OpenAI OpCo, LLC

THIS OPINION IS NOT A
PRECEDENT OF THE TTAB
Mailed: March 4, 2026
UNITED STATES PATENT AND TRADEMARK OFFICE
_____
Trademark Trial and Appeal Board
_____
In re OpenAI OpCo, LLC
_____
Serial No. 97733261
_____
Rebecca McCurry, Katie McKnight, and Christopher Weimer
of Pirkey Barber PLLC, for OpenAI OpCo, LLC.
Kelly Burke, Trademark Examining Attorney, Law Office 128,
Travis Wheatley, Managing Attorney.
_____
Before English, Elgin, and Lavache, Administrative Trademark Judges.
Opinion by Elgin, Administrative Trademark Judge:
Applicant OpenAI OpCo, LLC seeks registration on the Principal Register of the
mark CHATGPT, in standard characters, for the following goods and services:1
downloadable computer programs and downloadable computer software
for the artificial production of human speech and text; downloadable
computer programs and downloadable computer software for natural
language processing, generation, understanding and analysis;
downloadable computer programs and downloadable computer software
for machine-learning based language and speech processing software;
downloadable computer chatbot software for simulating conversations;
1 Ser. No. 97733261 was filed on Dec. 27, 2022 under Trademark Act Section 1(b), 15 U.S.C.
§ 1051(b), on the basis of intent to use the mark in commerce. An amendment to allege use
under Trademark Act Section 1(a), 15 U.S.C. § 1051(a), was filed on March 7, 2023, alleging
Nov. 30, 2022 as the dates of first use anywhere and in commerce in both classes.

Serial No. 97733261
- 2 -
downloadable computer programs and downloadable computer software
for creating and generating text, in International Class 9; and
providing online non-downloadable software for the artificial production
of human speech and text; providing online non-downloadable software
for natural language processing, generation, understanding and
analysis; providing online non-downloadable software for machine-
learning based language and speech processing software; providing
online non-downloadable chatbot software for simulating conversations;
providing online non-downloadable software for creating and generating
text; research and development services in the field of artificial
intelligence; research, design and development of computer programs
and software, in International Class 42.
The Trademark Examining Attorney refused registration of Applicant’s mark as
merely descriptive of the identified goods and services in both classes under
Trademark Act Section 2(e)(1), 15 U.S.C. § 1052(e)(1).
When the refusal was made final, Applicant appealed and requested
reconsideration of the refusal. In the alternative, Applicant claimed that CHATGPT
has acquired distinctiveness pursuant to Trademark Act Section 2(f), 15 U.S.C.
§ 1052(f). The Examining Attorney issued a Continuing Final Office Action in which
she maintained the mere descriptiveness refusal, but accepted the alternative claim
of acquired distinctiveness.2 The appeal resumed and is now fully briefed on the sole
issue of whether Applicant’s mark CHATGPT is inherently distinctive for the goods
and services identified in the application.
For the reasons set forth below, we affirm the refusal to register, and thus the
application will proceed under Section 2(f).
2 See 10 TTABVUE 2.

Serial No. 97733261
- 3 -
I. Preliminary Evidentiary Objections and Issues
The Examining Attorney has objected to four exhibits Applicant submitted for the
first time with its appeal brief:
• Exhibit A: A chart summarizing evidence relied upon by the Examining
Attorney during prosecution to show Applicant’s goods and services are
directed toward software developers, and copies of related webpages
printed on August 18, 2025;3
• Exhibit B: Screenshots of purportedly inaccessible websites and websites
displaying different information than those relied upon by the Examining
Attorney during prosecution;4
• Exhibit C: A Wikipedia page and news article relating to European privacy
and artificial intelligence regulations, intended to show that some foreign
users may have different experiences with Applicant’s goods and services
offered under the CHATGPT mark outside the United States;5 and
• Exhibit D: Screenshots printed on August 18, 2025 from certain websites
relied upon by the Examining Attorney during prosecution, to show the
owners are located outside the United States.6
We sustain the objection as to the new materials in Exhibits A-D. Only evidence
filed during examination is timely, Trademark Rule 2.142(d), 37 C.F.R. § 2.142(d),
and evidence should not be submitted on appeal.7 Applicant cannot rely on these
exhibits. We have, however, considered the three-page chart in Exhibit A8 only to the
3 8 TTABVUE 25-46.
4 Id. at 47-53.
5 Id. at 54-68.
6 Id. at 69-94.
7 “The proper procedure for an applicant or examining attorney to introduce evidence after
an appeal has been filed is to submit a written request with the Board to suspend the appeal
and remand the application for further examination.” In re Weiss, No. 88621608, 2024 TTAB
LEXIS 277, at *1 (citing Trademark Rule 2.142(d), 37 C.F.R. § 2.142(d)). That was not done
here.
8 8 TTABVUE 26-29.

Serial No. 97733261
- 4 -
extent that it comments on evidence in the record and would not, together with
Applicant’s brief, exceed the applicable twenty-five-page limitation. Trademark Rule
2.142(b)(2), 37 C.F.R. § 2.142(b)(2).
Applicant makes two objections to materials relied upon by the Examining
Attorney. First, Applicant argues that certain of the internet materials relied upon
by the Examiner are no longer available or are changed. However, the fact that
content on the internet is constantly in flux is the reason for our rule that, for such
content to be considered by the Board, a party must capture and submit a copy of the
specific matter to be considered together with the date accessed and the location
where the matter is found. In re Mueller Sports Med., Inc., No. 87209946, 2018 TTAB
LEXIS 156, at *7 (applying rule in Safer, Inc. v. OMS Invs., Inc., No. 91176445, 2010
TTAB LEXIS 51, at *22, to evidence submitted by examining attorneys in ex parte
cases and stating intention to hold applicants to the same standard). Because the
Examining Attorney did include the required information, we have considered the
evidence.
Applicant also objects to certain of the internet materials as being from foreign
websites.9 To the extent that websites appear to be English language foreign sources,
they nonetheless may be relevant to determine if a mark is merely descriptive. In re
Bayer AG, 488 F.3d 960, 969 (Fed. Cir. 2007) (“Information originating on foreign
websites or in foreign news publications that are accessible to the United States
public may be relevant to discern United States consumer impression of a proposed
9 Id. at 20.

Serial No. 97733261
- 5 -
mark.”) (citations omitted); In re Int’l Bus. Machs. Corp., No. 75581859, 2006 TTAB
LEXIS 483, at *15 n.7 (web page from an English-language website from another
country considered because case involved computer technology, and “it is reasonable
to consider a relevant article regarding computer hardware” from such a source). We
have considered this evidence.
Next, we address Applicant’s two expert reports. The first is a survey and expert
report by David T. Neal, Ph.D.;10 the second is an expert report by Phillip Carter,
Ph.D., a Professor of Linguistics.11 The captions of both reports reference a different
application serial number than at issue in this case, namely, Ser. No. 97733259 to
register the proposed mark GPT.12
Dr. Neal conducted a survey, in the “Teflon” format, of 213 software developers to
assess whether the term “GPT” is generic (i.e., a common name) or a brand name for
goods and services identified in application Ser. No. 97733259.13 “A ‘Teflon survey’ is
essentially a mini-course in the generic versus trademark distinction, followed by a
test.” Frito-Lay N. Am., Inc. v. Princeton Vanguard, LLC, No. 91195552, 2017 TTAB
10 Aug. 9, 2024 Request for Reconsideration at TSDR 28-93.
11 Id. at TSDR 94-218.
12 Id. at TSDR 29, 95. That application was not made of record. Dr. Carter reports that the
goods and services he considered were “downloadable computer programs and providing
software in non-downloadable form.” Id. at TSDR 97 (Carter Decl. ¶ 8). Dr. Neal appears to
have considered the term GPT for “web-based tools or services in the field of AI or machine
learning.” Id. at TSDR 46 (Neal Decl. ¶ 4.5).
13 Id. at TSDR 31-33 (Neal Decl. ¶¶ 1.7, 1.8, 2.2.3). Dr. Neal chose this universe because “The
relevant universe for a genericness survey is ‘those persons likely to rely on the challenged
mark in making purchase decisions.’ Accordingly, I screened respondents to ensure they were
software developers, and were either currently employed or self-employed full-time or were
currently freelance or contract-based part-time.” Id. at 37 (Neal Decl. ¶ 3.7).

