- A model compares words. A trademark examiner compares sound, appearance and meaning, and then asks whether your goods are related to someone else's.
- Identical names legally coexist across unrelated categories, which is why a name being in use somewhere is not by itself a reason to abandon it.
- The UK statutory test refuses a similar mark on similar goods where confusion is likely, and that includes the likelihood of association.
- UK registration starts at 205 pounds and takes roughly three to four months, with your application published so anyone can oppose it.
- Unregistered rights exist and appear in no register, so a clean search result is a narrower reassurance than it feels like.
- Print nothing until the opposition window has passed. The packaging is what makes a name change expensive rather than annoying.
The name arrives first. Somebody thinks of it in a car, checks that the domain is free, checks that the handle is free, and by the following week it is on a logo, a label, a shipping box and 2,000 printed pouches with a minimum order quantity that made them worth printing. The check that never happened is the one about whether anybody else already owns it for the thing you sell.
Asking an assistant to look into it feels like the modern version of that check. It is worth understanding exactly what that question can and cannot produce, because the answer it gives is quick and confident while being addressed to a different question than the one that matters.
What is a model actually doing when it checks a name?
Comparing strings and recalling what it has seen. Neither of those is the test, and the gap between them is where the money is lost.
The real test is set out plainly by the trademark office. The USPTO's explanation of likelihood of confusion puts it in two parts. First, similarity of the marks, and the marks need not be identical: they can be confusingly similar in sound, appearance or meaning, or create a similar commercial impression. Second, whether the goods or services are related, which covers goods that are identical, similar or competitive, used together, bought by the same purchasers, advertised together, or sold by the same dealer.
Read those two parts and the mismatch with a text comparison becomes obvious. Sound is not spelling. The office's own example is T. Markey against Tee Marquee, two strings with little in common on the page and almost nothing separating them when spoken. Meaning is not spelling either: a name translated into another language, or a synonym, can create the same commercial impression while sharing no characters at all.
Why can two businesses use the same name?
Because a trademark is a right over a name for particular goods or services, not over a word. This is the part that surprises people most, and it cuts both ways: it is why your name might be available despite an existing use, and why yours will not stop somebody in another field.
The office gives the examples directly: Dove soap and Dove ice cream, Delta faucets and Delta airlines. Same word, unrelated categories, both registered. What it also gives is the harder direction, where relatedness is not obvious to a layperson. Hot dog buns and hamburger buns are related. Pressure washing and lawn mowing are related. Neither pair is the same product, and in both cases an examiner would treat a shared name as a problem.
The machinery behind this is classification. The Nice Classification, established by the Nice Agreement in 1957, is the international classification of goods and services used for registering marks, and every application sits in one or more of its classes. Classes are administrative, not decisive: two goods in different classes can still be related, and two goods in the same class can be unrelated. A model asked whether a name is free will not have chosen your classes, and the classes are the shape of the right you are trying to obtain.
| The check | Who can do it | What it catches | What it misses |
|---|---|---|---|
| Domain and handle availability | You, in two minutes | Nothing legal at all | Every registered right that exists |
| Exact name search in a register | You, free | Identical marks in your classes | Similar sounds, similar meanings, similar spellings |
| Asking an assistant | You, instantly | Well known conflicts it happens to recall | Anything registered recently, anything local, anything unregistered |
| Variant and phonetic search | You, with a list of variants | Most near misses on sound and spelling | Relatedness judgements and unregistered use |
| Clearance search by an attorney | A professional, for a fee | Registered and common law rights, with an opinion | Nothing much, which is the point of paying |
The first row is there because it is the check most small businesses actually perform. A free domain proves that nobody registered that domain. It is evidence about a hosting database and about nothing else.
What does the law actually test?
Identity or similarity of the mark, combined with identity or similarity of the goods, and then whether confusion is likely. The UK statute is unusually readable on this point.
Section 5 of the Trade Marks Act 1994 refuses registration outright where a mark is identical to an earlier mark for identical goods. Where either element is merely similar rather than identical, registration is refused if there exists a likelihood of confusion on the part of the public, and the section adds that this includes the likelihood of association with the earlier mark.
