The pattern, itemised
It is worth being specific, because the pattern is remarkably consistent. Open ten AI company sites in ten tabs and you will find most of these on most of them:
- A near-black background with a violet-to-cyan gradient mesh behind the hero.
- Inter, or a geometric sans that is trying to be Inter without paying for it.
- A headline of the form “The AI-native platform for X”: where X is a category the reader has to already understand.
- An abstract node-and-edge diagram representing nothing in particular.
- A logo bar of customers, greyed to 40% opacity, several of whom are pilots.
- Three feature cards with thin-line icons, each describing a capability rather than an outcome.
- A subtle animated grid or particle field that costs 400ms of load time.
- A footer promising you can “start building in minutes”.
Why it happened
Convergence like this is never the result of laziness alone. Three forces produced it.
First, the product is invisible. A model has no shape. Faced with representing something with no physical form, everyone reaches for the same small set of available abstractions: networks, nodes, light, particles, and arrives at the same place independently. It is not plagiarism, it is a shortage of metaphors.
Second, the reference set is tiny. Founders benchmark against the four companies they admire, all of whom benchmark against each other. The reference pool for “what a serious AI company looks like” is smaller than the reference pool for almost any other category, and it is self-reinforcing every quarter.
Third, and most decisively: looking like the category signals membership of it. When a category is new and buyers are unsure who is legitimate, resembling the established players is a genuine trust strategy. It works, right up to the point where everyone does it, and resemblance stops carrying any information at all.
What it costs
The cost is not aesthetic. It is measurable in three places.
Recall. A buyer evaluating six vendors over three weeks needs to be able to tell them apart from memory. If your site is visually interchangeable with the other five, you are relying entirely on the sales conversation to do work that the brand should have done for free, and you are the vendor whose name they cannot remember on the call where the shortlist gets made.
Hiring. Senior engineers and designers read a company's website as evidence of how it thinks. A site assembled from category defaults reads as a company that has not made any decisions yet. You are competing for those people against companies with more money; visible conviction is one of the few advantages you can actually manufacture.
Pricing power. Undifferentiated presentation invites undifferentiated comparison, and undifferentiated comparison collapses to price. If nothing about how you present yourself suggests you are a different kind of thing, the procurement conversation will be about the number.
What to do instead
The way out is not to be weird for its own sake. Deliberate strangeness applied to a serious enterprise product reads as immaturity, and regulated buyers will simply not proceed. The way out is to build the identity from what is specifically true about your product rather than what is generally true about your category.
Ask what your buyer is actually purchasing. A compliance director buying a regulation rule engine is not buying intelligence: they are buying auditability, structure and the ability to defend a decision to a regulator eighteen months from now. That is a completely different visual brief from “powerful AI”, and it points somewhere the category defaults never will: toward precision, structure, and calm rather than glow and motion.
Then commit to the specifics. A real typeface with a point of view rather than the default. A colour system derived from something true about the business rather than from what tested well on a competitor. Imagery that shows the actual work: real interfaces, real outputs, real people, rather than an abstraction of it. Language that describes the change you produce rather than the technology that produces it.
None of this requires a bigger budget than the category default. It requires making decisions and then not undoing them the first time somebody says it does not look like the company they admire.
The test
There is a simple diagnostic. Take your homepage, remove the logo and the product name, and put it in front of someone who knows your market. If they cannot tell you which company it belongs to, the identity is doing no work, and everything you spend on driving traffic to it is being spent at a discount.