Gatekeeping in AI Commerce: Google’s Gemini–UCP Ecosystem

I. Introduction

In April 2026, Google enhanced the abilities of its AI assistant, Google Gemini, so that users can discover, compare, and consummate transactions from a single user interface. The main concern is that Google has become an intermediary, vertically integrated in AI-mediated commerce with Gemini as the interface making decisions and Universal Commerce Protocol (“UCP”) as the integration layer, controlling not only merchant eligibility but also their visibility to end-users. This two-fold control puts Google in a structurally favourable position to determine demand and allocation of visibility. In this context, stakes become particularly high considering the estimated projections of a USD 5 trillion global retail market by 2030; even a slight distortion of intermediation can have far-reaching implications for competition.

This article advances its argument in three parts. First, it deals with the delineation of the relevant market as AI-enabled platform intermediation with reference to Sections 19(7) and 19(4) of the Competition Act, 2002 (“Act”). Second, it argues that the interaction designed by Google between Gemini and UCP within the ecosystem could constitute a denial of market access “in any manner,” under Section 4(2)(c). Third, it draws inspiration from the European Union frameworks and accordingly proposes a three-pronged approach to deal with design-based exclusion to re-establish competitive neutrality.

II. AI-Mediated Commerce Market

An investigation into whether Google has abused its dominance under Section 4 of the Act should begin with the delineation of the Relevant Market as per Sections 2(r), 2(s) and 2(t) read in conjunction with Section 19(7) of the Act. Such an analysis should identify the functional nature of the platform and the conditions of competition in which it exists, especially in the digital markets where traditional product-based classifications may not reflect economic realities.

The Competition Commission of India (“CCI”) in Delhi Vyapar Mahasangh v. Amazon Seller Services Pvt. Ltd. has recognised that e-commerce platforms operate as marketplace-based models facilitating transactions between sellers and consumers. The CCI looked at the manner in which such platforms facilitate transactions and determine interactions between market participants, and this provides the starting point in evaluating digital markets. Similarly, Google performs the role of an intermediary by connecting users, merchants, and service providers within its AI-powered ecosystem.

Nonetheless, digital market intermediation is not homogeneous. Relevant Markets can be further delineated into sub-markets as held in Brown Shoe Co. v. United States, using the “practical indicia” test, such as functional characteristics, consumer perception and the particular circumstances of competition. These indicia become very important in this context, where the AI-mediated interface of Google reconfigures the process of intermediation itself, wherein the discovery, evaluation and transaction are built into a single system, and thus changes the way users behave in this ecosystem independent of any competitive constraints exerted by market participants.

Using this framework, Google’s AI-mediated commerce interface is a unique market in the larger space of digital intermediation. Contrary to the conventional platforms where users and sellers interact, the interface of Google’s AI ties together the ability to retrieve information, compare, evaluate and make transactions in one system of interaction. This integration changes the way users interact, and the conditions companies compete in, making Google’s AI non-substitutable with traditional search engines and independent e-commerce platforms. Therefore, this article argues that the relevant market must be delineated as “AI-enabled platform intermediation of the discovery, comparison and facilitation of transactions in goods and services,” in which Google is a structurally integrated provider.

After having identified the Relevant Market, the dominance analysis should be carried out in accordance with Section 19(4) of the Act. Among the factors listed therein, the barriers to entry under Section 19(4)(h) assume importance in digital markets that are characterised by scale, aggregation of data, and technological integration. The foundation of Google’s AI interface is the underlying data infrastructure that enables it to collect and constantly update the market information on a large scale. Google Shopping Graph exemplifies the nature of such infrastructural advantage, whereby it aggregates product information from across the web, which reportedly covers about 50 billion product listings and processes approximately 2 billion updates per hour. This allows Google to keep constantly updated data about prices, stock, and product attributes, which leads to the creation of an informational advantage.

The importance of this infrastructure is not only in its size but in its replicability because it requires continuous inputs of merchants, user interactions, and the computational facility at a scale not available to new entrants. All of these factors create significant entry and growth barriers, protecting Google from effective competition. These effects are reinforced over time, creating a feedback loop which further entrenches Google’s position. The more dependent on its AI-mediated interface, the more Google gathers granular data about users, their preferences, intent, and transaction behaviour. This information adds to the quality and perceived reliability of its outputs, strengthening user dependency and consolidating its role as the main interface to commerce. Hence, Dominance cannot be determined in such markets based on market share alone. Even small initial advantages in the digital markets can result in tipping, as has been noted by the Organisation for Economic Co-operation and Development, before conventional indicators of dominance fully materialise. Market power in the modern context is located in the form of the ecosystem of Google and its technological infrastructure, and is not only quantitative shares.

