In March of this year, the French competition authority (Autorité de la concurrence, hereinafter the “Autorité”) fined Google[1] €250 million for failing to comply with commitments it had adopted as the outcome of proceedings for abuse of a dominant position. The decision is interesting in relation to online use of publications by information society service providers, particularly their legitimate use in training artificial intelligence (AI) systems.
1. 2019 to 2022: Google’s commitments made binding by decision of the Autorité[2]
The Autorité began investigating Google’s conduct with news publishers and agencies in 2019, the year when DSM Copyright Directive[3] was transposed in France.[4] The DSM Copyright Directive recognized that press publishers shall be provided with the right to authorize the online use (and specifically reproduction and communication to the public) of their press publications.
In November 2019, prompted by reports from major French press publishers and news agency unions,[5] the Autorité opened an investigation into Google’s abuse of a dominant position pursuant to Article 102(c) TFEU.
In its preliminary report, the Autorité noted that Google was suspected of the following forms of abuse:
- imposing on press publishers and news agencies a “zero remuneration” policy for the online use of their publications within the framework of Google’s services (Google Search, Google News, and Google Discover), circumventing the rights provided with national transposition of the DSM Copyright Directive;
- having offered remuneration that did not consider Google’s actual use of the press publications, as some of the uses were tied to other Google services;[6]
- offering non-transparent remuneration by omitting information that was essential for good faith negotiations;[7]
- engaging in discriminatory conduct by dealing identically and without any substantial justification with counterparties having different situations while negotiating.[8]
Following notification of these objections by the Autorité, Google submitted a set of commitments that would remove the competition law concerns. The Autorité accepted those and made them binding in July 2022[9]. With the issues settled, the Autorité ended the proceedings without a finding of abuse and therefore without levying any sanctions (save for a previous fine for failure to comply with urgent interim measures that the Autorité imposed during the investigation).
2. Violation of commitments
In July 2023, the Autorité checked Google’s compliance with the commitments; in October, the Autorité notified the company of a report alleging violations of commitments 1, 2, 4, and 6.
In response, Google submitted corrective measures designed to remedy the alleged violation. These were deemed adequate to remedy the disputed conduct, but insufficient to overcome infringement and avoid the €250 million fine for failure to comply with the 2022 commitments.
3. Assessment of violation of commitments to good faith, transparency and non-discrimination in negotiations
Google previously undertook to negotiate in good faith[10] with press publishers and news agencies when determining remuneration for use of press publications in compliance with the applicable provisions[11] in a transparent, objective, and non-discriminatory way. The company also committed to updating the remuneration amount at least once a year.
According to Google’s submission to the Autorité, to calculate the remuneration due to press publishers and news agencies it would determine a total value for all newspapers and then divide that among the individual press publishers or news agencies based on website traffic data from Similarweb.[12]
The total remuneration included a portion due for display of press publications on Google Search, Google Discover, and Google News and another to reflect “indirect income,” i.e., revenues generated by traffic driven to Google services by the press publishers’ content.
Remuneration for both categories was calculated as a percentage of revenues from each. However, because Search is the only Google service that generates advertising revenue in France, it was the only one whose base could be calculated directly. The revenue base for Discover was estimated based on advertising revenue generated by Discover in other markets where Google receives such revenues. The base for News was a percentage of the Discover base.
However, according to the Autorité, none of the criteria of transparency, objectivity, and non-discrimination were met by Google in determining the remuneration due to press publishers and news agencies.
The data used were inaccurate and incomplete, leading to a lack of transparency. This was demonstrated in the methodological notes Google was required to attach to the remuneration proposal to illustrate its calculations.
This was also reflected in the violation of Google’s second commitment, which was to provide press publishers and news agencies the data they need to conduct negotiations, above all information on Google’s use of their publications.
With respect to lack of objectivity, the Autorité briefly[13] highlighted its concerns about the methodology used to calculate offers for Google’s services (Search, Discover, and News), pointing out that there was no mention in the methodological notes of other services such as Maps, Finance, or Trends, which nevertheless also used the press publishers’ content.
