agent that lives in the terminal. It looks intimidating, like the hacking scenes in movies, but it is simple to use. I told it: use this YouTube Apify scraper, get me one hundred videos on ChatGPT, I want to understand what performs well (34:09). It read the actor documentation, wrote its own fetch script, launched the run, and pulled the data.

While it worked, I showed the same analysis I had prepared on 100 Antigravity videos (37:30): top ten videos by views, performance broken down by video length, the top five channels to study, and key insights like the fact that there was no strong correlation between duration and views while extra-long videos got better engagement.

The ChatGPT dashboard came back during the session (40:11, 44:18) and the findings were sharp: fear and controversy hooks win, with a "truth about AI" style video at thirteen million views; strong personal brands and mega channels dominate; generic how-to-use ChatGPT tutorials do not work, which stung because I was running a series exactly like that; brand and corporate content performs worst; optimal video length is ten to twenty minutes; and winning title patterns use numbers, years, and all-caps words. One hundred videos analyzed in a couple of minutes.

Analytics and a content management system (41:13)

Analytics is where most marketers drown in KPIs. I showed the content management system I built for myself: every LinkedIn post and every YouTube video, with impressions, comments, shares, average view time, publish date, views, and watch time in one place. Setup was genuinely simple, I connected the YouTube API to Claude Code and Antigravity, and for LinkedIn I use Unipile, which connects your LinkedIn account via API and also supports Facebook and Instagram (42:44).

Live on the call, Claude Code ran scripts in parallel to refresh the numbers, updating view counts and computing sub conversion, how many views I need to gain one subscriber. Then I pushed further (44:18): deep analysis on every one of my videos, stacked performance graphs on the same time scale, cumulative channel growth, retention versus average views, and traffic source breakdown, all generated on the spot as an HTML dashboard (46:33). If you want to keep these dashboards, just bookmark the file or move it to your desktop.

The YouTube upload workflow (48:46)

My upload pipeline runs on transcription. Claude Code extracts the audio from a finished video, transcribes it with Whisper, OpenAI's model for turning audio into text (not to be confused with the similarly named dictation tool for talking to your computer), moves everything into the right folder, and uploads to YouTube. The descriptions are AI generated, and so are the timestamps: AI reads the transcript and writes the chapter markers you see on my channel.

On the audience question about creating shorts from long videos (50:54): yes and no. AI is excellent at finding the high-intensity moments in a transcript, the "here is the one thing you should not forget" beats. The hard part is format, because my videos are not shot for vertical. I pointed to Remotion (51:31) for programmatic format conversion and editing with code, with the honest caveat that I had not used it myself yet, only had it recommended by a friend.

Lead magnets and assets (54:02)

Content is an asset: reusable value, whether tutorials or templates. My position is that if lead magnets are your core business, create the core one by hand, then use AI to multiply it into niches. One solid "AI for marketing" asset becomes AI for marketing for content creators, for email agencies, for any niche you serve. One strong lead magnet, multiplied by AI.

Q&A highlights (55:49)

Why Claude Code as your main tool? Because it is the best way to build any type of workflow right now. It has become the AI assistant that is always next to me, and it saves me massive amounts of time. During the Q&A it even proposed lead magnet ideas I could have it build.

Can Claude Code create n8n workflows? Yes, but they are not very good end to end. It helps you test and saves time, but for precise workflows I advise against relying on it one hundred percent (57:45).

Where do the recordings and resources live? I shared my resource page with links to support tools like Apify and TheirStack, coupons when I find them, a Claude Code setup guide, and the webinar library (58:45). A participant flagged that some sessions were missing, which is because there are two libraries, one public and one exclusive to the MakerSchool community, and I was mid-migration from Google Drive to YouTube.

That was the session: content repurposing with Nano Banana Pro, trend research across four AI tools at once, a live scraping and analysis system, and the analytics stack behind my own channel. For the rest of the series, past and upcoming sessions are on the webinars page.