Introduction

The French Competition Authority (FCA) seeks stakeholder feedback on its market study on conversational agents and agentic commerce[1]. After a series of market studies by competition authorities worldwide on the upstream segments of the Artificial Intelligence (AI) value chain, focusing on the key inputs required to develop AI models and applications, this study turns, for the first time, to the downstream level[2]. It examines the dynamics of competition in conversational agents and agentic commerce, which are emerging with AI agents.

These agents, such as OpenAI ChatGPT Agent, can act on behalf of users with minimal human intervention. While still emerging, these agents are already reshaping market dynamics and market structures, shifting the economics of digital markets from human to agent-based interactions.

This submission provides a high-level analysis focused on the core questions raised in the public consultation and complements prior work on recent market developments in the sectors[3].

Conversational Agents

The FCA invites feedback on market conditions, vertical integration strategies, monetisation models, publisher partnerships, and platformisation.

Market Conditions

A conversational agent is a user-facing interface that enables end users to interact with AI models in natural language and complete tasks, such as discovering or purchasing products, within the interface itself. Unlike traditional chatbots, which merely sustain conversations, or virtual assistants, which process simple queries to access services or connected devices, conversational agents increasingly execute tasks directly. In practice, however, these distinctions are narrowing as chatbots and virtual assistants incorporate agentic capabilities.

A diverse range of actors now offers conversational agents. These include standalone answer engines such as Perplexity; general-purpose generative AI (GenAI) applications such as OpenAI ChatGPT; and established search engines such as Google Search and Microsoft Bing. They differ primarily in distribution channels, the scope of their agentic functionalities, and their partnerships with third parties.

Although many providers operate globally, certain features remain geographically limited. For example, OpenAI Instant Checkout is currently available only in the United States[4]. Providers may expand these features internationally over time. Commercial strategy often explains phased rollouts, but regulatory constraints can also delay deployment. In France, for instance, Google’s AI features, including AI Overviews and AI Mode, remain unavailable due to regulatory concerns stemming from a dispute over related rights for news publishers’ content[5].

Usage patterns remain difficult to assess due to limited public data. Users may rely on different providers for tasks such as search, editing, coding, or shopping. Ultimately, user choice depends on functionality, reliability, and integration with other services.

Entry and expansion conditions vary. As noted by market studies on the upstream level, developing conversational agents requires access to critical inputs such as data, models, and computing capacity. The degree of openness in access to these inputs significantly affects entry barriers. Open-source models, public datasets, and interoperable standards may lower these barriers.

Distribution also shapes competitive dynamics. Providers may rely on proprietary channels or third-party gateways, such as messaging services. In its ongoing investigation concerning Meta AI’s integration into WhatsApp, the European Commission has highlighted WhatsApp’s role as a key entry point for general-purpose AI assistants. The Commission has preliminarily concluded that restricting third-party AI assistants’ access to WhatsApp risks raising barriers to entry and expansion, potentially marginalising smaller competitors. It is therefore considering interim measures[6]. The Italian competition authority has already imposed such measures, requiring WhatsApp to maintain access for third-party AI assistants[7].

Vertical Integration

Several conversational agent providers also operate adjacent services, including search engines and e-commerce marketplaces. They may offer conversational agents as standalone services, such as Google Gemini, or integrate them into existing services, such as Amazon’s Rufus within its marketplace.

The competitive effects of vertical integration depend on context and thereby require a case-by-case analysis. Integration may enhance efficiency by improving user experience. Conversely, it may foreclose rivals through tying, bundling, or self-preferencing. For example, the Italian proceedings concerning Meta AI examine whether pre-installing Meta AI within WhatsApp constitutes an abuse of dominance by favouring Meta’s own service[8].

Developers of conversational agents often seek access to third-party data, software, or hardware, including operating systems and connected devices. In the absence of dominance or regulatory obligations, third parties generally retain discretion over such access. Under Article 6(7) of the Digital Markets Act (Regulation (EU) 2022/1925, DMA), designated gatekeepers for the provision of core platform services (CPSs) must provide access to certain software and hardware features free of charge. The Commission has already opened specification proceedings against Google to ensure that third-party AI developers can access Android-controlled AI features[9]. However, mandatory free access may weaken investment incentives if firms cannot recover development costs through direct access fees, as entitled by intellectual property rights.

Monetisation

Providers currently rely on four main revenue streams. Under an Application Programming Interface (API) model, providers license access to their underlying models on a usage basis. Under subscription models, business users and individuals pay recurring fees for premium features. Under advertising-based models, providers monetise attention through ad placements. Under commission-based models, providers offering checkout features charge brands when users complete purchases.