Serial No. 97733261
- 6 -
LEXIS 300, at *38 (quoting J. Thomas McCarthy, 2 MCCARTHY ON TRADEMARKS AND
UNFAIR COMPETITION § 12:16 (4th ed. June 2017 update)). In this case, the survey
explained the distinction between a type of product or service, or common name (such
as gasoline) and the source of a product or service, or brand name (such as Exxon).14
Respondents were then asked questions to test their ability to distinguish brand
names from common names. Respondents then were given a list of terms and asked
whether they were brand or common names; one of these terms was “GPT.”15 Dr. Neal
concluded that his survey “clearly support[s] a conclusion that a majority of working
software developers in the United States understand GPT to be a brand name for
web-based tools or services in the field of AI or machine learning and not a generic
(or ‘common’) term.”16
A Teflon survey typically is used in the context of a genericness inquiry, and may
have some relevance in the determining acquired distinctiveness under Trademark
Act Section 2(f). See In re Country Music Assoc., No. 78906900, 2011 TTAB LEXIS
343, at *32 (“[A]lthough the consumer survey … was submitted in connection with
the issue of genericness, the acquired distinctiveness of the term COUNTRY MUSIC
ASSOCIATION among the relevant purchasing public can be inferred from the
result.”). But neither genericness of Applicant’s mark (or a portion thereof) nor the
issue of acquired distinctiveness is before us.
14 Id. at TSDR 43 (Neal Decl. ¶ 4.5).
15 Id. at TSDR 43-47 (Neal Decl. ¶ 4.5).
16 Id. at TSDR 49 (Neal Decl. ¶ 5.1).

Serial No. 97733261
- 7 -
The only question before us is whether the term CHATGPT used in connection
with the goods and services in the application is merely descriptive. Associating a
term with a brand does not preclude the possibility that the respondents think it
conveys information about a quality, characteristic, purpose, or function of the
relevant goods or services. Cf. Rise-N-Shine, LLC v. Duner-Fenter, No. 14-CV-1305,
2015 U.S. Dist. LEXIS 24601, at *7 n.1 (S.D.N.Y. Feb. 28, 2015) (“Specifically, just
because a mark is a brand name (and perceived as such) does not mean that it is
suggestive rather than descriptive”), cited in J. Thomas McCarthy, 2 MCCARTHY ON
TRADEMARKS AND UNFAIR COMPETITION § 11:16 and n.26 (5th ed. Mar. 2026 update)
(“Only rarely is survey evidence used to try to prove the distinction between
descriptiveness and suggestiveness.”)). That is, a merely descriptive term that has
acquired distinctiveness functions as a source identifier (i.e. a brand name).
Nonetheless, we have considered Dr. Neal’s genericness survey for whatever
probative value it may have.
Likewise, Dr. Carter opines on whether the term GPT is generic. He concludes
that “the primary significance of the term GPT is not as a generic term for a category
of software, but rather as a brand name associated with [Applicant].”17 For the
reasons explained, his research and opinions are not directly relevant to our analysis,
but again we have considered them for whatever probative value they might have.18
17 Id. at TSDR 129 (Carter Decl. ¶ 87).
18 The Examining Attorney objects to the webpages replied upon by Dr. Carter as missing
“the dates of capture or print, which is not sufficient to introduce the underlying webpages
into the record.” 10 TTABVUE 3. Applicant counters the objection. 11 TTABVUE 6 n.5. The

Serial No. 97733261
- 8 -
II. Mere Descriptiveness
A. Applicable Law
Section 2(e)(1) of the Trademark Act excludes from registration any “mark which,
(1) when used on or in connection with the goods [or services] of the applicant is
merely descriptive ... of them.” 15 U.S.C. § 1052(e)(1). A mark is “merely descriptive”
within the meaning of Section 2(e)(1) “if it immediately conveys information
concerning a feature, quality, or characteristic of the goods or services for which
registration is sought.” In re N.C. Lottery, 866 F.3d 1363, 1367 (Fed. Cir. 2017); see
also In re Chamber of Com. of the U.S., 675 F.3d 1297, 1300 (Fed. Cir. 2012). “A mark
need not immediately convey an idea of each and every specific feature of the goods
[or services] in order to be considered merely descriptive; it is enough if it describes
one significant attribute, function or property of the goods [or services].” In re Fat
Boys Water Sports LLC, No. 86490930, 2016 TTAB LEXIS 150, at *4 (citing In re
Gyulay, 820 F.2d 1216, 1218 (Fed. Cir. 1987)).
Descriptiveness is not considered in the abstract; it is analyzed in relation to an
applicant’s identified goods or services, “the context in which the [term] is being used,
and the possible significance that the term would have to the average purchaser of
disputed materials are screen captures from a search of the Google Books database, namely,
“all books published on or after June 11, 2018, that contained the search term GPT and (1) the
word ‘chat,’ (2) the name OpenAI, (3) the word ‘chat’ and the name OpenAI or (4) neither the
word ‘chat’ nor the name OpenAI.” Aug. 9, 2024 Request for Reconsideration at TSDR 119
(Carter Decl. ¶ 65). The date the search was performed is not included for purposes of self-
authentication, but the screenshots are authenticated by the testifying witness. See In re
Fantasia Distrib., Inc., No. 86185623, 2016 TTAB LEXIS 471, *21 n.13 (Board considered
still frame information about an online video provided in witness declaration). The objection
is overruled and we have considered the materials for whatever probative value they possess.

Serial No. 97733261
- 9 -
the goods [and services] because of the manner of its use or intended use.” In re
NextGen Mgmt., LLC, No. 88098031, 2023 TTAB LEXIS 1, at *6-8 (quoting Bayer,
488 F.3d at 963-64); see also Chamber of Com., 675 F.3d at 1300; Fat Boys, 2016 TTAB
LEXIS 150, at *4. We ask “whether someone who knows what the goods and services
are will understand the mark to convey information about them.” Real Foods Pty Ltd.
v. Frito-Lay N. Am., Inc., 906 F.3d 965, 974 (Fed. Cir. 2018) (quoting DuoProSS
Meditech Corp. v. Inviro Med. Devices, Ltd., 695 F.3d 1247, 1254 (Fed. Cir.
2012) (internal quotation omitted)).
A mark is suggestive, and not merely descriptive, if it requires imagination,
thought, and perception on the part of someone who knows what the goods or services
are to reach a conclusion about their nature from the mark. See, e.g., Fat Boys, 2016
TTAB LEXIS 150, at *11. Because we consider the context of the use, that a term
may have other meanings in different contexts is not controlling on the question of
descriptiveness. In re RiseSmart Inc., No. 85050089, 2012 TTAB LEXIS 448, at
*8 (citing In re Chopper Indus., 273140, 1984 TTAB LEXIS 118, at *5-6). “So long as
any one of the meanings of a word is descriptive, the word may be merely descriptive.”
In re IP Carrier Consulting Grp., No. 78542726, 2007 TTAB LEXIS 66, at
*19 (citations omitted).
We also must “consider the commercial impression of a mark as a whole … viewed
through the eyes of a consumer.” Real Foods, 906 F.3d at 974 (cleaned up, quoting
DuoProSS Meditech, 695 F.3d at 1253) (emphasis in original omitted). “Where a mark
consists of multiple words, the mere combination of descriptive words does not

