When the Market Speaks: Why the 2024 Market Definition Notice Is Not Yet Fit for Language-Centric Systems

“Alexa, I have to ebook a physician’s appointment.”

The assistant solutions instantly with a single suggestion, voiced as a pleasant suggestion somewhat than as one choice amongst many. The consumer hears only one selection. No checklist of options, no rating, no seen logic. The interplay feels conversational and useful, virtually private. But behind this easy trade lies a profound shift in how digital markets perform. More and more, customers work together with the digital world not via screens however via language – asking, responding, clarifying, and counting on conversational methods to mediate entry to info and companies.

The definition of the „related market“ is a central ingredient when assessing mergers and most antitrust circumstances. The European Fee’s 2024 Market Definition Discover (henceforth “the Discover”) supplies steerage on the way in which the European Fee applies this idea. It’s the first revision of the Market Definition Discover since its adoption in 1997 and is mostly seen as a welcome step in the direction of modernising competitors evaluation for digital markets. It embraces ideas equivalent to multi-sided platforms (para 94), knowledge as a aggressive parameter, and zero-price markets (para 97), and recognises that competitors more and more happens via ecosystems (para 99) somewhat than remoted merchandise.

Regardless of these advances, the Discover finally treats massive language fashions (LLMs) – that are the premise for conversational methods like voice assistants or chatbots – as in the event that they have been variants of established digital companies. Whereas this isn’t explicitly said within the Discover, it implicitly treats them as such by inserting LLMs inside broader service classes (e.g., search, working methods, digital assistants, cloud companies) with out acknowledging that language-based intermediation is basically completely different from these companies. As a result of LLMs interact customers via dialog, they will form selections, habits, and behavior in ways in which create new types of dependence and, doubtlessly, market energy. Present market definition instruments are struggling to seize these constructions as a result of they have been designed for markets the place energy stems from costs, switching prices, or management over technical infrastructure – not from steady, personalised linguistic interplay. This interplay generates semantic, pragmatic, and private knowledge that repeatedly adapts the service to the consumer, producing suggestions loops and lock-in dynamics that the present Market Definition framework shouldn’t be designed to detect or assess.

This put up argues that the revised Discover, although modernised in lots of respects, stays ill-equipped to outline markets wherein undertakings compete not via value or interface design however via the phrasing, tone, and behavioural affect embedded in natural-language methods. By treating linguistic interplay as peripheral, the framework dangers inflicting authorities to miss key parameters via which corporations could train market energy. If conversational AI turns into a dominant gateway to digital companies, as current tendencies counsel, then competitors legislation might want to rethink how market boundaries are recognized in environments formed by linguistic interplay somewhat than conventional product options.

Development of the brand new Market Definition Discover

The Discover undoubtedly displays a major effort to adapt competitors legislation to digital realities. It incorporates insights about multi-sided markets (paras 94-95), acknowledges the significance of information, and supplies steerage for zero-price companies (paras 97-98) the place conventional price-based assessments can’t be utilized. It recognises that high quality (paras 23, 27, 30), innovation (paras 15, 23) and privateness (para 15) could signify key dimensions of competitors, and that ecosystem dynamics (para 99) typically decide consumer selection. These additions right most of the limitations of the earlier framework, which was written for conventional offline markets the place costs have been clear, switching prices low, and substitution patterns comparatively simple.

But, even these enhancements proceed to depend on assumptions about consumer behaviour that don’t translate effectively to conversational environments. The Discover implicitly assumes that digital companies are evaluated visually: customers see lists, evaluate options, and make selections primarily based on attributes equivalent to performance, rating, or value. This assumption collapses in language-centric markets. A voice assistant typically doesn’t current a set of choices; it synthesises a single reply. A chatbot doesn’t show competing suppliers of the products or companies it suggests; it frames info in narrative and dialogue-based kind. Most massive language fashions don’t provide a menu or search outcomes; they supply a linguistic development that feels tailor-made to the consumer, assured, and authoritative. These methods don’t merely ship content material. They carry out the reply, mirroring a novel trade with an emotional being.

