top of page
Search

Trademark Protection of Virtual Brand Mascots: Assessing Distinctiveness in AI-Generated Marks through a Bounded Variability Framework

sonalimukhia2002
11 minutes ago
16 min read

Author: Rishita jain, LNCT University


ABSTRACT

Virtual brands mascots are increasingly deployed as AI-generative personas whose visuals and behavioural form varies with each consumer interaction. This departs sharply from the fixed, reproducible signs that Indian trademark doctrine was built around, and the law has not caught up. Section 9 (1) of the Trademarks Act 1999, as read in Imperial Tobacco co. vs Registrar of Trademark  and Godfrey Phillips India Ltd vs. Girnar Food and Beveragesties acquired distinctiveness to a consistent commercial impression, a standard that assumes  consumers encounter  the same sign each time, not one regenerated anew  at every  interaction. This paper argues that the proviso, so interpreted, creates a structural obstacle to both the registering and retaining protection for the generative virtual brand mascots leaving such marks vulnerable at filing and exposed to non- use risk under section 47. The argument  proceeds through doctrinal analysis of the relevant  Indian statutory  and case laws  followed  by the comparative look at the United kingdoms series mark provision (section 41 of Trademarks Act 1994) whether its underlying  logic  can be adopted for Indian use and a more suggestive framework  will be discussed in this. The paper contribution is proposed bounded variability registration model applicants would declare a fixed core and defined range of permissible  variable  offering a narrow , workable fix rather than a call for wholesale  reform.


KEYWORDS: Virtual Mascots, Trade Distinctiveness, Acquired Distinctiveness, AI-Generated Brand marks, Trademarks Act 1999.


INTRODUCTION 

The protection of virtual brand mascots understood broadly implicates at least three distinct bodies of Indian law , trademark  law, which protects the  mascots  as a source  identifier, copyright law, with  the assistance  of artificial intelligence and  personality  rights , which in India remain confined to living  natural persons  under ICC development (International) Ltd. v Arvee Enterprises, leaving open the separated question of unauthorised cloning or misappropriation of a mascots  persona. Each of these presents its own doctrinal difficulties in the content of AI generated mascots, and each merits sustained treatment in its own right. This paper does not attempt to resolve all three. Its focus is confined to the first whether Indian trademark law acquired distinctiveness doctrine under section 9 of the Trade marks Act 1999 can accommodate a mascot whose commercial  manifestation varies with each consumer interaction. This narrower focus is deliberate. The trademark distinctiveness question is both the most immediate since it determines whether such a mascots  can be registered and retained as a protectable assets at all and the least examined in existing Indian scholarship within  has so far addressed static marks and the classification of virtual goods without addressing  the specific  problem of a mark whose form is generated a new  identity, after registration. The copyright  and personality rights dimensions of mascots protection are accordingly noted here  as important and unresolved but are left for separate treatment. This is not a marginal or speculative commercial  phenomenon. Whatever  their precision as  reports the market at $6.1B (2024), projected $11.8B by 2026, 40.8 percent CAGR through 2030 market forecasts confirm that AI generative brand personas have moved from the experimental deployment to a distinct and rapidly scaling commercial category. The doctrinal gap identified in this paper accordingly attaches not to an isolated feature of one company product but to an entire and expanding class of commercial assets  


LITERATURE REVIEW  

Existing literature  and authorities broadly agree that Indian trademark law protects non traditional signs where they can satisfy the basic requirements of representation and distinctiveness. Section 9 (1) of the Trademarks Act 1999 provides the doctrinal beginning point while section 2 1 (zb) and rules 23 and 26 demonstrates that law already accommodated form such as shape of goods, packaging marks. The disagreement is over where the relation of identity and fixity is located. Existing authorities do not expressly require a mark to be visually identical in each manifestation. But the literature and the administrative framework appear to presuppose a determinate representation. Comparative approaches reinforce rather than resolve this tension. The UK's series-mark mechanism permits controlled variation, while EUIPO treatment of motion marks still requires a reproducible representation. However, the notes themselves recognise that the UK model is designed for finite, pre-defined variations and therefore cannot simply be transplanted to open-ended AI generation.


