Two things happened to marketing at once, and they point in the same direction.
A client did something in a meeting a few months ago that I have since started doing myself, in advance, before every first session.
He opened his laptop, asked a chatbot what his company did, and read the answer out to the room.
It was not wrong, exactly. That was the problem. It was a flat, reasonable, faintly generic description of a business in his sector, with one detail in it that had not been true for two years, and none of the things he would have led with. It described a smaller, older, blander version of his company, in a confident tone, in about ninety words.
Nobody said anything for a moment. Then his marketing lead said the thing everybody was thinking, which was: where did it get that?
It got it from you
That is the uncomfortable answer. It got it from everything they had published, taken all together, with no idea which parts they meant most.
It could not know that the investor deck was the sharp one. It could not know the website was two years behind, or that the founder's own posts were the truest thing the company owned. It read the lot and produced an average.
Which is a genuinely new failure mode, and worth being precise about.
An inconsistent brand used to have a contained cost. It confused the person who happened to visit three of your channels, and most people never visited three. Now that inconsistency gets compressed into a summary and repeated, confidently, to people who will never visit any of them. They will read ninety words about you, form a view, and decide whether you are worth a conversation, and you will never know they were looking.
Coherence used to be about a human moving between your touchpoints. It is now also about what a machine concludes from reading all of them at once. Same discipline as before, considerably higher stakes, and much less forgiving of drift.
The other thing that happened
There is a second shift running alongside it, and the two are related in a way that is easy to miss.
Making things stopped being hard. Video, images, copy, voice. The floor came up so fast that a competent piece of content is now close to free.
When production was expensive, execution was a genuine signal. Good photography meant something, because bad photography was normal and good cost money. A well made video was evidence of investment, and investment was evidence that someone was taking this seriously.
That signal is gone. Everyone can make a competent thing now, so competence has stopped proving anything about anybody.
Which leaves the decision. What should exist, for whom, saying what, and what should not exist at all. That was always the valuable part. It was hidden behind the cost of production and got billed as though it came free with the making.
And there is a corollary in there that I think is the actual headline. If output is nearly free, output is no longer evidence of commitment. Publishing a great deal now says almost nothing about how seriously a company takes itself. What says something is what it chose not to do.
Both shifts, from opposite directions, land in the same place. Judgment is the scarce input, and consistency is what makes judgment legible.
What we found when we did this to ourselves
I would rather not write this as advice we have not taken, so here is what happened when we turned it on our own material this year.
We rewrote our case studies with the claims stated plainly instead of implied, mostly as an experiment. What it exposed was how much of our writing had been quietly relying on a reader who already liked us. Warm sentences carrying tone and almost no information. Confident phrasing that dissolved the second you asked what it was actually asserting.
The clearest example was a line on our old site saying the work had significantly boosted visibility and client engagement.
Read quickly by a friendly human, that sentence is fine. Read by anything that is not trying to be generous, it says nothing at all. It has a number behind it or it comes off the page.
It came off.
How do you change what AI says about you?
You cannot control what a machine says about you. You can remove the contradictions it is averaging, which in practice comes down to four habits.
Say what you are in language that survives being repeated by someone who is not trying to be kind. If a sentence only works when read generously, it will not survive the trip.
Make the same claim everywhere, in the same words. Reading all of you should produce one answer rather than a blend. Two different names for the same idea is how a company becomes uncitable, and most companies have three.
Put a number next to anything that sounds impressive. Unverifiable superlatives are the first thing a machine discards and the first thing a buyer discounts, so they were never doing the work you thought.
And cut the sentences that only hold up because the reader is already on side. Most website copy is written for someone who has decided to like you. Very little of your audience has.
None of this is new work. It is clarity and coherence, which were worth doing when the argument was purely about humans. What has changed is the price of ignoring them. It used to be a slow leak of trust that nobody ever quite noticed. It is now ninety words about your company, written by something that cannot be talked around over coffee, handed to people who will never give you the chance to explain.