In the short term, these models help recoup development and deployment costs, especially training and inference costs, as well as investment in AI infrastructure. Over time, new models may emerge. Providers may provide application marketplaces, such as OpenAI GPT Store[10], and charge developers enrollment or commission fees, mirroring traditional app store models. They may also develop proprietary hardware and monetise conversational agents indirectly through device sales. OpenAI, for instance, is already working on a device following its hiring of Johny Ive[11].

Advertising models remain experimental. Some providers, such as Google or Microsoft, insert ads within generated answers. Others are testing more interactive formats. OpenAI is piloting conversational advertising in the United States, enabling users to engage directly with ads by asking follow-up questions[12]. Google is testing direct offers in AI Mode, which allows advertisers to present exclusive offers. It is also deploying business agents, which allow users to interact with brands in their own voice[13]. Given the early stage of experimentation, any assessment of competitive risks is necessarily premature and therefore out of scope for this paper.

Partnerships

Providers have entered various non-exclusive partnerships. Some collaborate with news publishers, such as OpenAI with Le Monde and Prisa Media, to access publishers’ content to train their models and cite content in responses[14]. Others partner with brands, such as OpenAI with Instacart, to enable integrated shopping experiences[15]. Industry-wide initiatives, such as Google’s collaboration with Shopify, Etsy, Wayfair, Target, and Walmart to develop the Universal Commerce Protocol (UCP), aim to establish open standards for agentic commerce[16].

Such partnerships can promote new revenue streams for publishers and brands and provide standards that enable follow-on innovations. However, exclusive arrangements may restrict rivals’ access to essential inputs. Standard-setting initiatives also require careful governance to prevent the risks of exclusionary abuses or anticompetitive agreements that limit competition and innovation[17].

Platformisation

Some conversational agent providers now offer intermediary services, such as app stores, advertising, and embedded commerce, that connect business users with end users. Through network effects, these services may reinforce as more users adopt them, strengthening the platform’s overall value.

To succeed, providers must overcome the classic coordination challenge of attracting user groups. A substantial end-user base may therefore become a prerequisite for a platform to emerge. At this early stage, competition risks may arise if providers leverage dominance in adjacent markets to promote their platform services.

Agentic Commerce

The FCA also seeks feedback on developments in agentic commerce.

Agentic commerce refers to AI agents that autonomously perform shopping tasks on behalf of users, including product discovery and purchasing. Although still nascent, multiple actors participate in this ecosystem: search engines (e.g., Google Search), e-commerce marketplaces and platforms (e.g., Amazon, Shopify), payment providers (e.g., Visa, Mastercard), financial intermediaries (e.g., Stripe, PayPal), and conversational agent providers (e.g., OpenAI).

As a natural evolution of e-commerce, agentic commerce spans use cases across grocery, travel, retail, and entertainment. A Morgan Stanley report estimates that, by 2030, agentic commerce could account for between USD 190 and 385 billion in US e-commerce spending, accounting for approximately 10–20% of the market[18].

As of February 2026, deployment in France lags behind the United States. For instance, OpenAI offers shopping assistance in France, but not checkout functionality, and Google does not provide its agentic commerce features. Strategic rollouts, regulatory constraints, such as DMA self-preferencing obligations, and technical readiness with French actors may all contribute to these delays.

Agentic commerce may reshape competitive dynamics for intermediaries and sellers alike.

E-commerce marketplaces face competitive pressure from AI agents that replicate core buyer-facing functionalities, such as product discovery and purchasing. However, agents do not necessarily replicate logistics and fulfilment services that are important for sellers. Market definitions in e-commerce may therefore evolve to consider providers of AI agents. In response, incumbents are integrating agentic features. Shopify enables merchants to sell through AI channels[19]. Amazon allows US users to purchase products from non-Amazon merchants via its app[20].

For sellers, AI agents may become a critical access point to consumers. Conversational agents currently direct traffic to sellers’ websites. For instance, according to Digiday, OpenAI ChatGPT accounts for 20% of Walmart’s referral traffic[21]. However, agentic commerce increasingly shifts transactions into agent-mediated environments. In this context, sellers may need to adapt their commercial strategies to ensure visibility and effective participation in these environments[22].

From a technical perspective, agentic commerce relies on interoperability among merchants, payment systems, and agents. Initiatives such as the Universal Commerce Protocol (UCP), the Agent Payments Protocol (AP2), Agent2Agent (A2A), and the Model Context Protocol (MCP) aim to facilitate interoperability and secure transactions. At this early stage, however, it remains unclear whether any de facto standards will emerge[23].

Competition concerns in agentic commerce largely mirror established digital market risks. The OECD identifies risks in downstream AI markets, including algorithmic collusion, leveraging, foreclosure, personalised pricing, and predatory pricing[24].

