Don't Get Too Excited About AI-Generated Names

A while back, a friend in the food business came around to see me. He was buzzing. He opened a tool and more than twenty brand names tumbled across the screen — each one with a logo to match.

He said: help me pick one. Isn't AI naming brilliant?

I glanced at the screen and asked: these names — have you run a trademark search on any of them?

The light in his eyes dimmed half a beat.

That's what I want to talk about today. More and more teams, design studios, and outsourced shops are using generative AI to come up with brand names, logos, and taglines — because it's fast, because it's cheap. Fair enough. The trouble is, nearly everyone sees the "fast" and never sees the flip side that "fast" carries with it.

What "Collision" Means

Let me get the term straight first.

A collision is when the name you want to use already sits in the market looking a little too much like another already-registered trademark. The day that name lands, consumers stop and wonder: are these two the same company?

The kicker is that a collision is almost never intentional.

How does AI name things? It doesn't conjure names from nothing — during training it swallowed hundreds of millions of real brand names, logos, and taglines. Ask it to name a new tea-drink brand, and what it hands back is probably that whole family: "Tea In," "Tea Fun," "Tea Room," "Tea Roots." Why? Because those are the words that show up with the highest probability in the industry corpus it was trained on.

See it now? It isn't creating — it's reproducing probability. You think you cracked open a lucky box and found a brand-new name; what you actually did was re-harvest a cleverness a few thousand companies had already used.

The trademark law doesn't care where the name came from. Human-made or AI-made, the rule is exactly the same: the only thing that matters is whether it confuses consumers. If it confuses people, that's a risk. Even if you had no ill intent at all.

One sentence: where trademarks are concerned, the law treats AI and humans the same.

The Second Trap, the Easiest to Fall Into: "I Already Had It Checked"

The second trap is sneakier than the first.

A lot of teams, after they settle on a name, feel perfectly settled. Why? "Didn't the AI check it for me?"

I've seen more than one team whose first reaction after naming wasn't "let's wait for legal" but "the AI says it's fine."

And that's exactly where the problem sits.

What an AI can "check" and what actually sits in the trademark database are two different things. Ask the AI "is this name already taken?" and it'll probably cough up a pile of look-alike names. But whether a trademark can be registered isn't only a question of whether it looks similar — it also comes down to: are the goods and services in the same class, is there an industry link, and would it genuinely muddy the waters if it hit the real market?

That whole chain of judgment, AI can't do.

Here's a better way to picture it. Asking AI to check a trademark is a bit like finding a restaurant through a review app. It can tell you how many restaurants are on your block and what they're rated — but whether the food actually suits your taste, you only learn by sitting down and ordering the whole spread.

What AI gives you is a clue, not a verdict. The real stamp of approval still has to come from an experienced professional.

And here's a bit of pencil math: one proper, formal trademark search by a reputable firm runs a few hundred yuan. Skip it, wait until the packaging is printed and the goods are on the shelf, and only then discover the name collides — that single mistake costs you enough to start three more companies.

Saving a few hundred to blow a few hundred thousand — anyone can do that arithmetic.

The Volume Doesn't Make It Sound

One more thing that's easy to overlook: volume.

A human designer can churn out thirty names in one evening, and that's considered fast. AI? Minutes, and three thousand.

A flood of options looks great — but in practice it breeds one flaw: the team tells itself "with this much filtering, it must be sound." And in reality? Not a single one has been formally vetted.

Keep this sentence close: however many names you have, if none has passed the check, that equals zero. What you lack isn't volume — it's that "checked, verified" step, the vetting.

So before a company goes to market, you'd better lock the pipeline down first:

AI produces a first draft → filter out the obviously unusable ones → keep two or three → formal search → legal review → and only then does the final decision get stamped.

It all comes down to this: vet first, release later. Iron rule.

You Didn't Use AI — You Still Don't Get a Pass

Then there's the scenario that feels most unfair: you never used AI at all, and you still wound up in trouble.

You never touched it. But the design studio you outsourced to did. They pulled your logo into an AI, generated it, handed it to you, you used it, and something went wrong. They shrug: that's AI's doing.

In the eye of the law, nobody cares whether a logo was drawn by AI or by a human. The name carries yours, the packaging prints yours, the shelves stock yours — so you're the one who takes the fall.

Too many small companies go into this almost naked: the contract never mentions AI, never says who owns what, never says who's responsible. The unspoken assumption is "he's my old friend, he won't do me wrong."

An old friend won't do you wrong. But he just might.

So when you sign with a design studio, a supplier, or a fulfillment agency, at minimum get these four words down in black and white:

One, transparency. Which parts of the deliverables used AI must be stated plainly and written into the contract.

Two, guarantee. The supplier guarantees in writing that this work doesn't infringe anyone's rights.

Three, ownership. Whose intellectual property is the logo and the design? Write it down from the day you sign — don't wait until you've used it up to find out it isn't yours.

Four, compensation. If a third party genuinely comes at you, who foots the bill? If the contract doesn't say, it becomes "we'll see" — and "we'll see" means you pay.

Transparency, guarantee, ownership, compensation — four locks.

Even if you can't afford a marquee law firm, having these four words seen by a friend who knows the law beats slapping your chest and telling the other side "old relationship, absolutely no problem."

Someone Else Can Rip Off Your Good Name, Too

Flip the mirror around and hold it up one more time.

The tool that made inspiration cheap for you made copying cheap for your rivals, too.

In the past, someone who wanted to imitate your brand would have to get a designer to redraw it, a half-day per piece. Now? Drop your logo into an AI, type "make me something like this," and in a few seconds the edge-riding knockoff is done. Then it shows up on small e-commerce shops, social accounts, and short-video clips, with a name a little like yours, selling things a little like yours.

At this point, you have to strike first rather than wait:

• Set a monitoring rhythm. Every month, at a fixed time, search your brand name, homophones, and pinyin across a few platforms — screenshot each instance you spot.
• Put trademark registration on the schedule. Many platforms' takedown channels only open to "registered trademarks." If you haven't registered, you don't even get the right to complain.
• The registration fee, to most founders, isn't the money. The hassle is the passivity of having "not registered."

Five Rules

Let me wrap it up and give you something you can take away in one sentence each:

1. Set an internal rule. Whoever uses AI to make brand materials has to disclose it first — no one wields their own knife.
2. Review before you release. Any name, logo, or copy produced by AI has to pass legal review before it counts as delivered.
3. Never skip the formal search. AI is a stand-in; for a true "search everything," you need a human.
4. In supplier contracts, install all four locks above — transparency, guarantee, ownership, compensation — and don't leave a single one out.
5. Keep monitoring. Every month, open the platforms and search your name — it doesn't even cost five yuan, so don't save it.

Back to the Beginning

In the end, my friend did listen to me and sent all twenty or so names out for a formal search.

When it was done, he called: lucky we checked. Several of them already had owners in the same category.

So you see, that's the whole thing in a nutshell:

AI made "thinking of a good name" cheap. It didn't make "keeping a good name" cheap.

It magnifies your inspiration - and it magnifies your carelessness. The name can be generated by AI. The risk, no one can take on for you.

One last question he asked: so from now, I'll have AI quickly generate names and send them out for third-party verification one at a time?

I said: that's right, that's the answer.

Let AI create for you. Don't let AI decide for you.

May your name never need to be changed.