Unfair Trading Conditions Meet AI Search: The Turkish Competition Board's Interim Measures against Google in Comparative Perspective
November 19, 2026
Introduction: An Interim Measure Where Others Have Investigated, Regulated or Declined
In April 2026, Todd Davies argued on this blog that letting publishers opt out of AI Overviews would not cure the competitive harm those summaries cause, and that interim measures under Article 102 TFEU might be needed before a final decision arrives too late. Six months later, the first interim measure of that kind came not from Brussels but from Ankara. On 2 November 2026, the Turkish Competition Authority announced that the Competition Board had opened an investigation into Google's "AI Overviews" and "AI Mode" features and, by a decision of 8 October 2026, had imposed interim measures under Article 9(4) of Law No 4054 on the Protection of Competition.1 The Board's concern is twofold: that Google discriminates among comparable news publishers when selecting and licensing their content for its AI features, and that the terms on which that content is used are unfair. The second concern rests on a structure that European readers will recognise from the French neighbouring-rights cases. For six and a half months after the features launched in Türkiye, a publisher could keep its content out of AI Overviews only by risking its visibility in Google Search; and those offered a licence received a standard contract on terms the Board found they could not meaningfully negotiate.
As far as we can tell, this is the first binding measure adopted under an abuse-of-dominance provision, in Europe or elsewhere, that specifically targets AI-generated search summaries and their terms vis-à-vis news publishers. The qualifications matter and are best stated at the outset. The UK Competition and Markets Authority imposed a publisher conduct requirement on Google on 3 June 2026 under its ex ante digital markets regime; South Africa's Competition Commission had earlier obtained, through a market inquiry, a negotiated commitment giving its publishers parity with European publishers in controlling how their content appears in AI Overviews; interim measures on generative-AI conduct had already been imposed on Meta by the Italian, Brazilian, Turkish and EU authorities in the WhatsApp assistant cases; and the Autorité de la concurrence fined Google in March 2024 for, among other things, training Bard on press content with no genuine opt-out.2 None of these applied an abuse prohibition to the terms on which AI search summaries use publishers' content. The Turkish decision does, and it does so while the European Commission's investigation into the same conduct is pending, EU national authorities are barred from acting on it, and the only US court to have addressed the question has declined to.
This post asks what the Board borrowed from Europe, where it departed, and how durable the result is. Our reading is that it fused three European templates into one interim measure: the Commission's test for unfair trading conditions in Apple Music Streaming, the Autorité's negotiation injunctions from the neighbouring-rights cases, and the CMA's opt-out design. It cites only the first. The composite goes further than any of its models, which is both its interest and its vulnerability.
The Decision: Facts, Theories and Remedies
AI Overviews and AI Mode went live in Türkiye on 18 February 2026. Google's "search generative AI control", a Search Console setting that lets a site withhold its content from AI features without, on Google's account, affecting its ranking elsewhere, was announced on 3 June 2026, the day the CMA's conduct requirement was published; it was tested with UK sites first and reached all sites worldwide on 31 August 2026. The Board opened a preliminary inquiry on 27 August, inspected Google's Turkish subsidiary on 1 September and decided, unanimously, on 8 October; Google has three months to comply, with daily fines thereafter, and may seek judicial review in Ankara.
Two sets of facts carry the decision. The first is dependency. On the Board's (redacted) figures, search engines account for a substantial share of Turkish news sites' traffic, highest for publishers with weaker brands, and Google's share of that search-referred traffic is consistently high; the Board concludes that Google is an unavoidable trading partner for news publishers (para 309). It then finds that tying a publisher's choice about AI use to its visibility in Search limited publishers' ability to refuse (para 310), that the opt-out arrived six and a half months after the features did (para 313), and that the opt-out's existence does not settle the matter, since not using it is no evidence of a free choice where the consequences for ranking are not predictable (para 314). The Board also records that AI referrals account for less than 0.01% of publishers' traffic (para 348).