AI agents nevertheless introduce new analytical challenges. They may autonomously monitor and adjust competitive parameters, such as prices, potentially establishing and stabilising collusive outcomes without explicit coordination. Detecting such conduct and assigning liability may prove complex for competition authorities. Similarly, agents may engage in exclusionary practices, such as favouring affiliated products, without explicit human instruction, complicating assessments of intent and liability.

Upcoming Developments

Conversational agents and agentic commerce remain in an early, dynamic phase. Rapid innovation, market entry, and experimentation characterise the sector. As adoption increases, new use cases will likely emerge, accelerating the transition toward a full-fledged “agentic web”, whereby agents, not humans, interact with other agents and content[25].

The regulatory framework will significantly shape these developments. Existing European competition law provides tools to address traditional forms of human conduct. However, autonomously designed and implemented practices by agents may pose challenges for detection and enforcement.

The DMA already applies when conversational agents or agentic features are integrated into designated CPSs. While it can address certain risks, such as pre-installation, certain obligations, such as data-combination restrictions and self-preferencing rules, may affect gatekeepers’ development and deployment of AI applications.

The study takes place at a time when the debate over the DMA’s scope and applicability to AI services is ongoing[26]. Some argue that AI falls within existing CPS categories; others advocate for a new CPS category. Regarding AI agents, Professors Bostoen and Krämer have examined whether they could fall within the category of virtual assistants[27]. Stakeholders also diverge on the appropriate level of intervention, ranging from strict application of existing obligations to caution against premature regulation[28].

Ultimately, regulatory intervention should respond to demonstrable market failure, such as durable market power that harms competition. At this stage, rapid innovation and competitive intensity suggest that such failure has not yet materialised. Premature intervention risks constraining nascent market developments, thereby distorting the competitive process.

Competition authorities can nonetheless promote market developments. They can conduct market studies to monitor evolving dynamics, enforce swiftly, including through interim measures when competition harm occurs, issue guidance to enhance legal certainty, identify regulatory barriers, and ensure regulatory coherence across regulatory frameworks[29].

[1] The Autorité Launches a Public Consultation on Conversational Agents, Autorité de la concurrence, 29 January 2026 (accessed 19 February 2026). Available at: https://www.autoritedelaconcurrence.fr/en/article/autorite-launches-public-consultation-conversational-agents

[2] For a list of market studies, see Christophe Carugati, Generative Artificial Intelligence, Digital Competition (accessed 19 February 2026). Available at: https://www.digital-competition.com/generativeai

[3] Christophe Carugati, Why Does France Study Conversational Agents and Agentic Commerce?, Digital Competition, 27 January 2026 (accessed 19 February 2026). Available at: https://www.digital-competition.com//comment/why-does-france-study-conversational-agents-and-agentic-commerce

[4] Buy it in ChatGPT: Instant Checkout and the Agentic Commerce Protocol, OpenAI, 29 September 2025 (accessed 19 February 2026). Available at: https://openai.com/index/buy-it-in-chatgpt/

[5] Aymeric Geoffre-Rouland, La France Snobe Google : Les “Résumés par IA” de la Recherche Google se Déploient Partout, Sauf Chez Nous, Les Numériques, 8 October 2025 (accessed 19 February 2026). Available at : https://www.lesnumeriques.com/intelligence-artificielle/l-ai-mode-de-google-deferle-sur-le-monde-sauf-en-france-ou-les-resultats-de-recherche-resistent-aux-reponses-toutes-faites-n243689.html

[6] Commission Notifies Meta of Possible Interim Measures to Reverse Exclusion of Third-Party AI Assistants from WhatsApp, European Commission, 9 February 2026 (accessed 19 February 2026). Available at: https://ec.europa.eu/commission/presscorner/detail/en/ip_26_310

[7] A576 - Meta AI: The Italian Competition Authority Orders Meta to Suspend the Terms Excluding Competing AI Chatbots From WhatsApp, Autorità Garante della Concorrenza e del Mercato, 24 December 2025 (accessed 19 February 2026). Available at: https://en.agcm.it/en/media/press-releases/2025/12/A576

[8] A576 - The Italian Competition Authority Launches Investigation into Meta Over Abuse of Dominant Position, Autorità Garante della Concorrenza e del Mercato, 30 July 2025 (accessed 19 February 2026). Available at: https://en.agcm.it/en/media/press-releases/2025/7/A576

[9] Commission Opens Proceedings to Assist Google in Complying with Interoperability and Online Search Data Sharing Obligations Under the Digital Markets Act, European Commission, 27 January 2026 (accessed 19 February 2026). Available at: https://ec.europa.eu/commission/presscorner/detail/en/ip_26_202