The technological literature is comparatively thin. Duolingo's commercial deployment of AI-driven mascot interactions establishes that continuously generated brand-persona manifestations are no longer merely hypothetical. Yet the notes identify no Indian authority directly addressing AI-generative brand mascots. This is the gap the paper fills: rather than treating AI mascots simply as another non-traditional mark, it asks whether Indian distinctiveness doctrine can recognise a stable commercial identity despite controlled variability in manifestation, and develops a bounded-variability framework to address that problem.


METHODOLOGY

This paper adopts a doctrinal and comparative legal research methodology, supplemented by a limited analysis of contemporary AI-driven commercial practice. The doctrinal analysis examines the Trade Marks Act, 1999, particularly Sections 9(1), 2(zb), 47 and 57, together with the relevant provisions of the Trade Marks Rules, 2017. This paper analyzes Indian judicial authorities including Godfrey Phillips India Ltd. v. Girnar Food & Beverages (P) Ltd. and Imperial Tobacco Co. of India Ltd. v. Registrar of Trade Marks to examine how Indian law understands distinctiveness, acquired distinctiveness, identifiable subject matter and continued use.The comparative analysis takes into account the series-mark framework of the United Kingdom under Section 41 of the Trade Marks Act 1994 and the approach of EUIPO to motion marks. These jurisdictions are points of reference for comparison, not transplantation. Especially to observe the accommodation of controlled variation in trademark law. Finally, the paper provides a technology case study of AI-generative virtual mascots, with a factual foundation from Duolingo’s AI-powered mascot interactions to demonstrate that such commercial applications are no longer hypothetical. The analysis ultimately evaluates whether Indian trademark doctrine can accommodate a stable commercial identity alongside controlled variability, and proposes a bounded-variability framework to address the resulting doctrinal gap.


CONCEPTUAL FRAMEWORK– AI GENERATED VIRTUAL BRAND MASCOTS 

An AI generated virtual brand mascots may be understood as digitally created or digitally modified fictional persona, character avatar or anthropomorphic  representation that is generated, substantially developed, or continuously operated through Artificial intelligence and is used by business to identify , promote,  advertise, or communicate to the client and customer for delivering  the goods and services. The legal significance of such mascots lies not merely in their digital form, but in their capacity to function as recurring commercial identifiers while  their individual manifestation may be generated a new thought. AI generated mascots depart from the traditional model. Its visual presentation, tone, or conversational content is not stored and replayed but produced a new, in real time, for each individual  interaction.  Although  trademark law has progressively  accommodated  signs  such as sound  shapes colours and moving images  the principal concerned  remains whether the signs  can be  sufficiently represented moving images  and multimedia marks as forms non-traditional trademark and has identity  representation and description as important issues in their protection. The Trade Marks Act, 1999 defines a trademark by reference to its representability and capacity to distinguish goods or services, while the Trade Marks Rules, 2017 prescribe specific representational requirements for certain non-traditional marks, including sound marks. Duolingo’s  introduction of AI driven ‘Video call with lily’ and Adventures features. These features allow the motion marks, natural tone and conversational tone to the brand mascots which are dynamic in nature.  A single fixed sequence registered and reproduced identically on each use. Mascots whose commercially  manifestation  is generated  pre interaction rather than reproduced  from the fixed temples  that this paper addresses the distinction and its protection through legal framework  it matters as trademark distinctiveness  requirement  was developed entirely against the first category. This raises a question that conventional trademark doctrine does not expressly answer: whether distinctiveness should be assessed by reference to a fixed representation of the mascot, or by reference to the stable commercial identity that persists across its variable AI-generated manifestations. This paper argues that the latter question requires a more nuanced legal framework because the economic function of the mark may remain constant even when its individual manifestations are not. 


THE ARCHITECTURE OF SECTION 9 DISTINCTIVENESS

The first objection is the definition ‘ consistent commercial impression’ may require associative recognition rather than visual sameness  consumers  may identify a mascot by identity, not appearance , notwithstanding continuous  variation in form. This answered. Though not conclusively, by the doctrine's own structure. Godfrey Phillips India Ltd vs. Girnar Food and Beverages Ltd ties secondary meaning  is  an ascertainable sign and Imperial Tobacco Co. of India Ltd. v. Registrar of Trade Marks frames acquired distinctiveness as built through sustained consistent use of that sign over time  through which a mark comes to identify the goods or service  of a particular undertaking. These authorities do not establish that a mark  must  remain visually identical every instance they do, however support the proposition that there must be some identifiable subject matter in relation to which  distinctiveness is acquired . The requirement of a defined representation at the registration stage is consistent with this understanding, although it is not conclusive proof that Indian law necessarily excludes  a more associative conception of distinctiveness; the precise question whether its individual  manifestation continuously very has not yet been squarely addressed by Indian courts. 


