AI Brand Naming: A Studio Workflow With Domain and Trademark Checks

A studio can use AI to build the naming routes and generate a few hundred raw names, then screen them in three passes: scoring against the brief, domain and handle checks, and a trademark knockout search that a person runs on the real registers. Show the client six to eight survivors with their results, and send the final two to a lawyer.

Free name generators stop at the first step. They produce lists, and sometimes a green tick for a domain, but a name that cannot be registered, said down a phone line or kept clear of a competitor is worthless however good it sounds. The workflow below follows a family kitchen fitter launching a premium design-and-fit brand. The funnel runs from 190 raw names to two sent for legal clearance, and every figure is illustrative.

Follow me on Instagram@sagnikteaches

A naming brief that keeps 190 names from becoming 190 opinions

The brief is one page, filled in with the client on a 45-minute call. An assistant can turn your call notes into the first draft, but the client reads and approves it, because every later argument about a name goes back to this page. If the studio has already run a positioning exercise or a brand audit, its one-sentence positioning goes straight in. If not, the positioning and messaging workflow comes first.

Connect on LinkedInSagnik Bhattacharya

NAMING BRIEF (illustrative)
Client: family kitchen-fitting firm, 11 staff, launching a premium design-and-fit brand
Audience: homeowners spending $15,000 to $40,000 on a fitted kitchen
Positioning sentence: One team designs, builds and fits your kitchen, and stays until it is right.
Personality: warm, exact, unfussy
Markets and languages: English-speaking markets now; Spanish likely as a second language within five years
Draft classes: installation services (Nice class 37); the lawyer confirms any others
Domain policy: exact .com, or one recognised alternative the client accepts in writing
Handles needed: Instagram, Facebook, YouTube
Avoid: under 4 letters; puns on cooking; anything that needs explaining; names that resemble a rival's
Decision-makers: the owner and her daughter; choose by [date]

Three fields do the heavy lifting. Markets and languages set which registers you search and which languages you test for unfortunate meanings. The draft classes matter because trademark searches run by class: the Nice Classification, run by the world intellectual property body WIPO, sorts goods and services into 45 numbered classes (1 to 34 for goods, 35 to 45 for services), and a trademark application names the classes it covers. Class 37's heading begins "Construction services; installation and repair services", which is where a kitchen fitter's core work sits. The domain policy settles in advance whether only the exact .com will do. Most clients have not thought about any of the three, and each changes the screening.

Subscribe on YouTube@sagnikteaches

Five routes, then three rounds of generation

A route is a rule for how names are made, and it stops the assistant sliding into one style. For the kitchen fitter the studio chose five: craft and material words, home and gathering words, coined words of two or three syllables, a heritage route built on the family surname, and plain descriptive names. Each route goes into the assistant as its own instruction with its own examples of what to avoid.

Context: [paste the approved naming brief].
Task: generate 24 brand names for ROUTE 2, "home and gathering words".
Rules: 4 to 10 letters per word; at most two words; no names that need spelling out on a phone call; no hyphens; no made-up endings such as -ly, -ora or -ify; nothing that resembles a well-known kitchen retailer.
For each name give: the name, one line on the idea, and how you would say it phonetically.

1. Hearthwell - warmth plus a well you draw from. HART-wel
2. Tablehouse - where the family gathers. TAY-bul-howss
3. Gatherwell - a kitchen built for company. GATH-er-wel
4. Longtable - plain and welcoming. LONG-tay-bul
(4 of 24 shown. These names are invented for illustration; any of them may already be in use.)

That output is typical, and so are its faults. Assistants often drift towards the same families of invented words, they often reach for compound names that join two nouns, and they sometimes ignore a length rule halfway down a list. The studio therefore asks for 24 names per route in a first round, then runs a second round in which the team picks eight favourites and asks for names "shorter than these and less obviously two words". A third round combines fragments from the best survivors. Once duplicates and rule-breakers are struck out, the three rounds leave 120 names. A free generator such as Namelix, which describes itself as a free AI-powered naming tool, adds a further 40 names, and the team's own list adds 30. All 190 go into one spreadsheet before any of them is judged, so a name from the generator meets exactly the same screens as a name from the studio.

Run one cheap check on the assistant's contribution. Put the same brief into three fresh chats and note which names recur. Names that keep coming back are names other studios and founders are being offered too, which makes them weaker candidates than their sound suggests.

Pass one: scoring names against the brief, not against taste

The first screen is quick and happens in a spreadsheet. Each name gets 1 to 5 on five criteria, and two hard-fail rules sit on top. The point of writing the criteria down is that nobody on the studio side can quietly promote a name they simply like.

The hard fails are an unfortunate meaning in any language named in the brief, and a name that needs spelling twice on a phone call. The pass mark is 18 out of 25. Here is how six names scored, shown by label rather than by name so that no invented word is mistaken for a verdict on a real business.