That last clause is worth pausing on, because it is broader than most sellers assume. Association is not the same as mistake. A consumer who does not think your product is the other company's, but does think the two businesses are connected, licensed or related, is inside the test. A name that reads as a tribute, a variant, or an obvious cousin of an established brand fails on association even where nobody would literally confuse the products.
What does registration cost and how long does it take?
In the UK, at least 205 pounds and roughly three to four months, and the number that matters more than either is the publication step in between.
The official guidance is specific: the office checks that your mark is not the same as or similar to existing marks, the cost starts at 205 pounds and rises with the number of classes you cover, processing takes around three to four months, and your application is published in the journal in case anybody wants to oppose it. Objections have to be resolved before registration proceeds.
The publication step is the reason the timeline is not a formality you can ignore while printing. Your application becomes public, and the businesses most likely to care are the ones who monitor for new applications in their classes. An opposition is not an accusation you can shrug off: it is a process, it costs money to defend, and the cheapest outcome is frequently to change the name. If the packaging already exists at that point, the cost of the opposition is the printing bill rather than the legal one.
The cost that hurts is never the application fee. It is the artwork, the printed packaging, the labels, the signage, the stock already produced, and the marketplace listings and reviews attached to a name you have to abandon. Sequence the spending so that the reversible items come first.
What can you do yourself in an hour?
More than most people do, and it will not be a clearance opinion. The aim is to find the obvious problems cheaply so that you only pay a professional about names that survive.
Start by writing your goods description in plain terms before you search anything. Not your marketing category, the actual goods: what you sell, to whom, through what channel. Everything downstream depends on this, because relatedness is judged against it. A search run without a settled description is a search against a guess.
Then build a variant list rather than a name. Take your candidate and generate the spellings a person might use, the plural, the possessive, the version with the vowel changed, the version with a K for a C, the compound written as two words, and the homophone. This is one job where a model is genuinely good, because generating plausible variants is a language task with no legal content. Ask for twenty variants including phonetic equivalents, then search each one.
Search each variant in the official register for every market you sell into, and search the classes that cover your goods plus the adjacent ones. A candle maker should look at classes covering candles and also at cosmetics and home fragrance, because those are where a relatedness argument comes from.
Finally, search the open market rather than the register. A business trading under a name without registering it may still have rights, and those rights appear in no database. Search the marketplaces you sell on, the app stores if relevant, and ordinary web results for the name plus your product word. An active unregistered user in your category is a reason to stop even when the register is clean.
Where does AI earn its place in this?
In three narrow jobs, none of which produce a verdict. Generating candidates, generating variants to search, and drafting the goods description in the language classification uses.
Candidate generation is the obvious one and the least risky, because a name that turns out to be unavailable costs nothing at the idea stage. Ask for thirty candidates with a stated constraint, then run your own screen, and expect most of them to fail. A shortlist of three names that survive a register search is worth more than one name you fell in love with.
Variant generation is the highest value job, because it attacks the exact weakness of a manual search. People search the name they chose. Examiners consider the name a competitor might reasonably claim is too close. Variants and phonetic equivalents close most of that gap for nothing.
Drafting the goods description is the third and the most technical. Classification language is stylised, and a plain English description of your business needs translating into terms that map onto classes. A model can propose that wording and explain which class each term sits in, which turns a blank form into something you can review. Review it you must, because an over broad description invites objections and an over narrow one leaves your actual products unprotected.
What you should never accept is the conclusion. A model asked whether a name is safe will answer, because the question is well formed and the pattern is familiar, and its training data has no reliable picture of recent filings, local registers, or anything unregistered. The same structural failure applies wherever a confident answer stands in for a missing lookup, as in why language models produce confident wrong answers.
Does the logo carry its own problem?
Yes, and it is a separate question from the name with a separate answer. A word mark and a figurative mark protect different things, and generated artwork raises an ownership question the name does not.
The practical split is simple enough. Register the word if the word is your brand, because a word mark travels across every visual treatment you will use over the next decade. Register the logo as well only if the visual identity is itself distinctive and worth defending. A small business with limited budget almost always gets more protection per pound from the word.