Treating such interplay as functionally equal to screen-based visible interfaces obscures key features of linguistic intermediation. Voice and chat-based methods form behaviour via the way in which they formulate responses – via wording, intonation, pacing, and implied authority. The conversational format encourages belief and reduces cognitive effort, because the consumer is relieved of the necessity to evaluate options. It additionally integrates previous interactions into present ones, forging a way of continuity and familiarity that deepens consumer reliance.

Linguistic gaps within the Market Definition Discover

This hole turns into clearer when analyzing how the Discover approaches its core analytical instruments. The framework depends closely on demand-side substitution (paras 25 et seqq.), which asks whether or not customers would swap to a different service if the focal one worsened. However many customers don’t expertise a swap from Google Assistant to Google Search or from Siri to a browser as a pure substitution, even when each companies finally present info. Interplay via language feels completely different. Asking a query aloud shouldn’t be the identical as opening an app and typing. The interface is embodied via a voice, typically personalised, which strengthens the notion of help or recommendation. Substitution turns into troublesome to explain in financial phrases when the very nature of the interplay is distinct.

Equally, the Discover suggests using SSNDQ (a small however important non-transitory lower in high quality; see esp. para 98) the place financial costs are unavailable. But high quality is very arduous to quantify in conversational methods. What constitutes a decline in high quality: extra hallucinations, much less pure human-like phrasing, slower response instances, inconsistent tone, or lowered contextual understanding? These options have an effect on consumer behaviour in numerous methods, and no established metric exists for assessing them. Linguistic and pragmatic qualities don’t map neatly onto technical efficiency indicators.

Customers typically “pay” for digital companies with their knowledge. Whereas this framing is useful, it treats knowledge largely as a uniform enter. In language-centric ecosystems, nevertheless, language knowledge is uniquely invaluable: it incorporates semantic, pragmatic, emotional, and contextual layers that go far past what clickstream knowledge reveals. It discloses intentions, preferences, habits, and social that means. Customers produce this knowledge via spontaneous dialog, not by filling structured fields. It’s not publicly accessible and can’t be bought in uncooked kind. It could actually additionally not be replicated synthetically at comparable high quality. The Discover’s basic remedy of information doesn’t account for this specificity, nor for the numerous aggressive benefit that high-quality conversational knowledge confers.

Switching prices in conversational ecosystems additionally function in another way from these described within the Discover (see e.g. para 57). Customers discover ways to communicate to a particular assistant – what phrasing works, how the system behaves, and the way it responds to sure requests. Over time, they adapt to the system’s linguistic routines, constructing familiarity and expectations. Switching to a different assistant requires cognitive effort: one should be taught new instructions, alter to a brand new conversational fashion, and re-establish context. These behavioural switching prices are actual, but the Discover primarily focuses on technical or contractual switching limitations (esp. para 57-58).

Even the remedy of community results (paras 94-96) requires additional nuance. Classical knowledge community results are already troublesome to include into market definition, as extra customers generate extra knowledge, enhancing companies and additional attracting customers. In language-centric methods, these results are magnified. Every further consumer doesn’t simply generate knowledge; they generate conversational knowledge that instantly improves the system’s linguistic competence, typically in a personalised method. Efficiency positive aspects come up quickly from rising interplay quantity, creating suggestions loops which might be considerably stronger and extra entrenched than these in typical digital markets. These “linguistic community results” are usually not but addressed within the Discover, though they could decide market energy in AI-driven ecosystems.

This dynamic additionally illustrates why fashions equivalent to ChatGPT and Gemini could not, in observe, be robust substitutes, even when the Discover would place them in the identical product market primarily based on their comparable functionalities. When customers work together with one system, it adapts to their language, preferences and routines, creating behavioural sunk prices that make switching to a distinct mannequin really feel much less personalised or much less efficient. As switching turns into tougher and substitutability weakens, linguistic community results can entrench market energy in methods the present framework doesn’t but recognise.