Name F shows why the hard fails exist. It reached the pass mark on points and was still cut, because the owner would have been spelling it to every caller. Name D, the descriptive option, shows the quiet weakness of that route: a name that only describes the service scores 1 for distinctiveness, and a trademark has to distinguish one business's services from another's, which a purely descriptive name struggles to do. Across the whole pool, 56 of the 190 names cleared this pass.

The assistant helps with one part of pass one: meanings and sound-alikes. For each name it lists words it resembles in the languages named in the brief and anything an ordinary speaker might hear in it. Treat the answer as a list of things to check, never as a finding.

For each of these 56 names, list: (1) real words or names it sounds like, (2) any meaning it has in Spanish, (3) any rude, medical or negative association in English.
Mark each claim "certain" or "unsure". Where you are unsure, say so rather than guessing.

On the kitchen job the assistant flagged two coined names as resembling words in other languages. One flag was right and removed a name. The other was a confident translation that a native speaker, asked for five minutes, said was wrong. Models can state a wrong translation in the same assured tone as a right one, so any meaning that would kill a name gets confirmed by a person who speaks the language before it does.

Pass two: domain and handle checks in an order that saves time

This pass is manual and takes about an hour and a half for 56 names, because each check is a lookup, not a judgement. An assistant cannot do it for you. It has no reliable view of what is registered today, and a confident "this domain is probably free" is a guess. Do the checks in this order, because each one removes names before the next one costs you time.

1. Exact match on the agreed extension. Search the name at any registrar. Record one of four states: free, registered and parked, registered and for sale, or registered with a live site.
2. What is on the live site. Open it. A live business in the client's own field using the name on any extension is a trademark-style problem, not a domain problem, so mark the name red at once.
3. Who owns a parked domain. Registration lookups now run on RDAP, the protocol ICANN created to replace WHOIS, which returns WHOIS-like registration data. Owner details may be withheld, so you may see little more than the registrar and some dates, which can still hint at whether a broker or a long-term holder owns the name.
4. Near misses. Check the plural, a hyphenated version and the most likely misspelling. A confusable domain held by a competitor will send the client's visitors to the wrong place.
5. History. For any previously registered domain, look it up in the Internet Archive's Wayback Machine. A domain that once hosted spam or something unsuitable can carry that baggage into email deliverability and search.
6. Handles. Check the handles the brief lists. A dormant handle is a minor annoyance; an active account in the same trade is another red.

The matrix below shows six of the 56 names that reached this pass, again by label. Twenty-six names survived the pass overall.

Notice what the matrix does that a generator's tick cannot. Name H shows a free .com and is still dead, because someone in the same trade already trades under the name on a social platform. Name I has a taken .com but the owner is in an unrelated trade, so it survives as an amber if the client's domain policy allows an alternative. Record the date next to every lookup, because a free domain on Monday can be gone by Friday.

Pass three: a trademark knockout search in an afternoon

A knockout search is a fast first pass that removes names with obvious conflicts. It is not clearance, and your proposal should say so in those words. Trademark rights are generally granted market by market: WIPO's own guidance says that to protect a mark abroad you file with the trademark office of each country where you want protection, or use its Madrid System, a single procedure covering several territories. Either way, each office decides under its own law, so the brief's list of markets decides which registers you search. Do it in this order.

1. Search the exact name in WIPO's Global Brand Database, which gives access to international registrations under the Madrid system and trademarks from participating national and regional offices, and can be searched by name, number, goods and services and image similarity. Then repeat the search in the official search tool of each office covering a market on the brief.
2. Search variants. Conflicts in many places turn on whether people would be likely to confuse two businesses, so a near match in sound or spelling, used for similar services, can be as much of a problem as an identical name. This is the one step where the assistant earns its place: it can list the variants you would otherwise forget.
3. Search the open web for businesses using the name or a variant in the client's field. In some systems a business can acquire rights simply by trading under a name, usually within the area where it trades, so a clean register does not make a name free.
4. Log every hit with its owner, class or goods, status (live or dead) and how close it is, then give the name a colour.

Name: [Name A]. Business: kitchen design and installation.
List 20 spellings and sound-alikes of this name that someone could mistake for it when hearing it or reading it fast: letter swaps, double letters, plural or possessive forms, vowel changes, common prefixes and suffixes.
Group them by type. Do not tell me whether any are registered; I will search for them myself.

The last line of that prompt matters. Left alone, an assistant will volunteer an opinion on availability, and that opinion is the most dangerous output in the whole workflow. Use the variant list as a search checklist, and nothing more.

That is a sample of the log. Across the pool, 11 of the 26 names that reached this pass survived it, nine going out on identical live hits in a related field and six on near misses with live owners. The work took about three and a half hours.

When an assistant calls a name unique

The mistake that sinks naming projects is not a bad name. It is a bad name that arrived with a reassuring sentence. On the kitchen job, an assistant asked to rank the survivors described name B as "distinctive and unlikely to conflict with existing brands". The knockout search then found an identical live registration for installation services. The assistant was not lying in any ordinary sense. It produced a plausible sentence, and it had not searched a register.

