In December 2023 I put a one-page website up on carrd and sent the link to a few friends, asking them to tell me whether the concept was starting to get explained. The site was called TopicPlot, beta, and its tagline was Strategic SEO Content Plans. I was living in Wanaka, working 20 hours a week PAYE for a manufacturer, with one ad client on the side, and the plan, as I typed it to a friend on New Year’s Eve, was that my other work provides me consistent safe money, then I can pump all idiosyncratic impulses into TopicPlot, and TopicPlot can cruise without pressure. Asked where it was going, I said sell 10,205 topicplots.
The product was one document. A business filled in an intake form, and what came back was a plot of 49 pages of content for them to create, each page with its focus keyword, its place in the hierarchy and the other pages it should link to, which is the kind of plan an agency would have charged around $10,000 for, and which I had built by hand for clients several times, in several ways. The site was up so I could show my friends, and behind it I was building the internal tools, an AI chatbot pulling live SERP data via API, a fixed interview form so that every run started from the same place, the whole workflow automated from the first question to the finished plot if I could get it right. The same month, I was typing to another friend that AI replaces SEO, that when you have a personal GPT you ask it, not Google, and that whatever information everyone else’s AI enjoys using is who gets the traffic. I was building a keyword tool for a search engine I had already decided was on its way out. By March 2024 I was calling it a SaaS hobby business, a creative outlet, no commitments, and I never launched it.
Two competitors in the same niche fill in the same form. The same keyword database returns the same search volumes for the same phrases, the same rules rank the same topics, the same logic draws the same hierarchy, and both of them walk away with the same 49 pages. If done mathematically perfectly, everyone in the industry would have the same answer. Every improvement I made to the tool made the outputs more alike, so the better the product got, the less it was worth to any one customer, because a plan your competitor also holds is not a plan, it’s a starting line. Keyword tools and persona tools have the same property, the same data through the same method produces the same personas, which is why they all read alike. Perhaps the tension is why I could never finish and launch it? I just didn’t really believe in it.
Google rankings are competitive, the top 10 show at the top 10. Doesn’t matter if 1,000 follow best practices, there will still only be 10. You have to be different and better, not good. Give all 1,000 the same writing machine and it will produce good, quickly and in volume, for every one of them at once, and there will still only be 10. It incentivises lazy.
If my ad needs a unique headline, I talk to the team about how we do things, I find something weird they do, it becomes the headline, it becomes a brand position. It is unique, it is true. The weird thing was never in any dataset because nobody wrote it down, it lives in the building, in the way this particular team does the job, and the only route to it is to be in the room and ask. In-house means that you can adapt your headline/offer by changing the product, it is more genuine, as an ad agency you only really work with 1 P ‘promotion’. I ran an agency for a while and sat on the wrong side of that line, working the one P while the products I wrote about stayed exactly as they were, and the machine that writes has an agency’s view of the world: promotion only, product untouchable, input whatever it was handed.
A video project at work made the same point from the other end. A colleague had worked up a concept and refined it with ChatGPT, a second person was coming in to film it and direct the shots, and the finished video would be published into a competitive marketplace. What we had was essentially a one-brain idea. In that marketplace it would be up against videos that are five brain ideas, as in five different people collaborated on the core concept and then they passed it to different specialists, a scriptwriter, a director, a marketer, a subject expert. I said I could guarantee a better outcome, not because of the script or because of the filmer but because of the process: put collaboration early in the process and then let experts do their piece, everyone playing to their strengths. The machine had been used as the second brain, and it isn’t one. It is the average of every brain it was trained on, and an average is precisely what the five-brain competitors were not.
On 26 August 2026 I dictated a memo at work about why we could not choose a CRM. The problem with choosing a CRM provider at the moment is that the whole SaaS market is in a huge state of change. Everyone is adding AI into their products and their roadmap is changing frequently, and many companies are trying to pivot. The SaaS segment of the share market has been absolutely hammered, which will lead to acquisitions and mergers. CRM companies such as HubSpot are also looking to bundle their data so that they can make money as a data provider, and when a company gets sold or acquired the customer data gets sold along with it, and as far as I understand the new company can change the data usage rights. None of that justifies building a CRM ourselves, but it does make it difficult to choose and trust and be able to predict a company as a supplier. Every vendor we looked at was adding the same AI to the same features, so the products were converging the way my topic plots had, and the one thing that stayed different from vendor to vendor, our own customer data, was the thing being packaged for sale.
And then we have another wave of disruption arriving, which is the in-house boom. As AI lets more in-house people produce code and make programs, the theory is that this will lead to a boom of in-house productivity software, built by the people who hold the data and know the weird thing about how their team works, for nobody but themselves. The tool-makers converge on one product and the buyers walk in-house, which is TopicPlot again with the sign turned around: a perfect tool is worth the same to everyone, so it is worth nothing to anyone, and the value moves to whoever holds the input.
This collection is the input side of that. The words are mine, and an AI reconstructed the essays from them. Every draft was measured against my own corpus, the dictations, the plans, the messages, before it reached me, and then I read it and sent it back. A note I typed to myself, in capitals, said EXPERT INPUT CLOSER TO CONSUMER OUTPUT. It’s my ideas, for people to consume however they want it, in whatever language they want it, via AI. It sits behind one door, free and ungated, with no funnel behind it.
I think it will end the financialisation and advertising economy eventually, which could be a good thing. When 1,000 businesses can produce the same good content for nothing, attention stops being worth farming, because the farm grows the same crop in every field, and most of what pays for the internet stops paying. I don’t find myself mourning it, and I also can’t see what takes its place. Would that be a loss?
