How to Name a Brand That Can Actually Be Trademarked

Drew Mercer·March 18, 2026·10 min read

You have a product. You have a pitch deck. You have a domain name. What you might not have is a brand name that the USPTO will actually register.

Nearly 48.3% of all trademark applications filed with the United States Patent and Trademark Office are rejected. The success rate has fallen from 59.1% to 51.7% over the past five years. And the single most common reason for refusal? The name describes what the product does instead of identifying who makes it.

This is not an abstract legal problem. OpenAI cannot trademark "GPT." Google is being sued over "Gemini." xAI has 28 prior filings blocking "Grok." Microsoft quietly withdrew its trademark application for "Copilot." These are companies with billion-dollar legal budgets — and they still got the naming wrong.

The good news: the legal framework for choosing a strong name is well-established, intuitive, and free to use. It is called the Abercrombie spectrum, and once you understand it, you will never look at a brand name the same way.

The Abercrombie Spectrum: A Legal Cheat Sheet

In 1976, the Second Circuit decided Abercrombie & Fitch Co. v. Hunting World, Inc. and established the framework that trademark law still uses today. Every brand name falls somewhere on a five-tier spectrum of distinctiveness — and where it lands determines whether it can be protected.

Category	Definition	Protectable?	Example
Generic	The common name for the product or service itself	Never	"Email" for email, "GPT" for language models
Descriptive	Directly describes a feature, quality, or characteristic	Only with proof of secondary meaning	"ChatGPT" for a GPT-based chatbot, "SmartLock" for smart locks
Suggestive	Hints at a quality but requires imagination to connect	Yes — inherently distinctive	"Netflix" (internet + flicks), "Midjourney"
Arbitrary	A real word used in an unrelated context	Yes — strong protection	"Apple" for computers, "Claude" for AI
Fanciful	An invented word with no prior meaning	Yes — strongest protection	"Kodak," "Xerox," "Anthropic"

The spectrum moves from weakest (generic) to strongest (fanciful). The critical dividing line sits between descriptive and suggestive — everything above that line is inherently distinctive and eligible for trademark protection. Everything at or below it faces an uphill battle.

What the Data Shows: Where Applications Fail

The two most common substantive grounds for USPTO refusal tell the story clearly.

Most Common USPTO Trademark Refusal Grounds (% of Applications)

Section 2(d) — likelihood of confusion with an existing mark — hits roughly 1 in 5 applications. This is the "someone already has that name" problem, and it catches founders who skipped a clearance search.

Section 2(e)(1) — mere descriptiveness — catches names that simply describe what the product does. "ChatGPT" for a GPT-powered chatbot. "SmartScan" for a scanning tool. "AI Writer" for an AI writing assistant. These names might be clear to customers, but clarity is the problem: the USPTO will not grant one company exclusive rights to a term that competitors need to describe their own products.

The AI Naming Graveyard: Five Cautionary Tales

The AI industry has produced the most instructive trademark failures in recent memory. Each one maps directly to a different flaw on the Abercrombie spectrum.

GPT — Generic

OpenAI filed to trademark "GPT" and was refused. The USPTO determined that Generative Pre-trained Transformer is a generic term describing an entire category of language models — not a brand. Hundreds of third-party products already used "GPT" in their names. You cannot trademark the name of the thing itself.

ChatGPT — Descriptive

The ChatGPT trademark application received a final refusal for being merely descriptive: it tells consumers exactly what the product is (a chat interface using GPT). OpenAI has appealed to the Trademark Trial and Appeal Board, arguing acquired distinctiveness, but the initial refusal stands as a textbook example of the descriptiveness trap.

Gemini — Prior Art

Google renamed its AI chatbot from Bard to Gemini in February 2024. The problem: Gemini Data, Inc. had been using the "Gemini" trademark since 2011 for AI and data analytics software. The USPTO refused Google's applications based on likelihood of confusion. Gemini Data subsequently sued Google for trademark infringement. Even an arbitrary name fails if someone else got there first.

Grok — Crowded Space

xAI launched its chatbot under the name "Grok" — a term borrowed from Robert Heinlein's science fiction that means to understand intuitively. Distinctive in theory. In practice, the USPTO identified 28 prior filings too similar to allow registration, including "Groq" (an AI chip company), "Grok Health," and "Grokstream." xAI operates without a federal trademark registration to this day.

Copilot — Withdrawn

Microsoft filed to trademark "CoPilot" and then filed an Express Cancellation of its own application — an unusual move for a company with one of the largest IP legal teams in the world. Today, Microsoft uses the Copilot name extensively but never claims trademark rights in it. No TM symbol. No registered mark. The name is functionally unprotectable.