Build the habit into the process. Strip every availability opinion out of assistant output before it reaches a spreadsheet, and let a name earn each colour only from a lookup that a person ran and dated. The commercial terms of the AI vendors add a second reason for caution. Anthropic's commercial terms, for example, exclude trademark claims based on use of an output in trade or commerce from the indemnity it offers customers, so a vendor will not stand behind a generated name. Read the terms of whichever vendor you use, because they differ and change. For a deeper look at what automated trademark searching finds and misses, see AI trademark search for small IP firms.

Presenting survivors so the client chooses names, not risks

Seven names reached the presentation: the eleven survivors, cut to the strongest few in each of three routes. Before the meeting, read all seven aloud once to five people outside the studio and ask them to write each one down; the results go on the name's page. A client who sees fifteen names will argue about taste; a client who sees seven with their results beside them will argue about fit. Never show a name you have not screened, however good it is, because the client falls for it in the first minute and the lawyer's "no" then costs you the relationship.

1. One slide recapping the brief, including the positioning sentence the client approved.
2. One slide explaining the three routes with one example each.
3. One page per name, in the layout below.
4. A comparison grid: names down the side, criteria and check results across, as colours.
5. The studio's recommendation: a top three, each with a sentence on why.
6. A slide that says what has and has not been checked.
7. Next steps: the client ranks the top three alone, the lawyer clears the winner, and the domain and handles are registered the same day it clears.

The "what has and has not been checked" slide needs careful wording. Something like this works as a starting point, though your own lawyer should settle the final text.

We have run a first-pass search of the domains, social handles and trademark registers named in the brief. This is a screening step, not legal clearance. Before you adopt any name, ask a trademark lawyer to clear it for the goods, services and markets you will use it in.

Ask the client to rank their top three privately before any group discussion, and collect the rankings before anyone speaks. In a group discussion the loudest voice can set the answer, and a private ranking shows you whether the owner and her daughter actually disagree. If they do, that disagreement is a finding about the brief, and it is cheaper to discover now than after the logo is drawn.

Hours, scope and what the studio promises

Naming is easy to underprice because the visible part, the list of names, looks quick. The invisible part is the screening. These are planning figures for the kitchen fitter job. Your own timesheets will tell you more after two or three projects.

The saving is real but narrower than most people expect, and it sits in the generation stage. Everything that decides whether a name is usable costs the same with or without an assistant. Costed at an illustrative $75 an hour, the job costs the studio $1,050 with AI and $1,650 without, a difference of $600. A studio that treats the saving as a reason to skip screening has saved nothing: one name pulled after the logo is drawn costs more than the whole naming stage.

Write the promise into the proposal as plainly as the process. The studio delivers a brief, up to seven screened names per round, the results of its screening and a recommendation. It does not promise that any name can be registered, and the client's lawyer clears the final choice. A client who has been burnt by a rebrand will look for that last sentence first. If a client is naming their own business without a studio, point them to the owner's version of this shortlist and check, and for what comes next, what designers can promise about AI-generated logos covers the ownership questions logos raise.

Questions studios ask about AI naming and clearance

Can I trust the availability tick on a free name generator?

No. At best that tick says a domain looked free at that moment. It says nothing about trademarks, social handles or a live business already using the name in the client's field. Treat the generator as one source of raw names that then goes through the same three screens as every other name.

How many names should we show the client?

Six to eight, drawn from two or three routes, each with its check results beside it. Fewer than five feels thin, and more than eight tends to stall the decision because the client starts comparing details that do not matter. Ask the client to rank their top three alone before the group discussion.

Do we still need a lawyer if the knockout search is clean?

Yes, for any name the client will spend real money on. A knockout search is a fast first pass that removes obvious conflicts. A lawyer judges similarity, the right classes and the markets, and can see risks a name search misses. Say this in the scope and on the presentation slide.

Should we buy the domain before the presentation?

Check availability before, buy after. A domain gives no trademark rights, so paying for one on a name that later fails clearance wastes money. Once the client has chosen and the lawyer has cleared the name, register the domain and handles the same day.

Further reads

• AI for Branding Agencies: Where It Fits in a Brand Project — See where naming sits among research, strategy and identity work.
• Best AI for Brand Identity: Logos, Palettes and What Still Needs a Designer — What AI can do for logos and palettes once the name is chosen.
• Food Truck Name Generator AI: Prompts, Trademark and Domain Checks — A small-business version of the same checks, for a single trader.
• Plumbing Business Name Ideas: Shortlist With AI, Then Check Before Printing — How a trade business shortlists names before anything is printed.
• Design Studios Using AI: Briefs, Moodboards and Client Revisions — A studio-side workflow for moving from a messy brand brief to approved visual directions and controlled revision rounds.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: WIPO's pages on trademarks, the Madrid System, the Global Brand Database and the Nice Classification (NCL 13-2026), ICANN's RDAP page, a national trademark office's pages on use-based rights and likelihood of confusion, Anthropic's commercial terms and the Namelix site, all read in November 2026.