My first attempt at automating content marketing failed. You're reading the second.

Claude interviewed me for this post. Eleven questions, one at a time, about 30 minutes - then it drafted what you're reading straight from the transcript.

This is my second attempt at automating WaveGen's content marketing. The first one failed, and the failure turned out to be the useful part.

Two to three posts a week was eating my entire week

I keep a Notion library of content ideas. It's two months old and has 28 ideas in it. 11 have shipped.

The bottleneck isn't ideas, it's writing. A post takes me half a day when I'm interested in the topic, and if I'm not in the mood, good luck. I spent the last two weeks grinding the backlog properly and topped out at 2-3 posts a week - and that took most of my week.

I want content to carry SEO and AEO for WaveGen. At this speed I'm barely making ends meet, not scaling up.

Attempt one: mine my inbox for ideas. I never shipped a single post from it.

I started with Alex Lieberman's approach - the Morning Brew cofounder built a content machine by connecting Claude to his emails and meeting transcripts to mine ideas, then having a panel of AI podcast hosts interview him about each one. So I connected mine and let Claude dig out interesting bits as an oracle.

Here's how that died. A customer told me they found WaveGen through Gemini, so the proposed angle was: industry says Gemini is not useful for AEO, but I've got counterexamples. But after digging, we do see Gemini refers the least amount of traffic among all LLMs.

One customer is a story, not a trend. And that kept happening - the bits the oracle dug out were either not played out yet, or became uninteresting once the interview dug deeper. The mining looked automated and produced nothing worth publishing.

Attempt two: I pick the topic, Claude runs the interview

The fix wasn't better mining. It was putting my judgment back at the front of the pipeline.

Now I propose the topic - something I already know has a real story in it. Claude interviews me one question at a time, pushing past my first answers for the specific example, the number, the moment I changed my mind. Then it drafts straight from the transcript, pushes into the blog pipeline, and the post gets distributed through WaveGen itself.

You're reading the first run. The link I trust least is the one you're judging right now - whether we can draft something interesting and worth reading from this interview.

The bet is 15 minutes per post

This first run took about 30 minutes of my attention. The target is 15.

At half a day per post, 17 backlog ideas is two months of writing I don't have. At 15 minutes each, we can ship all my backlog ideas in a week.

The math only matters if the posts are worth reading, and I don't know that yet. I'm publishing this one as is - you're the reader, so you tell me.

What I actually learned

Use AI as stimulants/brainstorm partner to get more content out of me easily, and assistant to organize the content.

Attempt one asked AI to be the author. Attempt two asks it to get more author out of me.