What Good Naming Looks Like

Now look at the names that worked — and notice the pattern.

Claude (Anthropic) — An arbitrary mark. "Claude" is a common first name with zero connection to artificial intelligence. It is reportedly a reference to information theorist Claude Shannon. The USPTO registered it without issue. It sits comfortably in the "arbitrary" tier.

Midjourney — A suggestive mark. It hints at a creative process (a journey in the middle of creation) but requires imagination to connect it to image generation. The name suggests without describing.

Anthropic — A fanciful/suggestive mark. Derived from "anthropic" (relating to human existence), it is distinctive enough to function as a brand while subtly evoking the company's focus on human-aligned AI.

Apple — The canonical arbitrary mark. A common fruit, entirely unrelated to computers. Decades later, it remains one of the strongest trademarks in the world precisely because it describes nothing about the product.

Nike — Named after the Greek goddess of victory. Arbitrary in context, evocative in meaning, and impossible to confuse with a description of shoes.

The pattern: the best brand names describe nothing about the product and everything about the brand.

The .ai Domain Trap

One more naming hazard worth flagging: the .ai domain explosion has created a false sense of security around descriptive names.

.ai Domain Registrations Growth (Thousands)

From 60,000 in 2022 to over 551,000 by January 2025 — a roughly 10x increase driven by the AI boom. But here is the trap: domain availability has nothing to do with trademark availability. You can register smartscan.ai today and still have your trademark application refused tomorrow for descriptiveness. The domain gives you a URL. The trademark gives you legal protection. They are separate systems.

The most dangerous pattern we see is founders combining a descriptive term with "AI" — DataClean AI, SmartWrite AI, CodeHelper AI — and assuming the domain registration means the brand is protectable. It does not.

Name Strength Scorecard

Before you commit to a name, score it on these five dimensions. Each is worth 0 to 4 points. A score of 15 or above suggests a strong candidate; below 10 signals serious risk.

Criterion	0 (Weak)	2 (Moderate)	4 (Strong)
Distinctiveness	Generic or descriptive of your product	Suggestive — requires imagination	Arbitrary or fanciful — no connection to product
Clearance	Identical marks exist in your class	Similar marks exist but in different classes	No conflicting marks found in any relevant class
Linguistic flexibility	Unpronounceable or offensive in key markets	Works in English only	Pronounceable and neutral across target markets
Domain and handle availability	Neither .com nor key TLD available	One major TLD or social handle available	.com (or primary TLD) and social handles available
Scalability	Name ties you to one product or geography	Name works for current product line	Name works across future products, markets, and categories

Maximum score: 20. A name like "Claude" scores roughly 18 (arbitrary, cleared, globally pronounceable, domain secured, scales beyond any single product). A name like "ChatGPT" scores roughly 6 (descriptive, clearance issues with "GPT," scales poorly because it anchors to one technology).

The 5-Step Naming Process

Step 1: Generate Candidates on the Right Side of the Spectrum

Start with suggestive, arbitrary, and fanciful candidates only. Use metaphors, mythology, foreign languages, portmanteaus, and invented words. If your first instinct is a name that describes the product, set it aside — it belongs in your tagline, not your trademark.

Aim for at least 15-20 candidates before filtering.

Step 2: Run a Knockout Search

Search the USPTO's Trademark Electronic Search System (TESS) for each candidate. Look for identical and phonetically similar marks in your Nice classification classes. This free search takes five minutes per name and eliminates the obvious conflicts.

Also search Google, LinkedIn, and the major app stores. Common-law trademark rights exist even without registration.

Step 3: Score Against the Scorecard

Apply the Name Strength Scorecard above. Drop anything below 12 points. Your shortlist should be 3-5 names.

Step 4: Get a Professional Clearance Search

A comprehensive clearance search from a trademark attorney or professional search firm costs $500-1,500 per name and covers federal registrations, state registrations, common-law use, domain names, and international databases. Compare that to the $10,000-$180,000 cost of a forced rebrand, and the math is straightforward.

Step 5: File an Intent-to-Use Application

Once your attorney clears the name, file an Intent-to-Use (ITU) application with the USPTO to establish priority while you finalize your launch. This locks in your filing date — critical in a first-to-file race — for $350 per class.

Choosing a brand name is a creative act. Protecting it is a legal one. The founders who treat naming as both — who generate creative candidates and then validate them against the legal framework — are the ones whose brands survive contact with the USPTO.

The Abercrombie spectrum is not a constraint on creativity. It is a filter for it. The best names in business history (Apple, Nike, Kodak, Claude) are the ones that describe nothing about the product and become inseparable from it. That is the goal: a name so distinctive that it can only mean you.

Sources: