In 2024 I was building a side project that would turn SEO research into strategic content plans automatically, and I never finished it or launched it. I kept running into one thought I could not argue my way past: if the tool did its job mathematically perfectly, everyone in the industry who used it would end up with the same answer. I just didn't really believe in it.
That thought is now the ordinary condition of anyone using AI for marketing work. The tools solved the problem they were sold to solve, the cost of producing words, and in solving it they handed every business two new problems. Everyone's output starts to sound the same, and a mistake no longer stays where it was made.
Problem one: everyone gets the same answer
In July 2024 two researchers, Anil Doshi and Oliver Hauser, published an experiment in Science Advances that measured the first problem directly. They had 293 people write short stories, some of them with story ideas supplied by a large language model, and then had 600 separate readers rate the results. The stories written with AI ideas were rated more creative, better written and more enjoyable, and the biggest gains went to the writers who had been least creative to begin with. Those stories were also more similar to each other than the stories people wrote alone. Each writer came out ahead, while the collection as a whole got narrower.
Swap the short stories for product pages, ad copy and the monthly blog post, and you have the Collapse of Differentiation running through a whole category: every business that feeds a similar brief into a similar model gets a better draft than it would have written alone, and a draft that sits closer to its competitor's draft than before. The tool that was supposed to give you an edge gave everyone the same edge, which is no edge at all.
Long before AI, I held that capitalism runs on combative competition and that the end result is markets that monopolise or become commodities. A category where every business publishes the same improved average is walking toward the commodity end. In a commodity market the customer chooses on price, because nothing else is left to choose on.
Problem two: the mistake travels
In May 2024 Google's new AI Overviews answered some searches with advice to put glue on pizza to keep the cheese on, and to eat rocks. Liz Reid, Google's vice president of Search, explained afterwards that the glue had come from a discussion forum and the rocks from a satirical article that had been republished on a geological software provider's website. Google made more than a dozen technical improvements in response. A joke in one place had become an answer on Google Search.
A person who writes something wrong puts it in one place. A system that drafts from a bad input puts it in every place it drafts, at the speed of the machine, in the same fluent prose as everything true around it, and that is the Scaling of Errors. A business running AI across its website, its emails, its ads and its sales material has built the same machine at a smaller size, and one wrong figure in the brief becomes the wrong figure everywhere the brief reached.
Both problems share a root. The model is not the source of what makes a business different, and it is not the source of what makes a business right. Feed it the average and you get the average back, polished. Feed it a mistake and you get the mistake back, multiplied.
Another tool will not fix it
The instinct is to answer a tool problem with a better tool, a sharper prompt library, a newer model, an AI marketing assistant that promises the edge the last one did not deliver. Every one of those is available to your competitor on the same afternoon, which puts you straight back inside problem one.
Producing more does not fix it either. I don't like posting at volume, and I would rather build a smaller number of permanent pages that get better over time, because each of them keeps working instead of needing to be fed. Play the game if you can win it. Otherwise go back to work and do what you do best, which is better for the world than another website, another reel or another podcast appearance.
An operating order, not a tactic
What fixes both problems is the order the work happens in. The loop inside the 4-Quadrant Framework is that order, four steps with a rule about who leads each one. In Q1 a human decides what is true and what the business believes, and writes it down. In Q2 the machine executes from that, the drafts, the variations and the analysis, at whatever volume is useful. In Q3 a human checks every output against what was written down before anything goes live. In Q4 the system deploys, and what the market teaches goes back to Q1. Who leads what gives each step an owner.
The order answers each problem at the point where it starts. The collapse starts in Q1, because a business that skips it hands the machine nothing but the average, and the machine hands the average back. The scaling starts at Q3, because a business that skips the check lets the first wrong output become the template for every output after it.
My own notes on the four quadrants put the whole idea in one line: expert input closer to consumer output. The machine's job is to shorten the distance between what a business knows and what its customers read, not to replace the knowing.
The input is the only part you own
What a business writes down in Q1 is its System Seed, the expert knowledge, the position, the voice, the verified facts and the customer logic, held in one place where the machine reads from it in Q2 and the validator checks against it in Q3. The model is rented, and so are the prompt libraries and the assistants built on top of it. The seed is the one part a competitor cannot rent this afternoon, because it can only be built by doing the work. The System Seed is your only moat goes through what goes into one.
None of this alarms me. I think AI will eventually end the advertising economy as we know it, and that could be a good thing. The tools did not remove the hard part of marketing, though. They moved it upstream, to the two questions only a person inside the business can answer: what do we actually know, and is it true?