The second set of facts concerns Google's licensing contracts. Google's News AI pilot, announced on 10 December 2025 and said by Google to cover more than 400 publications by October 2026, pays selected publishers for enhanced content used to develop AI features. Examining the global template and the Turkish version, the Board found that Google alone decides which publishers are offered a contract; that the text is standard and non-negotiable; that the fee is derived from brand value and third-party traffic data by an undisclosed method; that the licence is worldwide, sublicensable and extends to training; that there is a single payment with no commitment to use the content or send traffic; and that attribution and linking are owed only "to the extent provided to similarly situated partners" and "where reasonably practicable" (paras 326-329). The global template contains a most-favoured-customer clause that the Turkish version omits (paras 36, 47). That publishers signed quickly, the Board observes, shows consent, not competitive terms (para 330).
On these facts the Board found a strong suspicion of two abuses under Article 6: discrimination among comparable publishers, both in the selection of sources for AI answers and in the offer and negotiation of contracts, and the imposition of unfair trading conditions. It noted that the same facts may serve both theories (para 195), and the interim measures track both. Google must not discriminate among equivalent publishers on criteria that are not objective, transparent and pre-determined; must negotiate in good faith with publishers that seek a contract, on objective, verifiable and transparent criteria reflecting the scope and intensity of use, user engagement and the effect on traffic, visibility and advertising revenue; must not penalise in ranking, visibility, indexing or traffic a publisher that declines a contract or its terms; must supply publishers periodically with verifiable data on how their content is used and what engagement and traffic it generates; must make the AI opt-out exercisable independently of visibility and indexing in other Google services; must not worsen existing contracts or apply divergent terms in new ones; and must report compliance to the Authority (para 396).
Two features of this list stand out. Article 9(4) permits interim measures that preserve the situation prior to the infringement, yet a duty to negotiate on value-reflecting criteria and to supply usage data creates obligations that did not previously exist; that is likely to be Google's first argument before the administrative court. European practice in digital cases shows what interim measures have been used for: to stop the conduct (Broadcom, 2019), to reopen access that a policy change had closed (the Commission's and the Italian authority's WhatsApp measures), and, in France alone, to impose duties of negotiation and disclosure anchored in a statutory right (Google 2020, Meta 2026). The one attempt to impose a positive duty to deal at the interim stage without such an anchor, the compulsory licence in IMS Health, was suspended by the EU courts and later withdrawn; the DMA's own interim-measures power has never been used. The Ankara 9th Administrative Court's Trendyol judgment states the same limit from the other side: an interim measure may neither punish the undertaking nor put complainants in a better position than before the alleged infringement, and may not substitute for the final decision; the court annulled the non-discrimination order for want of any analysis of whether the sellers concerned were in comparable situations.3 The Google decision cites that judgment for its standard of proof (paras 382-383), not for its ceiling. Its measures combine all three European types, and their negotiation and data duties have no statutory anchor; the comparability analysis that Trendyol lacked (paras 218-246) protects the discrimination limb, not those duties. The second feature is length: the list is longer than any comparable European instrument, combining the CMA's opt-out and anti-retaliation rule, the Autorité's negotiation and information duties, and a non-discrimination obligation that neither has.