The second objection is empirical. No reported Indian decision appears to have considered an AI generative brand mascot in the precise circumstances examined by this paper; the argument advanced here is therefore necessarily anticipatory. This limitation should not however be confused  with  lack  of legal significance; the absence of litigation may simply indicate that the technology has developed faster than the disputes capable of bringing the underlying doctrinal problem before a court. AI mediated commercial personas are already being deployed in consumers facing environments, demonstrating that the factual premise of the problem is no longer  hypothetical. The  contribution  of this paper is accordingly not to claim  that Indian  courts  have already recognised  a doctrinal conflict, but to identify a potential conflict between an existing  registration framework and an emerging form of commercial identity before that conflict is resolved through fragmented litigation. 


The third objection concerns the use of the United Kingdom’s series-mark mechanism as a comparative reference. Section 41 of the Trade Marks Act 1994 permits registration of a series of marks where the marks resemble each other in their material particulars and differ only in matters that do not substantially affect their identity. At first sight, however, the analogy appears imperfect. A series mark concerns a finite and identifiable group of variations that an applicant can specify at the time of registration. An AI-generative mascot may produce an effectively unlimited range of outputs, many of which cannot be predicted in advance. The difference is significant. The proposal advanced in this paper therefore does not seek to transplant the UK mechanism into Indian law. Rather, the relevance of the UK approach lies in a narrower principle: trademark protection need not necessarily depend upon absolute identity of every manifestation where the law can identify a sufficiently stable core and define the permissible limits of variation. Part VI develops this principle into a proposed bounded-variability framework adapted specifically to AI-generative marks.


A fourth objection is institutional. It may be argued that legislative reform is unnecessary because the Trade Marks Registry could address the problem through examination guidelines or administrative practice. Such an approach would undoubtedly be more immediate and less demanding than statutory reform. It would also allow the Registry to develop practical standards in response to technological developments without waiting for Parliament to intervene. The difficulty, however, is that administrative guidance cannot safely substitute for a statutory standard where the underlying question concerns the operation of Section 9 itself. Guidance that attempts to accommodate generative marks without a clear statutory basis may provide limited certainty to applicants and could remain vulnerable to challenge before appellate authorities or courts. A stronger version of the same objection is that Indian courts have previously interpreted statutory concepts such as “use” with sufficient flexibility to accommodate changing commercial practices. It is therefore possible that the existing statutory framework could evolve through adjudication rather than legislation. That possibility should not be dismissed. The argument for a bounded-variability framework is instead one of legal certainty: relying entirely on case-by-case judicial development would leave applicants to discover the permissible boundaries through successive disputes, whereas a defined framework could establish those boundaries prospectively.


The consequences of leaving the problem unresolved also extend beyond the initial registration question. If a dynamically generated mascot is registered on the basis of a particular representation, a later divergence between the registered representation and the manner in which the mascot is actually deployed could raise questions concerning the scope and continuity of use. Depending on the circumstances, this may intersect with the statutory provisions governing the duration and vulnerability of registered marks, including Sections 25 and 47. The precise consequences, however, remain fact-dependent and should not be overstated in the absence of Indian jurisprudence dealing with AI-generative marks. What can be said with greater confidence is that the absence of a framework leaves uncertainty at precisely the point where AI-generated commercial identities depart from the assumptions underlying conventional trademark registration.


The problem is not that artificial intelligence has made the current idea of trademark uniqueness no longer important and it is also not true that every different thing created by an AI system should be considered as something that can be protected by a trademark. The narrower problem is that the existing framework does not clearly explain how distinctiveness should be assessed where the commercial identity remains stable while its individual manifestations are deliberately variable. The bounded-variability approach proposed in Part VI responds to this gap by separating the protected identity of the mark from the permissible range of its manifestations. Its objective is not to relax distinctiveness, but to identify a different way in which distinctiveness may be demonstrated where technological design makes complete visual or textual consistency impossible.