Finally, non-compliance with the non-discrimination criterion was found in that revenues were distributed among publishers without taking into account the difference in the draw of various types of content. In other words, each impression was evaluated by Google in terms of an absolute value (using a 1:1 approach) without weighting the position of the content on the results page, even though a user is more likely to click on results at the top. In addition, a further profile of discrimination lay in the use of Similarweb data. According to the Autorité, there was no detectable correlation “between the allocation of the remuneration coefficient per visit calculated by Google on the basis of impressions of the entire class” of publishers and “the application of this coefficient to the volume of visits to the website as measured by the Similarweb audience indicator.”
Therefore, the Autorité did not believe the remuneration offered by Google met the criteria of good faith negotiation imposed by commitment 1. The Autorité also found that commitment 4 had been violated, as it required Google to offer remuneration in a transparent, objective, and non-discriminatory manner within three months of the start of negotiations.
3.1. The use of press publications in Google’s AI service (previously known as Bard, now Gemini)
The Autorité examined Google’s use of the press publications on Bard, but only to a limited extent. Particularly, it examined the issue from the viewpoint of potential infringement of the provisions on related rights. Amidst the unsettled debate as to whether the use of press publications in an AI service is eligible for protection from such a related-right perspective, the Autorité specified that it had not ascertained whether such use constituted an infringement of press publishers’ rights.[14] Instead, it focused on investigating whether the 2022 commitments, in particular 1 and 6, had been violated.
Bard, rebranded as Gemini in February 2024, is an AI chatbot that answers user questions, launched by Google in France in July 2023. Google stated that the underlying language model was trained on data from the websites of news publishers and agencies (though no specific timeline was given). The AI system is also linked to another Google service—Bard asks Search for the information it needs to answer user questions. These activities were carried out unbeknownst to the press publishers and news agencies vested with the rights to the protected content in question. These rightsholders only learned of the exploitation later on. Upon using the Bard service themselves, they witnessed the system summarizing major news stories in part by providing headlines from their own press publications.
According to the Autorité, the core issue with Bard’s use of such press publications was the lack of an effective technical method for press publishers and news agencies to selectively opt out. They were given no way to object to the use of their protected content by Google’s AI without opting out of other services. Indeed, opting out[15] would affect the “readability” of a publisher’s publications on all the online services, including those for which use had specifically been authorized and remuneration was being negotiated, i.e., Search, Discover, and News.
This was the situation until September 2023, when Google introduced the Google-Extended feature. Publishers could use the feature to opt out of the AI service without affecting the viewing of their content on other Google services. However, this option could only be used going forward, with the consequence that what had already been used could not be removed from the service. AI had already “learned” the information contained in those press publications. And even this function did not prove particularly effective. In October 2023, Bard continued to respond positively to user requests for major news stories published on the websites of French broadcaster TF1,[16] even though it had exercised its right to exclude its news websites (tf1.fr and tf1info.fr) from text and data mining (TDM) activities via the new Google-Extended feature.
So, in relation to the Bard service, the Autorité found Google non-compliant with commitment 1, as it had breached its obligation to be transparent with publishers and agencies about online use of their press publications within the services offered—and in this case even to “form and then operate a new service.”[17]
The absence of a solution that would have allowed press publishers to opt out without affecting the use of protected content in different services for which authorization and remuneration were pending (Search, Discover, and News) was a critical element in this decision. The Autorité found that Google tied the pending negotiations concerning remuneration for the online use of press publications within certain services to the use of the same content on other services. This violated commitment 6, whereby Google was to keep negotiations on rights independent of—and thus not affecting—any other economic relationships between Google and the publishers. The commitment required such use to be covered by separate agreements.
4. Conclusions
The Autorité decision imposed a sanction of €250 million for breach of Google’s 2022 commitments and accepted Google’s new set of remedial measures. It also factored in those measures when calculating the fine.
However, the Autorité’s argument establishing violation of the commitments regarding online use of press publications through Google’s AI service appears strained: Bard was launched in July 2023, so the service was not available at the time of the decision on the breached commitments, nor was it available during the investigation that ended with Decision No. 22-D-13. Therefore, linking Google’s use of press publications to its violation of its 2022 commitments appears critical.
Further, the first commitment expressly excluded from negotiations any services that were not operational at the outset of the negotiations (such as Bard). It stated that the scope of negotiations between Google and publishers were limited to existing services (such as Search, Discover, and News) unless the negotiating parties requested otherwise.