Serial No. 97733261
- 10 -
necessarily create a non-descriptive word or phrase.” In re Zuma Array Ltd.,
No. 79288888, 2022 TTAB LEXIS 281, at *8 (quoting In re Omniome, Inc.,
No. 87661190, 2019 TTAB LEXIS 414, at *12). “[I]f each component retains its merely
descriptive significance in relation to the goods or services, the combination results
in a composite that is itself merely descriptive.” Omniome, 2019 TTAB LEXIS 414, at
*12. Only where the combination of merely descriptive terms creates a unitary mark,
or a mark with a unique, incongruous, or otherwise nondescriptive meaning in
relation to the goods or services is the mark registrable as suggestive. Id.
Refusals for mere descriptiveness, like other refusals, are based on evidence
bearing on public perception of the mark. See, e.g., Real Foods, 906 F.3d at
974 (descriptiveness inquiry “consider[s] the commercial impression of a mark as a
whole, viewed through the eyes of a consumer”) (cleaned up, citation and italics
omitted); In re Nett Designs, Inc., 236 F.3d 1339, 1341 (Fed. Cir. 2001) (“The
perception of the relevant purchasing public sets the standard for determining
descriptiveness.”). Evidence that a term is merely descriptive to the relevant
purchasing public may be obtained from any competent source, Nett Designs, 236
F.3d at 1341, such as dictionaries, newspapers, or surveys. Bayer, 488 F.3d at 964. It
also may be obtained from websites and publications, see N.C. Lottery, 866 F.3d at
1368, including an applicant’s own materials, website, or news articles. In re Promo
Ink, No. 76541018, 2006 TTAB LEXIS 138, at *4-5 (examining attorney may
introduce evidence that applicant’s own literature supports descriptiveness of term

Serial No. 97733261
- 11 -
even though application was based on intent-to-use under Trademark Act Section
1(b)).
B. Discussion
In determining the overall meaning of CHATGPT, we first consider how relevant
consumers would view the mark for the goods and services identified in the
application.19 Although CHATGPT is a compound term with no space, we find that
consumers would recognize the separate component terms: CHAT, which is a
recognizable word to consumers; and GPT, which (as discussed below) is an initialism.
Cf. In re Cox Enters. Inc., No. 76591278, 2007 TTAB LEXIS 6, at *10 (“THEATL is
simply a compressed version of the descriptive term THE ATL without a space
between the two words. Without the space, THEATL is equivalent in sound, meaning
and impression to THE ATL and is equally descriptive of applicant’s goods”).
We next consider whether the separate terms, CHAT and GPT, have any
meanings; and if so, whether they each are merely descriptive of Applicant’s goods
and services. In re Petroglyph Games, Inc., No. 78806413, 2009 TTAB LEXIS 465, at
*14 (“First, we examine the evidence concerning the meanings that would be ascribed
to the term BATTLECAM and the separate terms BATTLE and CAM, when used
19 A multiclass application, like the one here, is the equivalent of separate applications. G&W
Labs., Inc. v. G W Pharma Ltd., No. 91169571, 2009 TTAB LEXIS 2, at *6 (“[A] multiple-
class application can be viewed as a series of applications for registration of a mark in
connection with goods or services in each class, combined into one application.”) We must
determine the issue of descriptiveness as to each class of goods and services. Although the
classes in Applicant’s application cover different types of software (downloadable (Class 9) v.
non-downloadable (Class 42)), the purpose of the software is the same. For this reason, we do
not break our analysis out by class of goods and services.

Serial No. 97733261
- 12 -
with applicant’s goods.”) (citing In re Zanova Inc., No. 75519495, 2001 TTAB LEXIS
340 (ITOOL would be perceived as short for “internet tools” and was refused as
descriptive for goods and services including software for creating Web pages).
1. Is CHAT Merely Descriptive of the Identified Goods and
Services?
The Examining Attorney made of record several definitions of CHAT, including:
• THE AMERICAN HERITAGE DICTIONARY: “a synchronous exchange of
remarks over a computer network”;20
• Techopedia.com: “[T]he process of communicating, interacting,
and/or exchanging messages over the Internet ... Chat may be
delivered through text, verbal, audio, visual (A/V) communication via
the Internet.... Online chat services have purpose-built chat
interfaces that manage the entire communication processes…. Chat
is also known as chatting, online chat or Internet chat”;21
• COLLINS DICTIONARY: “When people chat, they exchange short
written messages on the internet or on their phones”;22
• CAMBRIDGE DICTIONARY: “to take part in a discussion that involves
sending messages over the internet, by smartphone, etc.”23
• PC MAGAZINE: “A text communication via keyboard in real time
between two or more users on a local network (LAN) or over the
Internet. Although the original use of the term was only for text,
‘chat’ became popular for every two-way communication.”24
The Examining Attorney points out that the identifications of goods and services
in both classes of the subject application include downloadable and non-downloadable
20 May 25, 2023 Office Action at TSDR 7.
21 Feb. 9, 2024 Final Office Action at TSDR 31-32.
22 Apr. 7, 2025 Continuing Final Office Action at TSDR 102.
23 Id. at TSDR 112.
24 Feb. 9, 2024 Final Office Action at TSDR 27.

Serial No. 97733261
- 13 -
“chatbot software for simulating conversations,”25 and argues that, because
identifications are required to specify the common commercial or generic name of the
services, TRADEMARK MANUAL OF EXAMINING PROCEDURE (TMEP) § 1402.01 (Nov.
2025), the inclusion of a portion of the mark within the identification itself generally
is a reliable indicator that such term is, at a minimum, merely descriptive.26
The Examining Attorney’s argument is persuasive. One dictionary defines
“chatbot” as “a computer program that can have conversations with humans.”27
Another similar definition is “a software program capable of mimicking human
language through text or voice in order to interact with users….”28 The evidence of
record also reveals that a “chatbot” is “used by a wider audience to answer questions,
and even engage in casual conversation.”29 Moreover, in its specimen of use, Applicant
explains its goods and services as “conservational”:
We’ve trained a model called ChatGPT which interacts in a
conversational way. The dialogue format makes it possible for ChatGPT
to answer the followup questions, admit its mistakes, challenge
incorrect premises, and reject inappropriate requests. ChatGPT is a
25 10 TTABVUE 4.
26 Id.
27 COLLINS COBUILD ADV. LEARNER’S DICTIONARY (collinsdictionary.com/us/
dictionary/english/chatbot, accessed Mar. 4, 2026). The Board may take judicial notice of
dictionary definitions, including online dictionaries that exist in printed format. Cordua
Rests., No. 85214191, 2014 TTAB LEXIS 94, at *6 n.4; see also B.V.D. Licensing Corp. v. Body
Action Design Inc., 846 F.2d 727, 728 (Fed. Cir. 1988) (“dictionaries and encyclopedias may
be consulted”).
28 WEBSTER’S NEW WORLD COLLEGE DICTIONARY (5th digital ed.) (collinsdictionary.com/us/
dictionary/english/chatbot, accessed Mar. 4, 2026).
29 Apr. 7, 2025 Continuing Final Office Action at TSDR 14.