From actual world examples to idea

The aggressive significance of those dynamics turns into seen in sensible examples: Experiences have indicated that Amazon Alexa tends to favour Amazon-affiliated merchandise in response to client queries, framing these options as conversational recommendation somewhat than as search outcomes. As a result of Alexa normally supplies just one suggestion, the consumer doesn’t see competing choices, and the system’s linguistic framing exerts substantial affect over client behaviour. Google Assistant equally integrates voice-based responses with Google’s personal companies, typically phrased in ways in which current the advice as impartial fact-finding somewhat than preferential routing. And though the CJEU’s 2024 Meta Platforms ruling didn’t contain conversational AI, it underscored the aggressive relevance of cross-platform behavioural knowledge – an perception that applies with explicit drive in linguistic ecosystems, the place behavioural cues are encoded in interplay itself. Equally, the OLG Cologne handled points in regards to the DMA’s ban on “combining” private knowledge in a current nationwide determination. Though this case truly involved massive language AI fashions, linguistic or behavioural features of language knowledge haven’t been addressed within the judgment.

These examples spotlight a recurring theme: conversational interfaces don’t merely mediate entry to digital companies; they form the way in which markets perform. They decide which choices the consumer does or doesn’t see. They affect the framing of selections. They accumulate knowledge that opponents can not match. Such interfaces can create or reinforce market energy, not via exclusionary contracts or value methods, however via the linguistic structure of interplay.

The way in which ahead

The important thing query, due to this fact, is how market definition ought to adapt to this new setting. Importantly, doing so doesn’t essentially require revising the Market Definition Discover. The instruments for a extra behaviourally knowledgeable strategy exist already inside EU competitors legislation; they merely must be utilized with an understanding of linguistic intermediation.

A extra practical strategy would concentrate on how customers truly use conversational methods. The related market could not coincide with conventional classes equivalent to “search engines like google and yahoo” or “digital assistants”, however may as an alternative replicate duties like product discovery, service navigation, or info retrieval. A behavioural perspective would incorporate linguistic switching prices and consumer dependence on personalised dialogue, treating them as related indicators of substitutability. And a extra nuanced understanding of information would recognise that conversational knowledge shouldn’t be merely an enter however a behavioural useful resource with important aggressive implications.

Competitors authorities may additionally place better emphasis on assessing intermediation energy:  Which undertakings management the conversational channel(s) via which customers entry digital markets? Who determines the default pathways for voice-based product discovery or AI-mediated reserving processes? These questions are more and more pressing as a result of the linguistic interface is turning into the primary – and in some circumstances solely – gateway via which customers encounter companies, info, and business selections. As management over this gateway expands, so does the potential for undertakings to steer, form or restrict customers’ decision-making in ways in which conventional competitors legislation instruments battle to seize.

Conclusions

The 2024 Market Definition Discover is a major achievement, however linguistic intermediation exposes conceptual blind spots that the up to date framework doesn’t but resolve. As conversational methods develop into extra built-in into every day life, they are going to more and more function gateways to important markets. Market energy could emerge not from conventional conduct however from management over language itself: the phrasing of responses, the buildup of conversational knowledge, the behavioural lock-in created by steady dialogue.

If market definition fails to replicate these dynamics, downstream assessments underneath Article 102 TFEU or merger management could overlook important sources of aggressive benefit. EU competitors legislation has all the time tailored to technological change; it’s now confronted with a shift from screen-based to language-based market mediation. As this transition accelerates, competitors legislation should be taught to interpret the linguistic alerts reshaping digital markets.

The market is talking. Whether or not the legislation is ready to pay attention stays an open – and more and more pressing – query.

Barbara Justen is a PhD pupil on the College of Vienna, specializing in competitors legislation and new applied sciences.