Digital Markets Act Workshops: Key Takeaways from Alphabet, ByteDance, and Meta

Megan Kirkwood is a fellow at Tech Policy Press.

Following part one, this summary covers the second set of the European Commission's Digital Markets Act (DMA) compliance workshops, focusing on Alphabet, ByteDance, and Meta. Held between July 1 and July 3, 2025, these sessions provided further insight into how each gatekeeper is adapting to the DMA obligations in its second year of enforcement.

Alphabet

Alphabet opened its workshop by arguing that the DMA is deterring innovation in the EU, delaying the rollout of new technologies, and making EU citizens “second-class” digital citizens. While affirming its commitment to the region, citing seven EU data centers and six subsea cables connecting to Europe, Alphabet noted that some product launches have been delayed due to potential implications under EU digital regulations.

Alphabet representatives stated that they have been forced into unprecedented levels of regulatory engagement, which they argue is not sustainable and stifles innovation, as budgets are redirected toward compliance. They added, “economic growth driven by regulatory compliance is not the sort of economic growth we understand that the EU wants or needs.”

Alphabet also argued that businesses and users are suffering under the DMA. For example, they stated that changes to Google Search listings have led to businesses losing direct bookings and relying more on intermediaries, which may charge a commission. Relatedly, they noted that users are paying higher prices through such intermediaries and have a worse experience finding hotels.

These arguments follow a search results test carried out by Alphabet, which released only a brief blog post rather than detailed findings. Following the workshop, Bloomberg reported that the company is highlighting search results from other shopping and travel platforms at the top of search results pages to avoid fines, after being preliminarily warned that it is self-preferencing its own travel comparison service.

Despite Alphabet’s belief that providing users with increased choice might not automatically result in users acting on this, Alphabet representatives revealed that 82% of Android devices have downloaded an app from outside the Google Play Store, original equipment manufacturers have pre-installed non-Chrome browsers on 70% of EU Android devices, and two-thirds of EU Android devices have a third-party app store downloaded.

AI services and the DMA

Alphabet had to spend a significant amount of time discussing how AI is integrated into its many core platform services. Unlike other gatekeepers, its representatives began by explaining the term “AI,” acknowledging that it is a floating signifier without a definitive meaning. They emphasized that “AI is a technology, not a product,” suggesting it may not make sense to classify AI as a core platform service.

They also described the evolution of YouTube’s recommender system since 2008, moving from simple popularity-based ranking to personalized recommendations and machine learning–based content moderation. Additional examples included Google Play Store’s automated safety checks and policy classifiers, real-time speech translation in Google Meet, and Gmail features that generate email responses. These examples reinforced Google’s view that AI is deeply embedded across technologies and, in many cases, is not a particularly new innovation, making it difficult to regulate in isolation.

When asked if Google’s AI Overview, which is integrated into Google Search, should be subject to switching to other providers to avoid non-compliance by self-preferencing, the representatives argued that its overview should not be viewed as a separate service, but rather as part of the search. However, the line between a core platform service and its integrated services is thin. Arguably, if Google Shopping and Flights services are subject to the DMA, AI Overviews could also be subject to the same regulation.

Data scraping

At the workshop, a representative from a news media coalition raised concerns that AI Overviews compete directly with publishers, many of whom have seen referral traffic drop by 50–60%, impacting publisher revenue. Alphabet denied the claim but declined to release data to counter it.

The representative added that while Overviews rely on publisher content, Alphabet neither seeks consent nor offers a meaningful opt-out, only the option to withdraw from Google Search entirely. Alphabet pointed to robot directives that give publishers control over their content, such as the no-snippet tag, which blocks snippet display and use in AI Overviews and AI Mode, although it does not stop Google from crawling the content.

Finally, the news media representative argued Alphabet acts as both gatekeeper and competitor, yet fails to provide access on fair, reasonable, and non-discriminatory terms, as required under the DMA. Since Alphabet claims Overviews are part of Search, the news media representative argued they should be subject to the DMA’s ban on self-preferencing.

Data consent and data access

Two main issues were raised regarding data consent and data access. First, on Article 5(2), which places further obligations on gatekeepers to seek explicit consent on cross-using and combining data, various questions were asked on consent rates either for AI training data collection or rates of users opting into their data being cross-used over different Google services, such as Google Maps and YouTube. The representatives confirmed that Alphabet ensures that users grant consent to using their Google information for product development, which would cover AI training, and refused to give opt-in consent rates for cross-linking Google service information for personalization. Similar to Amazon, they believe that user adoption of expanded choices under the DMA is not indicative of compliance.

Second, regarding search data licensing and access on fair, reasonable, and non-discriminatory terms. Two rival search providers argued that the method used by Google to anonymize its click and query dataset removes around 98-99% of queries, rendering it “useless”. Google uses frequency thresholding, which the privacy-focused search engine DuckDuckGo has previously illustrated filters out the useful, less common queries, despite them not containing personal identifiers. Alphabet reiterated the company’s privacy concerns and said that it welcomes suggestions on improvements.

ByteDance

ByteDance’s workshop, held on July 2, 2025, was notably shorter than most, as TikTok is its only designated core platform service. As a result, the discussion was limited, and with little probing from the Commission, there was minimal progress compared to last year’s workshop. Like other gatekeepers, ByteDance declined to share opt-in rates for data combination on TikTok, but did present a consent screen that allows users to choose between generic and personalized ads. They explained that generic ads rely solely on first-party data from the platform, while personalized ads combine first- and third-party data.