Serial No. 97733261
- 14 -
sibling model to InstructGPT, which is trained to follow an instruction
in a prompt and provide a detailed response.30
In other words, the evidence shows that “chatbots,” as used in the identification of
Applicant’s goods and services, electronically converse or “chat” with humans in the
ordinary sense of that word. There is no multi-stage reasoning needed; rather, CHAT
immediately describes Applicant’s downloadable and non-downloadable software “for
simulating conversations” or “chatting.”
Applicant argues, however, that the term CHAT in the mark is suggestive because
it has several meanings relevant in the context of Applicant’s goods and services,
including “to talk in an informal or familiar manner” and “to take part in an online
discussion in a chat room.”31 As we discussed above, other potential meanings of
CHAT are irrelevant if one meaning is descriptive. IP Carrier Consulting, 2007 TTAB
LEXIS 66, at *19 (that a word has multiple meanings, some of which are not
descriptive, is not controlling; if any one of the meanings of a word is descriptive, the
word may be merely descriptive). But in this case, the definitions of “chat” relied upon
by Applicant are equally descriptive of downloadable or non-downloadable “computer
programs and … computer software for the artificial production of human speech and
text” and “for natural language processing, generation, understanding and
analysis.”32
30 Mar. 7, 2023 Specimen at TSDR 18.
31 8 TTABVUE 8 (citing Aug. 9, 2024 Request for Reconsideration at TSDR 248 (definition
from MERRIAM-WEBSTER DICTIONARY)).
32 We take judicial notice that “natural language” means “a language that is the native speech
of a people (as English, Tamil, Samoan)”; “the language of ordinary speaking and writing.”

Serial No. 97733261
- 15 -
Applicant also argues:
[W]hile “CHAT” may also refer to a type of casual conversation, the way
users interact with Applicant’s Goods and Services is not so familiar. To
use Applicant’s Goods and Services, users must write a prompt or series
of prompts .… Effective prompts are not casual but rather require
thought and planning: “[i]f you really want to generate something that
is going to be useful for you [with ChatGPT], you need to do more than
just write a generic sentence” .… Users refer to this method of creating
effective prompts to use in connection with Applicant’s Goods and
Services as “prompt engineering.” … In view of its multiple meanings,
“CHAT” may suggest, but does not directly describe, how users may
interact with Applicant’s Goods and Services.33
We find Applicant’s arguments unpersuasive in view of the evidence discussed
above that consumers are accustomed to “chatting” with “chatbots,” including
through use of Applicant’s goods and services. In addition, we must consider
descriptiveness in the context of the “average purchaser” of the goods and services.
Coach Servs. v. Triumph Learning, LLC, 668 F.3d 1356, 1378 (Fed. Cir. 2012)
(descriptiveness considered in the context of “the possible significance that the term
would have to the average purchaser of the goods because of the manner of its use or
intended use”) (quoting Bayer, 488 F.3d at 963). As discussed more fully below,
Applicant’s goods and services are not limited for use by software developers or other
experts in artificial intelligence. Based on the identifications of goods and services,
we must presume that in addition to being used by software developers, Applicant’s
downloadable and non-downloadable software is available to the general public and
casual users who do not take time to craft “effective prompts,” but rather are likely
MERRIAM-WEBSTER DICTIONARY (merriam-webster.com/dictionary/natural%20language,
accessed March 4, 2026).
33 8 TTABVUE 9.

Serial No. 97733261
- 16 -
to interact with the software to facilitate everyday communication or “chatting.” For
example, one prompt example in Applicant’s specimen is: “Got any creative ideas for
a 10 year old’s birthday?”34
In sum, we find that the term CHAT in Applicant’s mark CHATGPT is merely
descriptive of a purpose or function of Applicant’s software, i.e., to assist or facilitate
communication in a conversational manner, as reflected in Applicant’s identifications
of goods and services.
2. Is GPT Merely Descriptive of the Identified Goods and
Services?
We next turn to the second term of the applied-for mark: GPT. A term, an
initialism, or an acronym is merely descriptive when it is a recognized abbreviation
for the goods or services in the application, or is otherwise understood as being
“substantially synonymous” with the merely descriptive words it represents.35 In re
Thomas Nelson, Inc., No. 76681269, 2011 TTAB LEXIS 9, at *14-15 (citing Modern
Optics, Inc. v. Univis Lens Co., 234 F.2d 504, 506 (CCPA 1956)); see also In re BetaBatt
Inc., No. 77036122, 2008 TTAB LEXIS 59, at *9-11 (holding DEC substantially
synonymous with merely descriptive term “direct energy conversion” and thus merely
descriptive of a type of batteries and battery related services). We consider whether:
34 Mar. 7, 2023 Specimen at TSDR 7.
35 An acronym is “a word (such as NATO, radar, or laser) formed from the initial letter or
letters of each of the successive parts or major parts of a compound term,” whereas an
initialism is “an abbreviation (such as FBI) formed from initial letters.” MERRIAM-WEBSTER
DICTIONARY (merriam-webster.com/dictionary/acronym, accessed Mar. 4, 2026). See also
Aug. 9, 2024 Request for Reconsideration at TSDR 98 (Carter Decl. ¶ 15) (“One way coinages
are created is through initialism, where an underlying long-form is represented with the first
letter of each discrete form in the expression….GPT is among this group.”). Thus, GPT is
accurately termed an initialism.

Serial No. 97733261
- 17 -
(1) the letters GPT are an abbreviation for specific wording;
(2) the specific wording is merely descriptive of the applied-for goods and
services; and
(3) the relevant consumers will recognize the abbreviation as the merely
descriptive wording it represents.
Thomas Nelson, 2011 TTAB LEXIS 9, at *15.
a. Is GPT an Abbreviation for Specific Wording?
Applicant and the Examining Attorney agree that GPT stands for “generative pre-
trained transformer.”36 This meaning is confirmed by three dictionary definitions of
record for the term “GPT”:
• CAMBRIDGE DICTIONARY: “Generative Pretrained Transformer: a
complex mathematical representation of text or other types of media
that allows a computer to perform some tasks, such as interpreting
and producing language, recognizing or creating images, and solving
problems, in a way that seems similar to the way a human brain
works”;37
• Reverso English Dictionary: “(acr: generative pretrained
transformer) AI that creates text using learned patterns”;38
• Dictionary.com: “generative pre-trained transformer: a type of
machine learning algorithm that uses deep learning and a large
database of training text in order to generate new text in response to
a user’s prompt;”39 and
36 8 TTABVUE 11; 10 TTABVUE 5.
37 Apr. 7, 2025 Continuing Final Office Action at TSDR 10.
38 Id. at TSDR 43.
39 Id. at TSDR 96. That “GPT” does not appear in other dictionaries, such as THE OXFORD
ENGLISH DICTIONARY or MERRIAM-WEBSTER DICTIONARY, see Aug. 9, 2024 Request for
Reconsideration at TSDR 98-99, 128 (Carter Decl. ¶¶ 16, 84), is noted but not determinative.