Besides, consistently with commitment 6, which was designed to keep negotiations independent of other economic relationships, parties were given the opportunity to request that such services be subject to negotiation. Obviously, in order to exercise that right parties had to be duly informed of all Google services where their content was used.
The absence of a valid opt out method for publishers to object to the use of their content on Bard/Gemini—without jeopardizing their right to negotiated remuneration—played a key role, in the eyes of the Autorité to ascertain the violation of commitment 6. Indeed, according to the decision, Google effectively tied the two practices by forcing publishers to choose whether to allow tout court the use of their content on Bard and thus receive remuneration (limited to Search, Discover, and News), or to prohibit en blanc such use—including for the negotiated services—and consequently forgo the negotiated remuneration.
Therefore, above all, the case seems to highlight the risk of the inadequacy of tools for opting out.
The decision in question prompts thoughts on the choice the European legislature made to cite the provisions governing text and data mining (TDM)[18] in the DSM Copyright Directive in the AI Act. It requires an adequate opt out system that allows legitimately accessed works and other protected materials to be summarized and reproduced unless the rightsholders expressly reserve that right. The case suggests that rightsholders are not always equipped with adequate technical, commercial, and legal tools to express that type of reservation without affecting their right to dispose of their works and protected content. With AI applicable to so many fields, questions arise about how appropriate that system is and whether it might be better to have a more traditional opt in system that would allow a rightsholder to agree to the use of works or materials for the purpose of training AI models, as is generally provided for copyright and related rights. This would be consistent with another fundamental principle of copyright law, namely the independence and severance of exclusive rights, according to which the exercise of one should not prejudice the exercise of another.[19] It follows that even when it comes to use of protected content for TDM activities, rightsholders subject to opt out rules can and should be able to exercise their rights independently over the content.
[1] “Google” meaning Alphabet Inc, Google LLC, Google Ireland Ltd, and Google France.
[2] Decision No. 22-D-13.
[3] Directive (EU) 2019/790 on copyright and related rights in the Digital Single Market.
[4] Law No. 2019-775.
[5] Syndicat des Éditeurs de la Presse Magazine, Alliance de la Presse d’Information Générale, Syndicat de la presse quotidienne nationale, Syndicat de la presse quotidienne régionale, Syndicat de la presse quotidienne départementale, Syndicat de la presse hebdomadaire régionale, and Agence France-Presse.
[6] Par. 79 Decision No. 22-D-13.
[7] Par. 81 Decision No. 22-D-13.
[8] Par. 83 Decision No. 22-D-13.
[9] Decision No. 22-D-13.
[10] Commitment no. 1 Decision 22-D-13.
[11] Article L. 218-4 French Intellectual Property Law “The remuneration due in connection with related rights for the reproduction and communication to the public of journalistic publications in digital format shall be based on the revenues derived from any kind of exploitation, direct or indirect, or, failing that, assessed on a lump-sum basis, particularly in the cases provided for in Article L. 131-4.
“The determination of the amount of such remuneration takes into account factors such as the human, material, and financial investment made by publishers and news agencies, the contribution of print publications to political and general information, and the extent to which print publications are used by online public communication services.
“Online public communication services shall be obliged to provide publishers and news agencies with all information regarding the use of news publications by their users, as well as all other information necessary for a transparent assessment of the remuneration referred to in the first paragraph of this article and its distribution.” [authors’ translation].
[12] Similarweb (www.similarweb.com) is an information technology company that collects data to analyze and measure user engagement on websites.
[13] Since there was no dispute, the grounds are of a summary nature, which differs from those for a decision on the merits.
[14] Par. 267 Decision No. 24-D-03.
[15] A robots exclusion protocol is a set of instructions and directives that the owners of a website use to tell crawlers which parts they may visit.
[16] French TV channel that produces France’s most watched news program.
[17] Par. 268, Decision No. 24-D-03.
[18] Art. 4, para. 3 DSM Copyright Directive: “The exception or limitation in paragraph 1 shall apply provided that the use of the works and other materials referred to in that paragraph has not been expressly reserved by the rightsholders in an appropriate manner, e.g., through means that allow automated reading in the case of content made publicly available online.”
[19] See Art. 19 of Law April 22, 1941, No. 633.