The Board's Framework: An Imported Test, a Dependency-Based Application
Unlike Article 102(a) TFEU, Article 6 of Law No 4054 contains no express reference to unfair prices or trading conditions. The Board derives the theory from the general prohibition, noting that the examples listed in Article 6 are not exhaustive (para 305). It then adopts, almost verbatim, the Commission's test from Apple Music Streaming: trading conditions are unfair where they are (i) imposed by a dominant undertaking on its trading partners, (ii) harm the interests of those partners or of third parties, including consumers, and (iii) are not necessary to achieve a legitimate objective or go beyond what is necessary for it (para 303, citing Apple, paras 529-530). As in Apple, the Board adds that a finding of unfair conditions suffices and that no separate inquiry into a restriction of competition is required; it is enough that the terms adversely affect a parameter of competition such as price, choice, quality or innovation, to the detriment of trading partners or consumers (para 304). A few paragraphs earlier, however, the Board had said that competition-law intervention against unfair terms presupposes conduct that is not isolated and that produces appreciable "restrictive effects on the market" (para 293). The two can be reconciled if "restrictive effects" means harm to a parameter of competition rather than exclusion, and paragraphs 295 (the special responsibility of a dominant undertaking extends to "all directly harmful conduct") and 302 (unfair terms may be exploitative and exclusionary at once) point that way. But paragraph 293 also speaks of risks to "market structure and the effective competitive process", and the Board's own reasoning in Biletix, to which we return, pulled the other way. The decision notes the point without resolving it.
The application follows the three limbs. Under the first, the Board does not ask whether a non-dominant undertaking could have imposed the same terms; it asks whether publishers could escape them, and answers by reference to traffic dependency, the delayed opt-out and the standard contract. Under the second, it points to publishers left without any remuneration, to the information asymmetry created by Google's not sharing click data, and to smaller publishers below the contracting threshold who face a choice between uncompensated use and the risk of losing visibility (paras 333-337). Under the third, it rejects Google's explanation that a single control for all AI features avoids complexity, observing that a unified mechanism could still allow separate choices for AI Overviews and AI Mode (paras 346-347), and finds that the breadth of the licence, the opacity of the fee, the single-payment structure and the absence of any usage or referral commitment raise doubts, taken together, about necessity (para 352).
One sentence in the decision pulls against this dependency-based application. Introducing the assessment, the Board defines unfair conditions as terms that a dominant undertaking applies by taking advantage of its position "and that a non-dominant undertaking could not put forward" (para 306). That is a benchmark-flavoured definition in a decision that never performs a benchmark. Google will notice.
The Defence to Watch: "Everyone in AI Does This"
Google's most natural defence in Ankara is to argue that its terms are the industry's terms: OpenAI, Perplexity and Microsoft use publishers' content on comparable conditions, so the terms cannot be attributed to dominance in search. The definition in paragraph 306 invites the argument. How has it fared elsewhere?
In France, when Google appealed the Autorité's 2020 neighbouring-rights injunctions (discussed below), it argued that other search engines had not concluded remuneration agreements either. The Paris Court of Appeal answered first that Bing had in fact agreed to negotiate paid licences, and in any event that non-dominant operators could hardly avoid aligning themselves with the conduct of a quasi-monopolist, since departing from it would impose costs the market leader does not bear; Google's position allowed it to act free of competitive pressure.4 In Germany, the Federal Court of Justice in Facebook rejected the argument that the contested terms were customary in the market: terms used by firms outside the relevant market proved nothing, and the imitation of a dominant firm's terms across an industry says nothing about what effective competition would produce. The court accepted that a stricter standard governs purely exploitative abuses, but did not apply it, because Facebook's terms also impaired competition.5 That was an interim ruling; the case closed in October 2024, when Meta withdrew its appeal, without a judgment on the merits. At EU level, the Commission in Apple Music Streaming treated Apple's position as an unavoidable trading partner as what enabled the terms, held that whether non-dominant players used similar terms was immaterial, and described benchmarking as a feature of excessive-pricing cases and not a mandatory one. The Commission reasoned in the same way in Facebook Marketplace.6
Turkish practice has been less settled. In Biletix (2021), the Board examined a dominant ticketing platform's drip pricing as a possible unfair condition. It acknowledged harm to consumers but looked for a competitive dimension before competition law could intervene, found that competitors used or could use the same practice, and declined to find an abuse, referring the matter to consumer law. It also framed the question in terms of causality: an abuse may arise where the undertaking can engage in the conduct because it is dominant and/or where the anti-competitive effects stem from dominance.7 In Yemeksepeti (2025), the Board assessed a platform's requirement that restaurants use its courier service, found the platform no longer dominant and, for the sake of completeness, asked whether restaurants had realistic alternatives, concluding that they did. The Google decision continues Yemeksepeti's vocabulary, repeating that unfair terms can persist only in markets with entry barriers and a quasi-monopolist (para 306), but the real shift is the adoption of the Commission's test; of Biletix's two limbs, it is the first, dominance enabling the conduct, that does the work. Under that test, as under Yemeksepeti's, an "everyone does it" defence has little purchase: the question is dependency, not replicability. The loose thread is the definition itself, which still speaks of what a non-dominant undertaking could put forward.