Serial No. 97733261
- 18 -
• URBAN DICTIONARY: “GPT, which stands for ‘Generative Pretrained
Transformer,’ was developed by the company OpenAI.”40
Evidence of record from an acronym finder also includes an entry for an abbreviation
of “GPT” as “Generative Pre-Trained Transformer (language processing).”41
Additional evidence of record confirms that third parties, including mainstream
media, use “GPT” as an initialism for “generative pre-trained transformer,” including:
• THE ECONOMIST: “But how will GPTS (short for generative pre-
trained transformers) and other such services make money? The
answer is by providing their technology to companies. These firms
can use them to create content, such as articles, blog posts and entire
books. They can help with customer service, providing quick and
accurate responses to queries. And they can be used to help with
research, providing insights into consumer behaviour and
preferences.”42
• FORBES MAGAZINE: “GPT-1, the model that was introduced in June
2018, was the first iteration of the GPT (generative pre-trained
transformer) series and consisted of 117 million parameters.”43
• THE NEW YORK TIMES: “Transformer models are the key to systems
like ChatGPT — whose full acronym stands for ‘Generative
Pretrained Transformer.’”44
40 Id. at TSDR 106-07 (Carter Decl. ¶ 35). Dr. Carter opines that “Numerous linguists have
written specifically about the value of Urban Dictionary in conducting linguistic research on
word meanings.” Id. at 105 (Carter Decl. ¶ 34 n.6).
41 May 25, 2023 Office Action at TSDR 9.
42 Nov. 27, 2023 Office Action Response at TSDR 141 (economist.com/
business/2022/12/08/how-good-is-chatgpt). This article reports that Applicant’s CHATGPT
software product, in response to a query, stated that (albeit in Shakespearean trope) “GPTS
be a type of AI technology that can generate natural language in response to input.” Id.
(emphasis added)
43 Id. at TSDR 308 (forbes.com/sites/bernardmarr/2023/05/19/a-short-history-of-chatgpt-how-
we-got-to-where-we-are-today/?shcc2815b097674f).
44 Id. at TSDR 361 (nytimes.com/2023/03/28/technology/ai-chatbots-chatgpt-bing-bard-
llm.html); see also id. at TSDR 134-35, 675-76, 741 (nytimes.com/2022/12/
05/technology/chatgpt-ai-twitter.html).

Serial No. 97733261
- 19 -
• BBC SCIENCE FOCUS MAGAZINE: “GPT-3 (Generative Pretrained
Transformer 3), GPT-3.5 and GPT-4 are state-of-the-art language
processing AI models developed by OpenAI. They are capable of
generating human-like text and have a wide range of applications,
including language translation, language modelling, and generating
text for applications such as chatbots.”45
• BBC SCIENCE FOCUS MAGAZINE: “The ‘GPT’ In ChatGPT stands for
Generative Pre-trained Transformer.”46
• USA TODAY: “ChatGPT (Generative Pre-trained Transformer) is the
latest viral sensation out of San Francisco-based startup OpenAI.”47
• BBC NEWS: “ChatGPT is the latest in a series of AIs which the firm
refers to as GPTs, an acronym which stands for Generative Pre-
Trained Transformer.”48
• Ideamotive: “As AI technology advances, the development of GPT
(Generative Pretrained Transformer) models has become
increasingly popular. In this article, we will take you through the
process of building your own GPT project, from planning to
deployment.”49
• BMC: “We leverage these tribal resolutions by tapping into ticket
worklogs. We separate redundant entries from actual resolutions
and use these resolutions to train a large neural network that
complements a Generative Pretrained Transformer (GPT) model, as
shown in the below diagram…”50
• RealestatecareerHQ.com: “One example of this technology is
RoomGPT.io, an AI-powered virtual staging tool. RoomGPT.io uses
GPT technology, which stands for Generative Pretrained
Transformer, to generate realistic images of staged rooms. This
45 Id. at TSDR 454 (sciencefocus.com/future-technology/gpt-3). See supra text accompanying
n.9 (discussing Applicant’s objection to consideration of foreign English language websites).
46 Id. at TSDR 456 (sciencefocus.com/future-technology/gpt-3).
47 Id. at TSDR 567, 729 (usatoday.com/story/tech/2023/01/27/chatgpt-buzzfeed-ai/
11129947002).
48 Id. at TSDR 667 (bbc.com/ news/technology-638613220).
49 May 25, 2023 Office Action at TSDR 32 (ideamotive.com/gpt-software).
50 Id. at TSDR 39 (bmc.com/blogs/bmc-helixgpt-expert-of-your-systems).

Serial No. 97733261
- 20 -
technology is able to create images that are so realistic that it can be
hard to distinguish them from actual photos.”51
• Digitalinnov.com: “Artificial Intelligence: GPT (Generative Pre-
trained Transformer) … GPT, on the other hand, is a type of AI
language model that can generate text in response to a given prompt.
It is trained on a large corpus of text and can generate coherent and
natural-sounding sentences.”52
• H2O.ai: “What is GPT (Generative Pre-Trained Transformers)? GPT,
short for Generative Pre-Trained Transformers, is an advanced open-
source language model that utilizes transformer architectures to
generate human-like text. It is trained on vast amounts of unlabeled
text data from the internet, enabling it to understand and generate
coherent and contextually relevant text. Unlike rule-based systems,
GPT learns patterns and structures in text data to generate human-
like responses.”53
• NEWSWEEK: “In today’s rapidly evolving business landscape,
companies are increasingly harnessing the power of GPT (Generative
Pre-trained Transformer) and AI (Artificial Intelligence)
technologies to revolutionize their operations and drive
transformation. GPT, exemplified by models like GPT-3, is a cutting-
edge language model that has taken the field of natural language
processing to new heights. It can understand context, generate
human-like text responses, and perform language-related tasks with
astounding accuracy…. GPT and AI are different but closely related
technologies transforming the business’s future. GPT is a machine
learning model pre-trained on vast amounts of data. It can
understand the context and generate natural-sounding language,
making it an ideal tool for tasks such as chatbots, virtual assistants,
and content generation.”54
• Nostalab.com: “As technological advancements continue to disrupt
various industries, the field of artificial intelligence (AI) has become
a vital tool in revolutionizing how we approach and solve complex
problems. One such example is the emergence of Generative Pre-
51 Id. at TSDR 47 (realestatecareerHQ.com/AI-virtual-staging).
52 Feb. 9, 2024 Office Action at TSDR 14 (digitalinnov.com/blog/search-engine-vs-artificial-
intelligence-gpt).
53 Id. at TSDR 36 (h2o-ai-wiki-gpt-17043767469591).
54 Id. at TSDR 41-42 (newsweek.com/harnessing-power-gpt-ai-transforming-future-business-
1805363).

Serial No. 97733261
- 21 -
Trained Transformer (GPT), a powerful deep learning algorithm that
has shown remarkable success in various fields. GPT is a type of
neural network that can generate human-like language, allowing it
to understand context and meaning in text-based inputs.”55
• Amazon (AWS): “What is GPT? Generative Pre-trained
Transformers, commonly known as GPT, are a family of neural
network models that uses the transformer architecture and is a key
advancement in artificial intelligence (AI) powering generative AI
applications such as ChatGPT. GPT models give applications the
ability to create human-like text and content (images, music, and
more), and answer questions in a conversational manner.
Organizations across industries are using GPT models and
generative AI for Q&A bots, text summarization, content generation,
and search.”56
• Google Cloud: “GPT, or a generative pre-trained transformer, is a
type of large language model (LLM) that utilizes deep learning to
produce human-like text. Neural networks are trained on massive
datasets containing text and code, enabling them to understand and
generate coherent and contextually relevant responses. As a key
component in the field of in the field of generative AI, GPT pushes
the boundaries of what’s possible with Al. enabling machines to
produce creative and human-quality content.”57
• SolveForce: “Generative Pre-trained Transformer (GPT): A large-
scale language model that can be fine-tuned for a wide range of
natural language processing tasks, it can also be used for text
generation.”58
• Lumenalta: “Custom GPTs, or customizable generative pretrained
transformers, represent a powerful tool for businesses and
professionals looking to tailor artificial intelligence (Al) for their
specific needs. At their core, GPTs are large language models capable
of understanding and generating human-like text based on the input
they receive.”59
55 Id. at TSDR 51 (nostalab.com/gpt).
56 Id. at TSDR 62 (aws.amazon.com/what-is/gpt).
57 Apr. 7, 2025 Continuing Final Office Action at TSDR 13 (cloud.google.com/discover/what-
is-gpt).
58 Id. at TSDR 18 (solveforce.com/generative-pre-trained-transformer-gpt-definition).
59 Id. at TSDR 24-25 (lumenalta.com/insights/what-is-a-custom-gpt-key-features-benefits).