The ownership question sits underneath that, and it matters before you file. Who owns a logo produced by a generative tool is not settled in the same way as one drawn by a designer under contract, and the answer affects what you are able to assign or license later. We went through that in detail in what you actually own when a logo was generated, and the short version is that the trademark question and the copyright question have different answers.
What happens if you find a conflict after launch?
You get a letter, usually, and the speed of your response decides the cost. The worst reaction is to ignore it and keep selling, because continued use after notice changes the character of the dispute.
Three outcomes are common and they are not equally bad. A coexistence arrangement, where both businesses continue in defined categories or territories, is the good case and is more available than people expect when the categories genuinely differ. A rebrand is the middling case, and its cost is entirely a function of how much printed and published material carries the name. A dispute you fight is the expensive case, and it is rarely worth it for a business at this scale unless the name is the asset.
The related scenario is the reverse one, where somebody starts using your name. That is a monitoring and enforcement problem rather than a clearance one, and the tools and the tactics are different, which is why it sits in how a small brand gets a copy taken down rather than here.
Does any of this apply if you only sell on marketplaces?
It applies more, not less, because the marketplace gives a rights holder a faster route to act against you than a court does. A complaint can remove your listings before anybody establishes who is right.
The mechanism is worth understanding before it happens. Marketplace intellectual property programmes operate on notice. A rights holder submits a complaint naming your listing, the platform removes or suppresses it, and you then have to respond to restore it. The platform is not adjudicating the underlying question of whether your mark infringes theirs, and it has no obligation to wait while you argue. For a business whose entire revenue arrives through one channel, that sequence is more dangerous than a legal letter, because it interrupts trading immediately.
Two consequences follow. A registered mark of your own is worth more to a marketplace seller than to a purely direct seller, because registration is what most platform brand programmes require before they will act for you or listen to you properly. And an unresolved similarity to an active brand in your category is a standing risk to your listings rather than a theoretical legal exposure, which changes the calculation about whether to proceed with a borderline name.
The suspension mechanics on the other side of that, where your account rather than a listing is affected, run on the same notice and response pattern, and we set that out in what you are actually owed when a seller account is suspended.
How many classes should a small seller file in?
The ones covering what you sell now, plus the one covering what you will plausibly sell within the registration's life, and no more. Each additional class costs money and widens the surface where somebody can object.
The temptation runs both ways and both errors are real. Filing too narrowly protects a description that no longer matches your business two years later, which is common for makers who start with one product line and expand. Filing too broadly costs more, invites objections from more directions, and in some jurisdictions leaves the registration vulnerable where you never used the mark for the goods you claimed.
A reasonable working rule for a small business is to file in the class covering the goods you sell today, add the class covering the obvious adjacent product you have already discussed making, and add a retail services class only if you sell other people's products alongside your own. Anything beyond that is speculative protection paid for in advance, and the money is usually better spent on registering in a second country where you actually ship.
Which country to add is a commercial question rather than a legal one. Register where you sell in volume, where you manufacture, and where counterfeits of products like yours typically originate. A right that exists only in your home market is of limited use against a problem that arrives from somewhere else.
The sequencing that saves the money
Everything in this article reduces to one rule about order. Spend on reversible things first, and on irreversible things last.
A domain is cheap and reversible. A website is work but changeable. Listings can be edited. An application fee is small. Printed packaging, custom tooling, embroidered labels, branded shipping supplies and physical signage are none of those, and they are usually ordered early because they have long lead times and volume discounts. That is the trap: the items with the worst cancellation economics are the ones a launch schedule wants to order first.
So build the launch backwards from the publication window. File, build the digital side while the application is pending, and hold the print order until the opposition period is behind you. If a lead time genuinely forces an earlier print run, print the minimum rather than the economic quantity and treat the difference as insurance. And if you are standing up the storefront around the name in the meantime, do it in a way that lets you change the brand in one place rather than forty, which is one of the quieter arguments for building the shop from a description you can revise rather than hand editing a theme.
One honest limit. Nothing here is legal advice and none of it substitutes for a clearance search where the stakes justify one. What it does is tell you which names to stop considering before you pay anybody, which is most of the value a small business can extract from this process on its own.