In this regard, when an enterprise has control over a non-replicable data infrastructure, enjoys strong network effects, and is centrally placed in the organisation of commercial interactions, it has the capacity to act independently of competitive constraints in any meaningful sense. Based on the above-discussed factors, Google meets the following conditions and, thus, enjoys a dominant position in the market of AI-based intermediation of platforms. This dominance and its implications on access to fair participation will be discussed in the next part.

III. Denial Through Traffic Gatekeeping

Section 4(2)(c) of the Act prevents a dominant enterprise from engaging in conduct that results in the denial of market access “in any manner.” This provision should be understood to include not only explicit exclusion like refusal to deal or denial of essential facilities but also structural and interface-mediated forms of foreclosure. The ability to influence visibility and choice becomes a form of gatekeeping in digital markets where access to consumers is mediated through platform-controlled interfaces. The denial of market access happens through technical design, opaque conditions, and conditions that limit effective participation, as was held in Umar Javeed v. Google LLC. In this context, where Google can dictate the processes by which users make discovery and complete transactions, being excluded from this may constitute a denial of market access despite no explicit refusal.

Conventionally, Google’s search engine used to act as a gatekeeper by offering a ranked list of results to the users in a single interface. Although position bias affected user behaviour, the other competing firms were still visible, and users had the option to scroll, compare and assess alternatives displayed on the SERP. Such interaction involved what behavioural economics refers to as System 2 reasoning, which is deliberative, effortful, and comparative, whereby users were able to move beyond the top-ranked results and make independent decisions. Dominance, in that context, was exercised by giving favourable placement but did not eliminate competitors.

This structure is essentially reorganised in AI-mediated interfaces like Google Gemini. Such systems combine discovery, comparison, and recommendation into one interface, instead of ranked lists, providing curated outputs and a limited number of choices as a complete response. Comparative and evaluative functions are carried out in the upstream, shifting user interaction to System 1 processing, which is intuitive, automatic, and acceptance-oriented, where minimal independent verification is done. Accordingly, users would be less inclined to seek alternatives beyond the generated responses, and the interaction will be focused on the options chosen by the AI. Empirical data indicates this change, as reported growth in click-through rates of surfaced results, with about 35% uplift for organic links and 91% for paid placements, and a larger increase in zero-clicks. The ranking in this environment is no longer the determinant of traffic but rather a selection and access to the market based on inclusion in the decision interface.

This transformation also alters the operation of self-preferencing. Earlier cases addressed preferential conduct within a still-contestable interface. In the Google Shopping case, Google was found to promote its own comparison-shopping engine and demote competitors in the search results. Gemini breaks out of this paradigm. Google not only dictates the ranking of search results but also the content of the results, through selective inclusion that favours Google or its partners to the detriment of rivals. The harm that it causes is not comparative disadvantage in a marketplace, but non-existence in the marketplace.

Google has an exclusionary dynamic that is structurally made possible by the UCP, a standardised integration protocol that regulates how merchants present their capabilities to the Google systems. While user interaction is done via Gemini, terms of market access are decided upstream via UCP. Integrated merchants are made technically readable to Google’s AI and thus are included in the generated results. On the other hand, non-integrated merchants or those non-conforming with the UCP requirements have reduced or no presence on Gemini. UCP, in this sense, operates as a gatekeeping protocol layer that determines a pool of entities from which Google chooses and offers to users.

This dynamic is also supported by the integration of such features as “Direct Offers,” through which a participating merchant can incorporate intent-sensitive promotions directly into the AI-generated response, which practically occupies the decision interface at the point of user choice. The aggregation of UCP-based eligibility and Gemini-based selection focuses visibility on a set of participants that are closed and determined by a platform. This structure is reinforced with behavioural dynamics that enhance the dependency of the user on outputs generated. Gemini presents its responses in a way that makes them seem coherent and authoritative, prompting users to perceive them as complete and reliable, disincentivising the need for further research. These features enhance acceptance-based interaction where users are less inclined to find alternatives that are not displayed. Inclusion in Google-generated output can therefore be viewed as endorsement, whereas exclusion forecloses meaningful consideration.