Serial No. 97733261
- 22 -
• Microsoft: “What are GPTs (and how can I make one)? GPTs, or
generative pre-trained transformers, are versatile Al-powered tools
that are easily customized to suit your specific needs. For example,
if you’re looking for a GPT to help with homework, create your own
GPT and train it on your study materials. Learn more about GPTs,
how to make your own, and how to leverage them as your ultimate
digital assistant.”60
• Fingoweb.com: “GPT applications (Generative Pre-trained
Transformer) utilize advanced artlfic1al Intelligence models to
generate text that mimics the way humans write. These applications
are based on natural language processing (NLP) technology and can
produce high-quality, coherent, and contextually relevant content
based on previously Inputted data. GPT can be applied in numerous
fields, from automating customer service and creating marketing
content to generating programming code.”61
• OpenTrain.ai: “Generative Pretrained Transformer (GPT)
Definition: The Generative Pretrained Transformer (GPT) is a state-
of-the-art language model that utilizes the transformer architecture
to generate coherent and contextually relevant text based on a given
prompt.”62
• Poll The People: “Custom GPT, or Generative Pre-trained
Transformers, is advanced AI model tailored to specific tasks or
Industries, enhancing their effectiveness in those areas.”63
We find the evidence of record amply demonstrates that “GPT” is a recognized
initialism for “generative pre-trained [or pretrained] transformer.”64
60 Id. at TSDR 39 (microsoft.com/en-us/microsoft-365-life-hacks/everyday-ai/what-is-a-gpt).
61 Id. at TSDR 47 (fingoweb.com/it-dictionary/gpt-application).
62 Id. at TSDR 50 (opentrain.ai/glossary/generative-pretrained-transformer-gpt).
63 Id. at TSDR 60 (pollthepeople.app/custom-gpt-software-code-analysis).
64 Again, that GPT may be an initialism for other terms in other contexts, see 8 TTABVUE
12, is not relevant to our analysis. IP Carrier Consulting, 2007 TTAB LEXIS 66, at *19. For
simplicity, we use “pre-trained” in the remainder of this opinion; the hyphen omitted by some
third parties is not relevant to our analysis.

Serial No. 97733261
- 23 -
b. Is “Generative Pre-Trained Transformer” Merely
Descriptive of the Identified Goods and Services?
We next turn to the question of whether the wording “generative pre-trained
transformer” is merely descriptive of the identified goods and services.
Descriptiveness is determined from the perspective of a relevant consumer who is
familiar with the services. See, e.g., Bayer, 488 F.3d at 964 (descriptiveness evaluated
from perspective of an “average purchaser of the goods [or services]”). As discussed
above, the identified goods and services are not limited to a particular type of user
and thus include all relevant consumers, including software developers and the
general public.
The internet evidence discussed above establishes that “generative pre-trained
transformer” refers to a machine learning algorithm that is pre-trained to produce
human-like text in response to a human’s prompts. We also note that, taking the
terms separately, “generative” describes a feature of Applicant’s consumer-facing
large language models (LLMs), namely, that they “generate” new content such as
text, images, or code in response to user prompts. “Pre-trained” refers to the fact that
LLMs such as Applicant’s model are trained on a large body of data (i.e., text and
images), so they do not start from scratch.65 Once training is complete, “LLMs
undergo the process of deep learning through neural network models known as
65 8 TTABVUE 5 (“Large language models [including GPT] are neural-network based systems
trained on a vast amount of data using self-supervised learning techniques.”).

Serial No. 97733261
- 24 -
transformers, which rapidly transform one type of input to a different type of
output.”66
Because “generative pre-trained transformer” refers to a machine learning
algorithm that is pre-trained to produce human-like text in response to a human’s
prompts, we find that it merely describes a feature or characteristic of Applicant’s
downloadable and non-downloadable software, namely, that it facilitates “the
artificial production of human speech and text, … natural language processing,
generation, understanding and analysis,” and it simulates conversations.67
c. Will Consumers Recognize GPT as the Merely Descriptive
Wording It Represents?
We turn to the last element of the test: that is, whether relevant consumers will
recognize the initialism GPT as the merely descriptive wording it represents,
“generative pre-trained transformer.” Applicant argues that the evidence of record,
66 “Large Language Model,” ENCYCLOPEDIA BRITTANICA (britannica.com/topic/large-
language-model, accessed Mar. 4, 2026). According to the evidence of record, the
“transformer” neural network was invented by Google in 2017 and “turbocharged” the power
of chatbots; they are “key” to AI systems like Applicant’s CHATGPT offering. Nov. 27, 2023
Office Action Response, at TSDR 149, 153, 361, 697. See also ENCYCLOPEDIA BRITTANICA,
supra (“The first large language models emerged as a consequence of the introduction of
transformer models in 2017”).
To the extent that we rely on this or other evidence of record for the truth of the matter
asserted therein, there is no rule against consideration of hearsay evidence in ex parte
proceedings. In re Epstein, 32 F.3d 1559, 1566 (Fed. Cir. 1994). The Board “generally takes a
somewhat more permissive stance with respect to the admissibility and probative value of
evidence in an ex parte proceeding than it does in an inter partes proceeding....” In re Canine
Caviar Pet Foods, Inc., No. 85710350, 2018 TTAB LEXIS 155, at *18 (quoting TRADEMARK
TRIAL AND APPEAL BOARD MANUAL OF PROCEDURE (TBMP) § 1208 (June 2017)).
67 We need not and do not consider whether “generative pre-trained transformer” or GPT are
generic – “the ultimate in descriptiveness” – for Applicant’s goods and services. H. Marvin
Ginn Corp. v. Int’l Ass’n of Fire Chiefs, Inc., 782 F.2d 987, 989 (Fed. Cir. 1986).

Serial No. 97733261
- 25 -
much of which is discussed above, is largely directed towards software developers and
not average consumers who are unlikely to understand the meaning of GPT.68 We
find this argument unpersuasive for several reasons.
First, the evidence demonstrates that many users of Applicant’s goods and
services actually are sophisticated users. Applicant’s specimen of use advertises
intended use of the identified goods and services by software developers and other
sophisticated users.69 Applicant argued during prosecution that: “Each of
[Applicant’s] software models incorporates ‘GPT’ within the brand name and is used
by software developers to build applications.”70 To the extent the genericness survey
submitted by Applicant is relevant, Dr. Neal selected software developers as the
“relevant consumer public”:
I understand that software developers are important target consumers
for the AI and machine-learning software offered by OpenAI under the
GPT mark. For this reason, I ensured that 100% of respondents in my
survey were software developers, and were currently either employed or
self-employed full-time, or were currently freelance or contract-based
part-time.71
We conclude that it is not improper to focus on software developers and other
sophisticated users of Applicant’s goods and services as part of the relevant consumer
class.
68 8 TTABVUE 13; id. at 25-29 (Exhibit A); 11 TTABVUE 3-4.
69 See Mar. 7, 2023 Specimen at TSDR 4, 12, 15, 18-19.
70 Aug. 9, 2024 Request for Reconsideration at TSDR 15.
71 Id. at TSDR 32 (Neal Decl. ¶ 1.9).