In a forthcoming article we examine whether causality between dominance and the ability to impose a term should be a necessary element of exploitative abuse, and distinguish approaches that answer by comparing the dominant firm's terms with those of its rivals from approaches that ask only whether customers can escape.8 For present purposes the point is narrower. The Board's reasoning is of the second kind. If Google wishes to displace it, the productive argument is not that others behave the same way but that the dependency findings are wrong or stale: that publishers who have used the opt-out since 31 August 2026 have not lost visibility, that AI referrals are growing, that the pilot's terms are in fact negotiated. That is an evidentiary contest, and the Board's data-provision order will, somewhat paradoxically, generate the evidence on which it is fought.
Four Contemporaneous Responses: Brussels, Paris, London, Washington
The decision will also be judged against what four other systems have done with the same conduct over the same period.
The European Commission. On 9 December 2025 the Commission opened formal proceedings against Google under Article 102 to examine whether AI Overviews and AI Mode are built on publishers' content without appropriate compensation and without the possibility for publishers to refuse without losing access to Google Search, and whether Google thereby imposes unfair terms and conditions on publishers. That is the Turkish theory in the Commission's words. Complaints have accumulated: independent publishers asked for interim measures in June 2025, the European Publishers Council filed a formal complaint in February 2026, and five French publishers' organisations joined in October 2026. The Commission's only visible steps are requests for information sent to publishers in the summer of 2026 on whether they would use Google's new opt-out and whether their content is adequately attributed. It has used its interim-measures power once in this field, in June 2026, to restore rival AI assistants' free access to WhatsApp, an exclusionary case. Under Article 11(6) of Regulation 1/2003, the opening of proceedings relieves national authorities of competence over the same practices; when the Polish authority charged Google in October 2026 over neighbouring-rights negotiations, it said expressly that AI features were a matter for the Commission. Türkiye sits outside that architecture, which is why its authority was free to act first. The DMA runs in parallel: the specification decisions of July 2026 on Android interoperability for AI assistants and on search-data sharing, both analysed on this blog, and the non-compliance fines of 23 July 2026, whose press release "takes note of Google's proposals on how it plans to apply the principles of the decision to AI Overviews and AI Mode". The DMA's self-preferencing principles may reach AI answers before Article 102 does; neither instrument addresses the terms on which publishers' content is used.
The Autorité de la concurrence. Its interim measures of April 2020 ordered Google to negotiate in good faith with press publishers and agencies, to provide the information needed to assess remuneration, and to ensure that the negotiations had no adverse effect on the indexing, ranking and display of their content; the Paris Court of Appeal upheld them in October 2020. Non-compliance cost Google €500 million in 2021; commitments followed in 2022; and in March 2024 the Autorité fined Google €250 million for breaching them, finding among other things that Bard had been trained on press content without publishers being told and that the only way to keep content out of Bard was to block Google's crawling altogether, Search, Discover and News included. That is the structure the Turkish Board found two and a half years later, inside the results page itself. In July 2026 the Autorité imposed interim measures on Meta on an unfair-trading-conditions theory arising from the neighbouring-rights negotiations, ordering good-faith negotiation on transparent, objective and non-discriminatory criteria, disclosure of information within fifteen days, no degradation of display and regular reporting; the Paris Court of Appeal confirmed them in full on 29 October 2026. AI Overviews reached France only on 22 July 2026, and the Alliance de la presse d'information générale asked the Autorité on 11 August to enforce the 2022 commitments against them; no step has been made public. The Turkish decision cites none of this. Its remedies nonetheless read as a translation of the French injunctions, with the AI opt-out added.