[10] Introducing the GPT Store, OpenAI, 10 January 2024 (accessed 23 February 2026). Available at: https://openai.com/index/introducing-the-gpt-store/

[11] A Letter from Sam & Jony, OpenAI, 9 July 2025 (accessed 23 February 2026). Available at: https://openai.com/sam-and-jony/

[12] Our Approach to Advertising and Expanding Access to ChatGPT, OpenAI, 16 January 2026 (accessed 23 February 2026). Available at: https://openai.com/index/our-approach-to-advertising-and-expanding-access/

[13] Vidhya Srinivasan, New Tech and Tools for Retailers to Succeed in an Agentic Shopping Era, Google Blog, 11 January 2026 (accessed 23 February 2026). Available at: https://blog.google/products/ads-commerce/agentic-commerce-ai-tools-protocol-retailers-platforms/

[14] Global News Partnerships: Le Monde and Prisa Media, OpenAI, 13 March 2024 (accessed 23 February 2026). Available at: https://openai.com/index/global-news-partnerships-le-monde-and-prisa-media/

[15] Instacart And OpenAI Partner on AI Shopping Experiences, OpenAI, 8 December 2025 (accessed 23 February 2026). Available at: https://openai.com/index/instacart-partnership/

[16] Universal Commerce Protocol (accessed 23 February 2026). Available at: https://ucp.dev/latest/

[17] OECD, Standard Setting, Series Roundtables on Competition Policy, 2010 (accessed 16 January 2026). Available at: https://www.oecd.org/content/dam/oecd/en/publications/reports/2011/03/standard-setting_24c2823d/8b423811-en.pdf 

[18] Here Come the Shopping Bots, Morgan Stanley, 8 December 2025 (accessed 14 January 2026). Available at: https://www.morganstanley.com/insights/articles/agentic-commerce-market-impact-outlook

[19] The Agentic Commerce Platform: Shopify Connects any Merchant to Every AI Conversation, Shopify, 11 January 2026 (accessed 28 January 2026). Available at: https://www.shopify.com/news/ai-commerce-at-scale

[20] Amazon’s New 'Buy for Me' Feature Helps Customers Find and Buy Products from Other Brands’ Sites, Amazon Blog, 3 April 2025 (accessed 28 January 2026). Available at: https://www.aboutamazon.com/news/retail/amazon-shopping-app-buy-for-me-brands

[21] Allison Smith, ChatGPT is Now 20% of Walmart’s Referral Traffic — While Amazon Wards Off AI Shopping Agents, Digiday, 25 September 2025 (accessed 23 February 2026). Available at: https://digiday.com/media/chatgpt-is-now-20-of-walmarts-referral-traffic-while-amazon-wards-off-ai-shopping-agents/

[22] For a deeper discussion on the impact of AI agents on a mock e-commerce, see Amine Allouah et al., What Is Your AI Agent Buying? Evaluation, Implications and Emerging Questions for Agentic E-Commerce, 2025 (accessed 26 February 2026). Available at: https://arxiv.org/abs/2508.02630

[23] For a deeper discussion of the technical aspects, see Christophe Carugati, Why Does France Study Conversational Agents and Agentic Commerce?, Digital Competition, 27 January 2026 (accessed 19 February 2026).

[24] OECD, Artificial Intelligence and Competitive Dynamics in Downstream Markets, 1^st December 2025 (accessed 24 February 2026). Available at: https://www.oecd.org/en/events/2025/12/artificial-intelligence-and-competitive-dynamics-in-downstream-markets.html

[25] Yingxuan Yang et al., Agentic Web: Weaving the Next Web with AI Agents, 2025 (accessed 24 February 2026). Available at: https://arxiv.org/abs/2507.21206

[26] Consultation on the first review of the Digital Markets Act, European Commission (accessed 24 January 2026). Available at: https://digital-markets-act.ec.europa.eu/consultation-first-review-digital-markets-act_en

[27] Friso Bostoen and Jan Krämer, Is the DMA Ready for Agentic AI?, CERRE, 3 July 2025 (accessed 24 February 2026). Available at: https://cerre.eu/publications/is-the-dma-ready-for-agentic-ai/

[28] Consultation on the first review of the Digital Markets Act, European Commission (accessed 24 January 2026).

[29] For a study on regulatory coherence, see Europe’s Progress in the Digital Single Market: A Proposal for Consistency, Digital Competition, 17 February 2025 (accessed 24 February 2026). Available at: https://www.digital-competition.com/comment/europe-progress-in-the-digital-single-market-a-proposal-for-consistency