Serial No. 97733261
- 26 -
Second, the evidence of record supporting the meaning of the initialism GPT
comes not only from materials targeted at sophisticated users, but also from sources
targeted at everyday consumers. The term GPT is included in mainstream (i.e., non-
technical) dictionaries such as CAMBRIDGE DICTIONARY, Dictionary.com, and URBAN
DICTIONARY.72 The meaning of the initialism also has been discussed in mainstream
publications such as THE NEW YORK TIMES, THE ECONOMIST, NEWSWEEK, FORBES
MAGAZINE, USA TODAY, and BBC NEWS.73 This evidence belies the argument that only
sophisticated consumers understand the meaning of GPT as an initialism for
“generative pre-trained transformer.”
Nor are we convinced by Applicant’s reliance on Baroness Small Estates, Inc. v.
Am. Wine Trade, Inc., No. 92051369, 2012 TTAB LEXIS 335.74 In that case, the Board
found that the initialism CMS (meaning cabernet, merlot and syrah) was suggestive
for wine because the “authors of the various articles and reviews and sales
information believed that they needed to spell out the connection between CMS and
the names of the varietals that are contained in the wine, and that readers would not
immediately understand that the mark CMS has the meaning of cabernet sauvignon,
merlot, syrah.” Id. at *23. In that case, there was no evidence that “C” was a
recognized abbreviation for “cabernet sauvignon” or that “M” was a recognized
abbreviation for “merlot” or that “S” was a recognized abbreviation for “syrah,” id. at
72 See supra text accompanying nn. 37, 39, and 40.
73 See supra text accompanying nn. 42, 43, 44, 45, 46, 47, 48, and 54.
74 8 TTABVUE 11.

Serial No. 97733261
- 27 -
*23-24, let alone evidence of common use of CMS. In this case, even if each individual
letter (G, P, and T) is not a well-recognized abbreviation for the word to which it
corresponds, the record demonstrates that the initialism GPT as a whole is commonly
used, including in mainstream dictionaries, acronym finders, and news sources. In
addition, there are examples in the record of articles using GPT on its own without
spelling out the meaning “generative pre-trained transformer,” indicating that the
abbreviation is recognized on its own.75
Applicant also relies on the expert opinion of Dr. Carter, who (in the course of
concluding that GPT is not generic) opines that GPT is a “neologism” coined by
Applicant.76 But just because Applicant may have been the first to adopt or use GPT
or “generative pre-trained transformer” does not render either term or phrase
inherently distinctive.77 See In re Fallon, No. 86882668, 2020 TTAB LEXIS 464, at
*32-33 (“‘The fact that Applicant may be the first or only user of a term does not
render that term distinctive’ if … it has been shown to be merely descriptive of the
goods identified in the application.”) (quoting Fat Boys, 2016 TTAB LEXIS 150, at
*10); In re Swatch Grp. Mgmt. Servs. AG, No. 85485359, 2014 TTAB LEXIS 131, at
75 See, e.g., Apr. 7, 2025 Continuing Final Action at TSDR 52-55 (using GPT alone, explaining
meaning on a separate FAQ page); id. at TSDR 72-81 (explaining how to build a GPT model
from scratch); id. at TSDR 83-95 (concept design for building a “GPT-powered tutor”); id. at
TSDR 119-22 (discussing GPT-based chatbots); id. at TSDR 141-44 (releasing a “a pre-built
application for commercial invoices on AI Hub which allows users to automate invoice
processing using GPT models”); id. at TSDR 145-53 (offering a “Private GPT AI solution …
designed to provide an intelligent private chat based local knowledgebase,” to “[u]se your
own, trusted GPT”).
76 8 TTABVUE 12-13; Aug. 9, 2024 Request for Reconsideration at TSDR 97-98, 128 (Carter
Decl. ¶¶ 13, 85). See also 11 TTABVUE 6 (discussing Carter Declaration).
77 See 8 TTABVUE 14.

Serial No. 97733261
- 28 -
*29 n.50 (“Being ‘the first and only one to adopt and use the mark sought to be
registered does not prove that the mark is not descriptive.’”) (quoting In re Bailey
Meter Co., 102 F.2d 843, 844 (CCPA 1939)).78
We conclude that relevant consumers will recognize GPT as the merely descriptive
wording it represents: “generative pre-trained transformer.”79
C. Is CHATGPT More Than the Sum of Its Parts?
Having determined that each term in the mark is merely descriptive of Applicant’s
goods and services, we look at the mark CHATGPT in its entirety. Applicant contends
that the Examining Attorney has improperly dissected the mark into its constituent
terms CHAT and GPT and has not considered the mark CHATGPT as a whole.80
As discussed above, it is entirely acceptable to look at each individual component
of a compound mark to divine its meaning as whole. A mark comprising a combination
of merely descriptive components is registrable, however, if the combination of terms
creates a unitary mark with a nondescriptive meaning. See, e.g., In re Colonial Stores
Inc., 394 F.2d 549 (CCPA 1968) (finding SUGAR & SPICE for “bakery products”
suggestive); In re Shutts, 1983 TTAB LEXIS 150, at *7 (finding SNO-RAKE for “a
snow removal hand tool having a handle with a snow-removing head at one end, the
head being of solid uninterrupted construction without prongs” suggestive). That is
not the case here.
78 Aug. 9, 2024 Request for Reconsideration at TSDR 119 (Carter Decl. ¶ 65).
79 As mentioned above, the issue of whether GPT is generic for any goods or services is not at
issue in this appeal and we make no such finding either way. See supra n.67.
80 8 TTABVUE 15-16, 18-19; 11 TTABVUE 2, 7, 8, 9.

Serial No. 97733261
- 29 -
As explained, the terms CHAT and GPT individually are descriptive of Applicant’s
goods and services. Viewed in its entirety, we find that CHATGPT does not present
an incongruity or lose its descriptive meaning, rather it immediately informs the
consumer that Applicant’s goods and services feature a conversational chatbot,
utilizing a generative, pre-trained transformer algorithm. Applicant points to no
other meaning of the compound term CHATGPT aside from that of a source identifier
for its goods and services.81
Moreover, the evidence of record reflects third-party use of “chat” and “chatbot” in
the context of conversations with “GPT” software models: the “GPT chat use-case
leverages generative models to deliver real-time, interactive conversations with GPT
models. This capability is particularly useful for chatbots and virtual assistants,
providing uses with responses that are contextually relevant and accurate.”82
Further, “GPT-based chatbots generally require system prompts, which are text
strings that are prepended to each conversation to set the scope and tone of the
chatbots’s [sic] interactions with human users.”83 One problem is “Chatty GPT” where
“your GPT model [is] giving excessively long responses….”84 Another software
81 Dr. Carter opines that “[a]s a part of the derivation process (coinage), the words
‘generative,’ ‘pre-trained,’ and ‘transformer’ were put into a unique sequence reflective of the
technology underlying this product.” Aug. 9, 2024 Request for Reconsideration at TSDR 98
(Carter Decl. ¶ 14). There are only two ways to sequence these descriptive adjectives
(generative, pre-trained) and noun (transformer) such that they make sense, and both are
descriptive. But even this “unique sequence” of words does nothing to change the meaning of
the terms, separately or together.
82 Apr. 7, 2025 Continuing Final Office Action at TSDR 119.
83 Id. at TSDR 120.
84 Id. at TSDR 121.