This results in broader implications for competition. With fewer organic avenues to discovery, rival companies have to pay more to gain visibility, be it through paid placement or continuous adjustment to obscured, Google-oriented standards of integration and selection. Smaller or less-endowed companies are disproportionately impacted, lacking the capacity to compete within these constraints. The exclusion of firms from the decision interface of Google Gemini is, in effect, the exclusion of them from any meaningful participation in the market, leading to progressive foreclosure and further reliance on the Google ecosystem to access demand.

To this end, the phrase “in any manner” in Section 4(2) (c) should be construed in such a way as to include these types of structural and interface-based exclusion. Where System 1-driven acceptance replaces System 2-driven evaluation, visibility is dictated by inclusion in the AI interface of Google, and access to demand can be dictated by visibility. The ability that Google has to lock out players in this decision interface, in effect, is the ability to foreclose competition.

IV. Way Forward

a) Restoring the Contestability in AI-Mediated Commerce:

The architecture of exclusion in digital markets has shifted from overt contractual restraints to more subtle, design-driven distortions embedded within platform systems. Although frequently justified by the reasons of efficiency in the short term and better user experience, the decision made by the European Commission in Google Shopping explains that these advantages cannot outweigh long-term harm to competitive structure and market access. Therefore, the main issue is not immediate user gains, but whether interface design systematically undermines rival visibility and contestability over time. Based on this, an ex-ante regulatory framework is needed, and this part suggests a three-pronged approach to deal with design-based exclusion and re-establish competitive neutrality.

b) Mandating Transparency in Selection

Where Google’s AI interface determines which merchants, products, or services are surfaced within its generated responses, the metrics that govern this visibility become decisive in terms of competition. The lack of transparency with respect to these criteria allows opaque and possibly discriminatory practices of inclusion, such as giving preferential treatment to affiliated or integrated organisations.

In response to this, the CCI, under Sections 64(1) and 64(2)(h) of the Act, should require Selection Disclosure Regulations for AI-mediated commerce interfaces, whereby there is transparency in the criteria of inclusion in generated outputs. Based on Article 5(1) of the EU Platform-to-Business (P2B) Regulation, these obligations must include periodic disclosure of: (i) the key parameters that guide selection and inclusion in AI-generated output; (ii) the relative weight that is given to factors like integration, affiliation, or fulfilment capabilities; and (iii) the systematic benefits that the platform or the preferred partners receive in inclusion. These obligations should be ongoing structural obligations and not conditional in the context of an investigation, given the fact that exclusion is reached at the stage of design itself.

c) Ensuring FRAND access to UCP

Google’s UCP is defined as the entry-point that allows merchants to be technically eligible to be included in AI-generated outputs. Structural foreclosure can be practised through discriminatory onboarding or favouritism through participation where such infrastructure happens to define the participation.

To counter this, the law must provide for mandatory non-discriminatory access to integration infrastructure. Through regulations under Section 64, the CCI should require that platforms: (i) post their eligibility criteria clearly and objectively to be integrated into systems like UCP; (ii) provide access to similarly placed merchants on equal terms; and (iii) should not condition their inclusion in AI outputs on their use of ancillary or affiliate services. Moreover, there should be FRAND-like commitments to make sure that the access to integration pathways is equitable, rational, and non-discriminatory. Importantly, the platform should not be allowed to rely on the integration status itself as a factor to preferential inclusion in the outputs of AI, and thus disconnect eligibility and visibility.

d) Enabling Data Portability and Reducing Ecosystem Lock-In

Even where access and selection are explicitly neutral, competitive pressures might be undermined when firms are data-dependent on a single ecosystem. In AI-mediated commerce, platforms gather granular data about the performance of merchants, user interaction, and transaction results, which subsequently influences subsequent inclusion in the outputs produced. This has switching costs that discourage multi-homing and entrench platform power.

To address this lock-in, the CCI needs to implement practices based on the ideas of Article 6(9) of the Digital Markets Act, which presupposes efficient and real-time data portability. It is required that platforms: (i) must allow merchants to export their product data, performance metrics and transaction histories in open and standardised formats; (ii) must be able to integrate with competing platforms using open and standardised APIs; and (iii) portability should not lead to a loss of functionality or analytical capabilities. Such steps would create a less reliant and more competitive firm between platforms.

About the Author

Vashmath Potluri and Nishaan Potluri are students at the National Academy of Legal Studies and Research, Hyderabad

Editor

Suprava Sahu, Senior Editor