The CMA. The authority designated Google with strategic market status in general search in October 2025, expressly including AI Overviews and AI Mode in the designated activity, and on 3 June 2026 imposed a publisher conduct requirement: effective controls to withhold content from search AI features and from Google's broader generative-AI services, covering training and grounding, at directory and page level; a prohibition on undermining the controls through ranking signals or retaliation; attribution; engagement metrics; and compliance reporting. The obligations take effect on 3 December 2026, with page-level grounding controls following in March 2027. The CMA considered and declined to require separate controls for AI Overviews and AI Mode, accepting that the features are integrated, and deferred the question of remuneration while reserving further measures on "a fair exchange of value". No Competition Act case was opened despite a complaint requesting interim measures; the regulatory route was chosen instead, and Google's worldwide control is its product. The Turkish decision records that control's arrival and treats it as insufficient: late, unverified, and not self-evidently neutral in its effects on ranking. On the two points where the CMA stopped, the Board went on. It rejected the integration argument for a single control, and it ordered negotiation on criteria that reflect the value of the content.
The US courts. In the remedies phase of United States v. Google, the plaintiffs proposed an order allowing publishers to opt out of crawling for generative AI while remaining in Search. Judge Mehta declined in September 2025: the proposal bore no relationship to the unlawful conduct found, the court had heard no testimony from a single publisher, and the demand was for "overly regulatory requirements". The final judgment of December 2025 is silent on publishers. On 30 September 2026 the same judge dismissed the private suits brought by Chegg and Penske Media over AI Overviews: the expectation of search traffic in return for content "is not an agreement", general search and AI Overviews form "an integrated search experience" rather than separate products, and publishers lack antitrust standing in the search market; the harms were real, the court said, but a matter for Congress or regulators.9 Appeals are pending. The contrast with the Turkish decision is doctrinal before it is political. Without an exploitative-abuse concept, the publishers' claim has no doctrinal home: it must be forced into agreement, tying or exclusion, and fails. With one, the question becomes whether the terms result from dominance and whether they are fair, which is where the Turkish decision and the Commission's investigation both sit.
Elsewhere the issue is as global as the product: South Africa's market-inquiry remedies of November 2025, Brazil's decision of April 2026 to open proceedings against Google on a theory that includes generative-AI summaries, and the Japanese authority's surveys of AI search services pursue the same concern through different instruments.
Conclusion: What to Watch
Three things will show whether the Turkish decision is an outlier or an early mover. The first is the final decision: whether the Board keeps the Apple framework, whether the effects language of paragraph 293 is given work to do or, as paragraph 302's acceptance that unfair terms can be exploitative and exclusionary at once suggests, quietly dropped, and whether discrimination, which needs no theory of causality at all, becomes the principal ground. The second is judicial review, where the question will be whether a duty to negotiate on value-reflecting criteria preserves a status quo or creates one; Trendyol and IMS Health give Google its script. The third is the Commission, whose investigation will meet the same questions: whether consent to a standard licence under conditions of dependency is free, whether an opt-out that arrived after the fact and whose effect on ranking is unverified cures the imposition, and whether a benchmark against other AI providers is relevant. The Turkish Board has given one set of answers under an abuse provision; the CMA has given another under regulation; the US courts have declined to give any under the Sherman Act. For the European debate on unfair trading conditions in digital markets, the interest of the Turkish decision is that it treats Apple Music Streaming not as a one-off but as a template, and applies it to the conduct that will define the next phase of that debate.
