There is a particular kind of website you have started seeing everywhere. Dark background. A headline in a geometric sans. Three feature cards with thin line icons. A gradient that fades from violet to blue. Copy that says something like “built for teams who move fast.”
You have seen it because it works. You have also seen it because everyone is now generating it from the same handful of tools, trained on the same corpus of what already existed.
This is worth paying attention to, because it changes where competitive advantage actually lives.
The models are trained on the average
A generative model does not invent a style. It learns the distribution of everything it was shown and then produces something near the middle of it. That is not a flaw. It is the entire mechanism. When you ask for a “modern minimal landing page for a B2B SaaS company,” you are not asking for an original idea. You are asking for the statistical center of every modern minimal B2B SaaS landing page that has ever been indexed.
The output is competent. It is usually better than what a small business would have produced on its own with no designer and no budget. That is a real gain and it is worth saying plainly.
But competent and average are the same coordinate. If your competitor down the street prompts the same tool with the same brief, they land on the same coordinate. You have both bought the median, and the median cannot differentiate you from someone standing on it too.
Production has stopped being the bottleneck
For most of the history of marketing, the constraint was making things. Copy took time. Design took time. A landing page took a developer. Because production was expensive, being able to produce at all was a genuine advantage. Agencies sold capacity.
That constraint is dissolving. A founder can now generate a month of social posts before lunch. The cost of producing marketing material is heading toward the cost of the electricity to run the prompt.
When a constraint disappears, the advantage it protected disappears with it. Nobody wins on the ability to produce a blog post anymore. Everyone can produce a blog post. What is scarce is knowing which post is worth producing, and having something to say in it that the model could not have averaged into existence.
The bottleneck moved from making to deciding.
What actually stays scarce
A few things do not come out of a model, because a model has no access to them.
Your actual position. Not your category, your position. A model can tell you that project management software should emphasize collaboration. It cannot tell you that your real advantage is that you are the only vendor in your space who will do migrations by hand for free, and that this should be the loudest thing on your homepage. That comes from knowing your business.
Specific proof. Generic trust language is now free and therefore worthless. “Trusted by industry leaders” costs nothing to write and signals nothing. A named customer, a specific number you can defend, a real screenshot of the thing working, a guarantee you will actually honor: these cannot be generated, because they have to be true.
Taste. Taste is the accumulated judgment of what to cut. It is the thing that says the gradient is doing nothing, the third feature card is filler, and the headline is one clause too long. Models are trained to produce, and producing more is what they are good at. Deciding to produce less is a human act.
A voice with a point of view. Averages have no opinions. A brand that will not say anything a competitor could not also say has no voice, only vocabulary. Having a position means some readers will disagree, which is precisely why an averaging process will never generate one.
This is an opening, not a threat
The useful way to read all of this is not defensively.
If the baseline of marketing output is rising toward a uniform competence, then anything with a genuine point of view stands out more than it used to, not less. Sameness is a background against which difference is easier to see. When every site on a category’s first page looks the same, the one that looks like a decision gets remembered.
That is a better environment for a small, specific business than the one we just left. You do not need a large budget to have a position. You need to have made choices.
What to do about it
A few things worth doing this quarter, in order.
Write down what you will not say. Most positioning work fails because it only adds. Take your homepage and identify every claim that a competitor could copy verbatim without it becoming untrue. Those claims are not doing work. Cut them and see what is left. What is left is your actual position, and it is usually shorter and sharper than what you had.
Audit your proof. Separate every trust signal on your site into things that are specific and verifiable, and things that are atmosphere. Keep the first category. Delete the second, or go earn a replacement. One real, checkable detail outperforms five pieces of confident-sounding filler.
Use the tools for volume, not for judgment. Generative tools are genuinely excellent at drafts, variations, summaries, and first passes. They are poor at deciding what matters, because deciding what matters requires a stake in the outcome. Let them do the production. Keep the deciding.
Put a human where the difference is. Your point of view, your proof, your voice, and your visual identity are the surfaces where averageness costs you the most. Those are the places to spend real attention. Everything downstream of them can be assisted.
The short version
The tools are going to keep getting better at producing the expected thing. That is what they are for. As they do, producing the expected thing keeps getting cheaper and keeps getting less valuable.
The work that holds its value is the work that requires knowing something specific and being willing to choose. Deciding what you actually are. Deciding what to leave out. Deciding what you will put your name on.
That was always the real work. It is just getting harder to hide from it.