The Bounded-Variability Framework- Separating Notice from Proof

The analysis has shown that the challenge facing AI-generative brand mascots is not that they do not have a stable commercial identity but that Indian trademark law currently has no way of recognizing that identity on its own without a single unchanging image. The framework proposed here solves this by splitting two roles that a traditional trademark registration currently handles together: the role of telling others what is protected and the role of proving that the mark has become distinctive. In the system one registered image has to do both things at the same time. It is what others check against and it is the image that must be shown to have been used consistently to prove secondary meaning under Godfrey Phillips India Ltd. V. Girnar Food & Beverages (P) Ltd. This works fine for a fixed mark, where one image can do both things. It does not work for a mascot that changes every time it is used because no single image can properly tell others what is protected and also show all the ways the mark has become distinctive.


This paper suggests that these two roles should be kept separate. For the purpose of notice an applicant would need to say what the fixed commercial core is at the time of registration: a set name, an image and a written statement of the mascot's main characteristics in terms of how it behaves or sounds. This declared core does the same job as a specimen does today. It is what others look at to know what must be avoided and it is what an examiner or court uses when checking for infringement. Importantly this core does not. In the case of a generative mascot cannot include every single use of the mark; it only includes the parts that stay the same across all uses. For the task of proving that the mark has become distinctive the applicant would be allowed to use evidence from all the different ways the mascot is used. The pattern of use across all its forms. Instead of being limited to showing that the declared core was used alone. This is similar to what's already allowed in Imperial Tobacco Co. Of India Ltd. V. Registrar of Trade Marks: the law needs proof that consumers have come to connect a thing with a certain source; it does not need every use to look exactly the same, only that the thing that becomes distinctive can be recognized. The bounded-variability framework does not change this rule. It just moves where the rule applies from a fixed image to a declared and limited set of stable features.


This framework is intentionally narrower than copying the United Kingdom’s series-mark system for the reasons already discussed in Part 5. Section 41 of the UK Trade Marks Act 1994 was made for a number of variations that the applicant lists in advance and its own manageability under the "material particulars" rule has been hard to handle in practice. The United Kingdom Intellectual Property Office has said it plans to stop accepting series-mark applications citing inconsistent results and legal confusion as key issues. A framework for AI- marks that copied this structure. Requiring the applicant to list variations in advance. Would bring in the same manageability problems that made the UK stop using it and it would do so in a setting of endless unlisted variations that the system was never meant for. The framework proposed here avoids this problem by declaring the stable core, not a list of possible outputs; the limit of protection is defined by what has to stay the same, not by a full list of what can change.


The key idea in this framework is the material-deviation threshold: the point where a change to the mascots form's no longer a normal variation and becomes a new sign that needs its own protection. Normal changes. Differences in how the mascot looks, acts or what is said in an interaction. Are still covered by the registered core as long as the features declared at registration stay the same. Changing the name, the main image or the characteristics listed at registration would be a material change making the new form not covered by the registration. This threshold does the job as a fixed image does for a traditional mark but it uses the features that are declared and stable instead of looking at every single use. Examination by the registry under this system would focus on two things: whether the declared core is distinct enough to meet Section 9 and whether the evidence shows that the core has been used consistently across all the ways the mascot appears. The registry would not need to check or approve every AI-generated version as a condition of registration or renewal; the owner would instead be expected to keep evidence of use that can show if challenged under Sections 25 or 47 that the declared core has been used in a way even though its individual forms change.


The goal of this framework is not to make it easier to prove that a mark is distinctive nor to protect things. Like friendliness, humor or common gestures. That is not themselves special enough to show where a product comes from. It is to find the right thing to look at when the technology used for a mark makes it impossible to have the look every time. The framework does not treat the AI- mascot as an exception to Section 9. It sees it as a category that still needs to follow the rules in Section 9. That consistent use of something that can be recognized is what proves distinctiveness. It just needs a way to collect evidence and register the mark than the one used for fixed signs.


ANTICIPATED OBJECTIONS AND SAFEGUARDS

The proposed bounded-variability framework may face the objection that allowing variation could make the scope of a trademark uncertain. If an AI mascot can continuously change its appearance or movement, competitors may find it difficult to determine what is actually protected. This concern can be addressed by requiring the applicant to identify a fixed commercial core and clearly specify the permissible range of variation. This preserves the certainty required for trademark registration while accommodating technological change.


