> ## Content Index > Fetch the complete content index at: https://madewithlove.com/blog/llms.txt > Use this file to discover other available public pages before exploring further. # My blog posts are 100% AI-generated - URL: https://madewithlove.com/blog/my-blog-posts-are-100-ai-generated/ - Published: 2026-11-12T19:03:00.000Z - Updated: 2026-11-12T19:03:01.000Z - Description: The final sentences are close to 100% AI-generated. The experiences, opinions, angles and decisions about what to publish are not. A walkthrough of a writing process built on interrogation, rejection and arguing over single lines, and why counting generated words says little about authorship. - Author: Brenden Cambier - Tags: AI, LLMs, AI development, Ideas, Communication, AI Agents, Claude, #ai-fast-track I'm coming clean. I use AI for my blog posts. Not just for grammar corrections. Not just to polish a sentence here and there. Pretty much every blog post you've read from me was written by an LLM. 100% AI-generated. 0% AI opinion. And I suspect most people don't realise how much AI is involved. Or how little of the actual thinking comes from it. So naturally, I asked AI to check all my blog post creation conversations. These are the five points it found: 1. **I rarely ask AI for an opinion.** I bring experiences, observations, ideas, and conclusions. Sometimes I know exactly what I want to say. Sometimes I have absolutely no idea how to say it. 2. **I reject a surprising amount of what AI writes.** Not because it's badly written, but because it's not what I think. Or because it sounds like a motivational speaker having a particularly productive day on LinkedIn. 3. **I use AI to discover the story, not invent the experience.** The events happened. The lessons are mine. Finding the interesting narrative is where AI helps. 4. **I'm ridiculously particular about my voice.** Dry humour, sarcasm, pacing, unnecessary drama. Apparently, I spend an unhealthy amount of time telling AI that I don't talk like that. 5. **I retain editorial authority.** AI can suggest ten directions. I'll reject nine, combine two, and occasionally decide the whole premise was wrong. Well, that's reassuring. But how does that actually work? ## It starts with a brain dump Most of my articles start with something that happened. Something I built. Something that went wrong. Something I learned. Occasionally, something that annoyed me enough to write 2,000 words about it. I dump everything into ChatGPT (yes, not Claude. I like GPT more for writing. Personal bias). The context, what happened, why I find it interesting, what surprised me, and what I think the lesson might be. Sometimes it's surprisingly structured. Sometimes it's an incoherent wall of text that would probably concern a psychologist. But I don't immediately ask AI to write the article. I ask it to help me figure out whether there's an interesting story or learning in there. ## Then comes the interrogation I explicitly ask AI to question me before drafting. When I was writing [*Three Claudes Walk Into a Codebase*](https://madewithlove.com/blog/three-claudes-walk-into-a-codebase/), GPT thought it was Sherlock. It started with eight groups of questions. I answered them. It came back with eight more targeted questions. And those questions matter. That article started with my experience comparing AI coding tools. Cursor had disappointed me. Claude Code had impressed me. It could easily have become another boring comparison of AI tools. But through the questions and answers, the interesting part changed. This wasn't just about which model was better at writing code. It was about developers becoming tech leads, directing agents instead of doing all the implementation themselves. That's the kind of shift I'm looking for. **Not AI telling me what to think. AI helping me discover what I was actually trying to say.** And it's not always right. Sometimes I reject its proposed lesson entirely. Which brings us to the next step. ## Finding the story Once there's an angle, we start working on the narrative. Headings first. What belongs where? What's the opening? What should the reader discover along the way? Where does the lesson land? This can take quite a few iterations. For my Junction diary, I wanted to describe what happened when I started introducing AI to a non-technical investment team. AI initially tried to turn the experience into an explicit lesson about how AI changes the way we teach people. I didn't want that. I wanted readers to discover the irony themselves: I'd spent [an entire blog post](https://madewithlove.com/blog/ai-chat-is-a-brain-in-a-jar/) arguing that AI should be more than a chatbot, only to build a chatbot to teach people that AI should be more than a chatbot. That was the story. The lesson didn't need a neon sign pointing at it. This is also where the conversation can change my own understanding of an experience. Sometimes I start with one conclusion and end with a more interesting one. But I have to recognise it as my conclusion. Otherwise it's just an LLM producing something that sounds profound. And LLMs are extremely good at sounding profound. ## Then we write. And argue. Once I like the structure, we start drafting. Usually section by section. AI suggests. I read. I adapt. I give feedback. It adapts my adaptations. I adapt those again. Sometimes the changes are substantial. Sometimes we spend an embarrassing amount of time debating one sentence. Another example: *Hacking the Human Psyche with AI*. I insisted we remove speculation about people's motivations and psychological labels. Observing how someone communicates is not the same as knowing why they behave that way. AI could generate all the beautiful prose it wanted. That ethical boundary wasn't negotiable. These aren't grammar corrections. They're decisions about accuracy, meaning, and what I'm willing to put my name under. ## Then we argue about the arguing Eventually, I have something that resembles a finished article. And I feed it back into AI. Review the pacing. Find contradictions. Cut repetition. Challenge the conclusion. Compare the tone with my other articles. Then I review its review. For [*I Miss a Good Stick*](https://madewithlove.com/blog/i-miss-a-good-stick/), I caught a joke about needing an agent to tell me to stop working and watch a movie. Great joke. Except the reader hadn't been introduced to the movie yet. So we moved it. Later, we debated the final two lines: "Find your stick. I'll go find mine." Or: "Find your stick. I've now found mine." AI preferred the first because the second implied I'd already solved the problem. I agreed. That tiny difference changes the entire emotional conclusion of the article. And once all of that is done, I usually have one final request. Stop putting every fucking sentence on its own line. That was past Brenden. I evolved. It hasn't caught up yet. And I haven't made a skill for it yet. I've literally instructed AI not to change a single word, just fix the paragraphs while preserving the important punchlines. Apparently, my AI has a serious addiction to imposing past Brenden's dramatic pauses. ## But sometimes I just write There are rare exceptions. *I Miss a Good Stick* and *Three Claudes Walk Into a Codebase* were largely born in the middle of the night. I woke up with the story already in my head and had to write it down. No interrogation. No searching for a narrative. Just getting the words out before they disappeared. AI still helped me polish, restructure, challenge, and refine them afterwards. Those are probably my favourite writing experiences. Unfortunately, inspiration doesn't always appear when I want it to. ## So who wrote this? Today I had to answer a company culture survey asking what percentage of my blog posts were written by AI. It's a surprisingly difficult question. If we're counting who generates the final sentences, the answer is close to 100%. If we're counting who experienced the events, formed the opinions, made the decisions, challenged the conclusions, and decided what was worth publishing, the answer is me. Of course, that doesn't make every AI-generated article authentic. You can ask AI to invent an opinion and publish it under your name. You can also write every word yourself while repeating somebody else's ideas. **I just don't think counting generated words tells you much about authorship.** I didn't necessarily type every word you've just read. But I mean every one of them. And yes, this article was written using exactly the process it describes. Including the part where I told AI its first version wasn't good enough.