Serial No. 97733261
- 30 -
offering enables users to “choose from the top LLM models to chat,” and those users
who “need additional information from the web, [to] seamlessly switch to chat with
web within a chat session.”85 In addition, Applicant explains that it trained its
CHATGPT model through “conversations that [human] AI trainers had with the
chatbot” and then ranking them.86 This evidence demonstrates that CHATGPT has
the same meaning as CHAT and GPT in the context of Applicant’s goods and services.
Applicant made of record five third-party registrations for goods and services in
Classes 9 and 42 for marks including the term CHAT followed by an “arguably
descriptive term for the relevant software or software services that the Office
considered inherently distinctive” to support its argument that compound marks such
as CHATGPT are suggestive:87
• CHATSTAT (standard characters) for “Consulting services in the
field of software as a service (SAAS); software as a service (SAAS)
services featuring software for the purpose of analyzing social media
posts and sending alert notifications; software as a service (SAAS)
services featuring software using artificial intelligence for the
purpose of analyzing social media posts and sending alert
notifications,” in Class 42;
• CHATBRIDGE (standard characters) for “Software as a service
(SAAS) services featuring software for text based correspondence
and instant messaging, real-time customer support, customer service
and customer engagement; Providing temporary use of on-line non-
downloadable software tools for text based correspondence and
instant messaging, real-time customer support, customer service and
customer engagement; Software as a service (SAAS) services
85 Id. at TSDR 130.
86 Mar. 7, 2023 Specimen at TSDR 19.
87 See 8 TTABVUE 9-10 (discussing Aug. 9, 2024 Request for Reconsideration at TSDR 273-
302). Applicant also made of record five pending applications for “chat” formative marks.
Applications, however, have no probative value other than as evidence that the application
was filed. In re Team Jesus LLC, No. 88105154, 2020 TTAB LEXIS 503, at *17 n.29.

Serial No. 97733261
- 31 -
featuring software for allowing customers to converse with a
customer service representative in real time via live messaging;
Providing temporary use of on-line non-downloadable software tools
for allowing customers to converse with a customer service
representative in real time via live messaging,” in Class 42;
• CHATMALL (standard characters) for “Downloadable mobile
application and computer software that can enable users to make
inquiries, post and respond to online classified advertisements,” in
Class 9, and “Operating on-line marketplaces for sellers and buyers
of goods and services featuring Automobiles, Automobile parts,
clothing, household electronics, personal computers, mobile phones
and mobile phone accessories,” in Class 35;
• CHATBASE (standard characters) for “Application service provider
featuring application programming interface (api) software that
provides analytics capabilities to measure user experience with
computer programs designed to simulate conversation with human
users to help users create better computer programs to simulate such
conversations,” in Class 42; and
• CHATATTACH (standard characters) for “Electronic hardware
accessories, namely, speakers, speakerphones and microphones,
cables, firmware and software that add two way audio conferencing
capability to multiple devices,” in Class 9.
These other CHAT-formative marks do not persuade us that Applicant’s proposed
mark CHATGPT is not merely descriptive. As Applicant acknowledges, the Board is
not bound by prior decisions made in connection with different marks, and each case
must be decided on the proceeding record before us and on its own merits. In re
Cordua Rests., Inc., 823 F.3d 594, 600 (Fed. Cir. 2016) (citing In re Shinnecock Smoke
Shop, 571 F.3d 1171, 1174 (Fed. Cir. 2009)).
We find the meaning of the compound term CHATGPT to be no more than the
sum of its individual parts, and thus it is merely descriptive of the identified goods
and services under Section 2(e)(1). See In re Polo Int’l Inc., No. 74729974, 1999 TTAB
LEXIS 59, at *5 (DOC-CONTROL for document management software is merely

Serial No. 97733261
- 32 -
descriptive because DOC-CONTROL is not incongruous, creates no double entendre,
and does not create or present a commercial impression or meaning other than
“document control.”); In re Pennzoil Prods. Co., No. 73670049, 1991 TTAB LEXIS 27,
at *8 (MULTI-VIS for multiple viscosity motor oil is merely descriptive because the
plain meaning of the components comprising the term MULTI-VIS as confirmed by
dictionary definitions, other evidence, and applicant’s specimens when considered in
the context of applicant’s goods, demonstrate there is nothing incongruous about the
term MULTI-VIS or requires the exercise of mental processing to perceive the
descriptive significance of such term as it relates to multiple viscosity motor oil).
D. Other Arguments
Applicant argues that “competitors do not need to use GPT,” because its
competitors have adopted other names (such as Perplexity, Copilot, Bard, and
Claude).88 That a competitor does not need to use a term is not the test for mere
descriptiveness. In re Carlson, No. 78752616, 2009 TTAB LEXIS 438, at *17.
In any event, the record demonstrates that third parties use GPT descriptively89
and have adopted GPT as part of brand names for related goods and services.
Examples of brand names include: RoomGPT.io for AI virtual staging;90 Context GPT
for a “GPT-powered contextual platform allows users to build custom prompts”;91
88 8 TTABVUE 14-15.
89 See supra Section II(B)(2).
90 May 25, 2023 Office Action at TSDR 47.
91 Feb. 9, 2024 Final Office Action at TSDR 11.

Serial No. 97733261
- 33 -
MageAI GPT for an AI chat solution;92 GPT Definity Ai for a “generative AI based
crypto auto-trading platform”;93 ZotGPT Chat for a custom generative artificial
intelligence solution;94 FINANCEGPT for building “your own AI co-pilots for
personalized finance insights…”;95 and DermGPT for an “AI powered tool specializing
in dermatology.”96 Other third parties offer customized “GPT” solutions.97
Thus, although competitive need is not the test, the record demonstrates that
third parties commonly use GPT in connection with goods and services related to
Applicant’s here, further supporting our conclusion that GPT is descriptive. Cf. Juice
Generation, Inc. v. GS Enters. LLC, 794 F.3d 1334, 1339 (Fed. Cir. 2015) (Board erred
in not considering evidence of third-party use and registrations bearing on whether
a portion or a mark carried a suggestive or descriptive connotation in the relevant
industry); see also In re Taverna Izakaya LLC, No. 88612441, 2021 TTAB LEXIS 427,
at *6 (third party uses demonstrating descriptiveness). Third-party use of this sort
illustrates how descriptive terms must be kept free for use by competitors and the
public. See In re Abcor Dev. Corp., 588 F.2d 811, 813 (CCPA 1978) (“The major reasons
for not protecting [merely descriptive] marks are: (1) to prevent the owner of a mark
from inhibiting competition in the sale of particular goods [or services]; and (2) to
92 Id. at TSDR 19.
93 Id. at TSDR 55.
94 Apr. 7, 2025 Continuing Final Office Action at TSDR 123.
95 Id. at TSDR 135.
96 Id. at TSDR 154.
97 See, e.g., id. at TSDR 24-35 (Lumenalta), 52-67 (Poll The People).

Serial No. 97733261
- 34 -
maintain freedom of the public to use the language involved, thus avoiding the
possibility of harassing infringement suits by the registrant against others who use
the mark when advertising or describing their own products [or services].”) (citation
omitted).
Finally, Applicant suggests that, to the extent doubt exists as to whether
Applicant’s mark CHATGPT is merely descriptive of the identified goods and
services, we must find that the mark is suggestive.98 We have no such doubt in this
case.
Decision
The refusal to register application Ser. No. 97733261 on the ground of mere
descriptiveness for the identified goods in Class 9 and services in Class 42 is
affirmed. The application will proceed with Applicant’s claim of acquired
distinctiveness under Section 2(f).
98 8 TTABVUE 20-21.