On the topic of AI integration, ByteDance, unlike Alphabet, did not classify TikTok’s use of machine learning to personalize recommender feeds as an AI feature. They argued that, since TikTok does not include chatbots, AI integration is “non-applicable” to their services.

Meta

A day before Meta’s workshop, the company released a blog post condemning the Commission’s €200 million fine over its pay or consent model. Under the DMA, Meta must offer users a choice between combining their personal data or using an equivalent, less-personalized alternative. The Commission found Meta’s binary option — paying for a subscription or consenting to personalized ads — was not sufficient, as it “did not allow users to exercise their right to freely consent to the combination of their personal data.” Although Meta has since offered a less-personalized ads service based on more contextual data, the Commission is still reviewing whether this satisfies compliance requirements.

Meta’s compliance “tweaks”

During the workshop, Meta detailed the changes made since the March 2024 compliance deadline, describing the past year as focused on “tweaks” rather than major overhauls, as it believed its services were already largely compliant. Key updates included introducing less personalized versions of its social media services and adding interoperable messaging functionalities to WhatsApp and Messenger.

Less personalized ads represent a “seismic shift” at Meta, offering an “unprecedented degree” of user choice and control, the Meta representatives said. They estimate that the less personalized ads option will use close to 90% less personal data compared to personalized ads. Meta also made it clear in their opening remarks that they see their obligation to make the company’s messaging services interoperable has been unfairly targeted — since only WhatsApp and Messenger have been designated under the DMA — and argued that the requirement is unnecessary, citing the fact that no third-party services have connected to either platform.

Meta also described the process of having one of its core platform services de-designated. Facebook Marketplace was found to no longer meet the core platform service threshold, and the Commission announced its de-designation on April 23. Meta argued that it should never have been designated in the first place, as Facebook Marketplace is a consumer-to-consumer service. The company said that the de-designation process took 14 months, with gaps of up to 8 months between requests for information, calling it a significant regulatory cost and burden.

AI services

Meta did not dedicate a session to AI but made clear its stance that AI “refines” the functionalities and capabilities of a service, and is not separate from the main platform. This is quite similar to Alphabet’s argument that AI is a technology, not a product, and operates within a platform. One of the representatives went as far as to say all digital companies are rolling out AI in the form of integration into existing functionalities to enhance them.

The workshop participants did push Meta to answer questions regarding its use of data to train its Llama large language model, which is being used in core platform service integrations. Participants asked if user data is used to build the company's models and asked when user consent is sought. Meta stated that they view this as falling under EU data protection law and therefore have provided users with an objection form, meaning it is opt-out rather than an explicit opt-in. The Meta representatives went on to argue that the data is de-identified, aggregated, and tokenized so that it cannot be considered personal data.

Data combinations

Meta representatives repeated the arguments outlined in the company's blog post to reiterate its disagreement with the non-compliance decision and confirmed that the company is appealing. A big part of Meta’s argument that the less-personalized ads service is harmful to consumers rested on its finding that users are actively clicking to close ads. Meta reported an 800% increase in ads being closed because users reported them as irrelevant. From this, Meta also argued that it is a worse deal for advertisers whose ads are not reaching desired audiences and might see a decline in click-through rates.

Overall, Meta views this situation as being at odds with the DMA, as it purportedly hinders new start-ups from relying on an ad-supported model and hinders a key marketing player for businesses. This perhaps raises questions about the tension inherent in the DMA between creating structural changes to rebalance power and allowing fairer competition at the edges of gatekeeper platforms. Is it a legitimate concern that advertisers are getting a worse deal, or should we be encouraging the breakdown of a gatekeeper's position as an unavoidable trading partner?

To deliver less personalized ads, Meta went into depth about what data is used. Representatives said that the less personalized option will not use demographic data like relationship status, school location, or education level. The company will also not use interaction with organic content on the platform (likes, comments, views, etc.), meaning that first-party content isn’t used in the less personalized option. The representatives stated that third-party data, including external website interactions, is not used for less personalized ads. Instead, only on-platform signals, such as user feedback on ads (e.g., liking or hiding an ad), are considered. Age, gender, and general location (at the country level), along with limited contextual data from the current session, are also used. They estimated that just 12% of the user data available to Meta is used in this mode. Like other gatekeepers, Meta declined to disclose how many users have opted for less personalized ads, repeating the argument that adoption rates do not equate to compliance.

Messaging interoperability

Meta spent the final session discussing the engineering efforts that have gone into making WhatsApp and Messenger interoperable with third parties. The company representatives ultimately argued that there is no user demand for this, pointing to the fact that no third parties have connected yet. Meta representatives suggested that businesses are uncertain about customer demand or product market fit and are therefore unsure if they are willing to invest in interoperability with Meta.

Despite this, Meta revealed that some third-party services have applied, illustrating that there are some businesses interested, with most applying specifically to interoperate with WhatsApp. They didn’t provide more information, but announced that the first messaging provider is in the final stages of connecting, meaning we will soon see this provision of the DMA in action.

What’s next

The next steps for the Commission are to review the latest compliance measures and decide whether to take action if they deem a measure insufficient. In particular, Meta’s less-personalized ads option will be scrutinized by the Commission, deciding whether or not it meets DMA compliance. Alphabet will also continue dialogue with the Commission regarding self-preferencing on search and its app store rules.

In tandem, the Commission is also actively seeking feedback on the effectiveness of the DMA so far in achieving its objectives of ensuring contestable and fair digital markets, including whether to modify its rules, add new core platform services, or extend the scope of its rules. Therefore, next year’s workshops could look quite different if the gatekeepers are forced to extend compliance measures to new services.