In the seventh season of Happy Days, Fonzie put on water skis, kept his leather jacket over his swim trunks, and jumped a shark tank. The phrase came from that. What most people using it do not know is that the episode was a hit. Roughly thirty million people watched it, and the show ran six more seasons after that night. The show was finished and the ratings were the best they had been, so no one felt pressure to look any further.
Everyone remembers the stunt as the marker of decline, but nobody said so at the time. The phrase was coined years after the episode aired, by people who already knew how the show ended and worked backward to find the spot where it turned.
The talk right now is about AI and content. Both LinkedIn and Substack have moved on detection. The prevailing read is that the platforms want better writing, that quality collapsed and they stepped in to fix it. That is not what happened. Users complained, in volume, about a feed that had become unreadable, and the platforms did what platforms do, which is give people what they say they want in a form that keeps them on the site longer. Detection is a product feature. It arrived because the channel stopped functioning, and a channel that stops functioning stops producing engagement.
The feature created a sport. Readers now hunt. They look for the tells, they find them, and they say so publicly, and the shaming produces more time on platform than the original content ever did.
Whatever anyone intended, what exists now is a system that manufactures suspicion and rewards accusation, and the piece of it that matters most commercially is that the detector reads only the sentence structure. A differentiated thought and an empty one get flagged the same way. So if you are scanning before you read, understand what you are doing. You are throwing out the baby with the bathwater and you will never know which one you had.
This puts idea-led firms in an uncomfortable position. Why?
Content was never the commercial system. It was evidence that a commercial system existed. Somebody was in the field, meeting clients, taking positions and getting corrected by the market. The writing was the residue of that. It carried weight because production cost something. You could not fake having done the work, because faking it took longer than doing it.
Firms confused the residue for the engine. They built publishing calendars and treated the calendar as the growth choice, and it worked for a while because the proxy still correlated with the work. When production cost went to zero the correlation broke, and every firm that had been running the proxy found out it had no engine. That is the actual event of the last two years and it has almost nothing to do with prose quality.
A client told me recently that he could not tell my writing was AI generated, and that in a recent piece I had been too hard on myself about using the tool. He meant it kindly and I took it that way. It settles nothing. He is a client, he likes me, so the signal there is not strong enough for me to pat myself on the back, but it is still nice. What his comment does tell me is narrower and still useful. A reader who knows these tools went looking for the seam and did not find one. There was nothing to find, because the argument was mine before any of it reached the tool.
Here is where I part company with most of what is being written.
Regression to the mean is not a defect in these systems. It is the function. The tools return the center of the distribution because the center is what they were built from, and asking them to do otherwise is asking them to be something they are not. Every instruction you give is local. The pull is constant. I have banned the same words repeatedly and I still cut them out of drafts months later, and I have stopped experiencing that as a failure of the tool. It is the tool working as designed, against a distribution that includes everything average anyone ever published.
The consequence is that the output is a direct read on what you brought. Bring nothing and you get the average of everyone, rendered fluently, which is exactly why so much of it is unreadable in the same way. It is not badly written. It is written well and about nothing, and there are ten thousand versions of it. Bring a real argument, a position you can defend, terminology that means something specific, and the tools become very good at execution against structure at speed.
So this is an accelerant. Good work becomes great. Average work disappears into the middle, permanently, because the middle is now infinite. Poor work is noticeably worse than it was, because bad thinking used to hide behind bad writing and now the writing is clean enough that everyone can see straight through to the argument.
None of that is free. It takes work, and it takes the same work it always took. What people wanted from these tools was permission to skip the thinking, which is the only scarce input in the first place.
There are two checks on your work.
The internal one is a question you can answer without asking anyone. Are you choosing a growth path built on your own skill and your own position, or are you going through the motions. You know. There is no version of this where you have done the work and are unsure whether you did.
The external one is not analytics. Impressions and likes and engagement rates are the Fonzie metric, and thirty million people watched that episode. What counts is consequence. A prospect reaches out and says they read the piece and want to talk. Someone repeats your terminology back to you, using your words to describe their own situation, which means the argument got inside their thinking and stayed. And then, further out and slower, a client uses your writing as the justification for a strategy they are actually implementing, spending money against it, restructuring something because of it. The last one is not gamed by a platform because it happens in real life.
I am as creative as I have ever been. The ideas are mine and the core writing is mine. What I handed over was the monotonous part, the construction of my thoughts. There are great contractors that never hammer a nail. There are also contractors that throw cookie cutter developments together and call it a day.
The move to AI detection is a tell, but not what you may think it is. AI is not a get out of jail free card that helps you fill the content calendar. Idea-led firms thrive on ideas. They need to be good ones.
If someone you know is scanning for the tells instead of reading the argument, send them this.
Tell me which check you trust more, the internal one or the consequence. I read every reply.
And if you have a publishing calendar but no engine underneath it, come find me.
Bossey is a growth studio. We build the commercial structure of your firm with you, the system that turns work arriving by luck into revenue you can count on, and then we hand it over so the firm can sell and grow without you carrying every deal. A studio builds it with you and leaves it in your hands.