A second objection is that the concept of commercial gestalt may become too broad and could result in protection over ordinary characteristics such as friendliness, humour, smiling or waving. The safeguard should therefore be that generic behaviour or personality traits cannot independently constitute the protected subject matter. Protection should depend upon the distinctive combination of features through which consumers identify the particular commercial source.


A further objection concerns continuous changes to the AI system. A proprietor might argue that every new manifestation is merely another variation of the registered mascot. To prevent this, the proposed material-deviation threshold should distinguish ordinary variation from substantial transformation. Changes to the mascot's essential visual identity, name or distinctive characteristics should be treated as material and should not automatically fall within the original registration.


CONCULSION

This paper started with a simple question: can Indian trademark law protect a brand mascot that never looks or sounds exactly the same twice? The answer, on the reading developed here, is yes  but not in the way most people would first assume.


Section 9 does not actually say that a mark must stay visually identical every time it is used. That requirement is something courts and the Registry have built up through practice, applying the section to marks that happened to be static. Nobody wrote it into the statute because, until recently, nobody needed to think about a mark that changes on its own. AI-generated mascots like Duolingo's Lily have changed that. The company itself designs Lily around a fixed personality and backstory, while letting every single conversation come out differently. That is not a legal problem for Duolingo right now — but it becomes one the moment someone tries to register that kind of mascot as a trademark, or the moment a competitor copies its personality without copying any single output exactly.


This paper rejected the obvious fix — copying the UK's series-mark system — because that system was built for a small, fixed list of variations decided in advance, and even that limited version of "variation" is proving too hard to manage. The UK is withdrawing it. A mascot that generates something new in real time, every time, is a much harder case than anything the UK system was ever designed to handle. Borrowing a doctrine that is already failing at an easier problem was never going to solve a harder one.


Instead, this paper argues that protection should follow what this paper calls the mascot's commercial gestalt: the stable identity (its name, its core look, its personality) that stays the same even though every individual appearance is different. A person's handwriting is never identical twice, but it is still recognisably theirs. The law should be able to treat a well-designed AI mascot the same way: prove the stable pattern behind it, not every single output it has ever produced. What goes on the trademark register is a short, fixed description of that stable pattern enough for a competitor to know what to avoid. What gets used as evidence to prove the mascot deserves protection can draw from its full range of appearances.


This also quietly fixes a second problem. Once a mascot is registered this way, a business does not have to worry about losing its trademark just because the AI produced an output slightly outside some pre-approved list. As long as the core identity, the name, the look, the personality stays consistent, the mark stays protected, no matter how many different things the AI actually says or does.


None of this has been tested in an Indian courtroom yet. No case has come up, and this paper does not pretend otherwise. But the technology is already here, already commercial, and already being used by companies with real trademark value at stake. The gap this paper identifies is not hypothetical — it is a gap the law will have to close eventually, and it is better closed by a clear framework now than by a messy dispute later, decided under a law never built with AI mascots in mind.


One thing this paper does not fully resolve: what happens when the mascot's core identity itself changes over time — a brand refresh, a personality update, a new model version? That question is left for future work, but it is worth flagging here, because it is the next place where this same tension between "fixed law" and "changing AI" is going to show up again.


REFERENCES

II. Legislation

  • The Trade Marks Act, No. 47 of 1999, §§ 9(1), 25, 47 (India).

  • Trade Marks Act 1994, c. 26, §§ 41 (UK).

II. Cases

  • Colgate Palmolive Co. v. Anchor Health & Beauty Care Pvt. Ltd., 2003 (27) P.T.C. 478 (Del.) (India).

  • Godfrey Phillips India Ltd. v. Girnar Food & Beverages (P) Ltd., (2004) 5 S.C.C. 257 (India).

  • ICC Development (International) Ltd. v. Arvee Enterprises*, 2003 (26) P.T.C. 245 (Del.) (India).

  • Imperial Tobacco Co. of India Ltd. v. Registrar of Trade Marks, A.I.R. 1977 Cal. 413 (India).

III. International and Institutional Materials

IV. Corporate and Industry Materials

V. Online / Media Sources








Comments

bottom of page