- 1Competition Board, decision of 8 October 2026, No 26-37/1000-423 (Google Interim Measures), announced on 2 November 2026 (announcement). Paragraph references are to the published reasoned decision, which is available only in Turkish; translations are the authors'. The preliminary inquiry was opened by decision of 27 August 2026, No 26-31/932-M.
- 2CMA, Publisher Conduct Requirement, 3 June 2026 (final decision); Competition Commission (South Africa), Media and Digital Platforms Market Inquiry, Final Report, Remedial Actions, 13 November 2025, Google clause 9.1 (Annexure 8); AGCM, Case A576, interim measure of 22 December 2025 (press release); CADE, preventive measure of 12 January 2026, upheld by the Tribunal on 4 March 2026; Competition Board, decision of 14 May 2026, No 26-18/536-193 (announcement); Commission decision of 9 June 2026, Case AT.41034 (IP/26/1276); Autorité de la concurrence, decision 24-D-03 of 15 March 2024 (press release).
- 3Ankara 9th Administrative Court, judgment of 25 May 2022, E. 2021/2069, K. 2022/1157 (DSM Grup (Trendyol)), on Competition Board decision of 30 September 2021, No 21-46/669-334; the court also cut the eight-year record-retention orders back to the final decision and upheld the self-preferencing and data-use measures. Commission decision of 16 October 2019, Case AT.40608 Broadcom (interim measures), followed by commitments of 7 October 2020. Commission decision of 3 July 2001, Case COMP D3/38.044 NDC Health/IMS Health (interim measures), suspended by order of the President of the Court of First Instance of 26 October 2001 in Case T-184/01 R, confirmed by order of the President of the Court of Justice of 11 April 2002 in Case C-481/01 P(R), and withdrawn by Commission Decision 2003/741/EC of 13 August 2003. For the general limits, Case 792/79 R Camera Care v Commission, EU:C:1980:18, paras 18-19, and Joined Cases 228/82 and 229/82 Ford v Commission [1984] ECR 1129; in France, Code de commerce, art. L.464-1.
- 4Cour d'appel de Paris, 8 October 2020, RG No 20/08071, paras 110 and 117-121 (judgment). Google did not lodge a further appeal and the judgment is final.
- 5Bundesgerichtshof, decision of 23 June 2020, KVR 69/19 (Facebook), paras 65, 72-82 and 88-91 (Bundeskartellamt's English translation). The proceedings before the Oberlandesgericht Düsseldorf ended on 10 October 2024 when Meta withdrew its appeal (Bundeskartellamt).
- 6Case AT.40437 Apple – App Store Practices (Music Streaming), Commission Decision C(2024) 1307 final of 4 March 2024, paras 532-544; see also Case AT.40684 Facebook Marketplace, Commission Decision C(2024) 8053 final of 14 November 2024, paras 1083-1093. Both decisions are under appeal before the General Court (T-260/24 and T-66/25).
- 7Competition Board, decision of 21 January 2021, No 21-04/53-22 (Biletix), paras 179-187 (English courtesy translation); decision of 22 May 2025, No 25-20/488-228 (Yemeksepeti), paras 233-234, 384-404 and 513-517.
- 8Doğa Küçükay and Ahmed Pehlivanlı, 'Dominance, Dependency and Unfair Terms: Rethinking Causality in Exploitative Abuse' (forthcoming).
- 9 United States v. Google LLC, No 1:20-cv-03010 (D.D.C.), Memorandum Opinion on remedies of 2 September 2025, at 5 and 203 ff., and Final Judgment of 5 December 2025 (DOJ case page); Chegg, Inc. v. Google LLC, No 1:25-cv-00543, and Penske Media Corp. v. Google LLC, No 1:25-cv-03192 (D.D.C.), Memorandum Opinion of 30 September 2026 (opinion). Both dismissals have been appealed to the D.C. Circuit.
You may also like