The impact of artificial intelligence (AI) on trade mark law has recently drawn increasing scholarly attention, with commentators exploring the challenges posed by this technology across a variety of domains – from its use by trade mark offices and in adjudication processes, to its broader implications in AI-driven behavioural advertising and the Digital Services Act (DSA). This study contributes to the legal analysis, from the point of view of the EU legal framework, of two specific scenarios: (a) the use of third parties’ trade marks by operators of online shopping platforms – including online marketplaces – that rely on AI systems either to recommend purchases to consumers or to execute purchases on their behalf; and (b) the use of third parties’ trade marks by online marketplace operators to promote or recommend competing commercial offers to users. The first scenario challenges the continued relevance of the “average consumer” as the legal benchmark for assessing both the existence of a likelihood of confusion and the liability of online shopping agents. The second scenario, which has emerged in case law across several jurisdictions, raises interpretative ambiguities that warrant closer examination. These concern the extent to which existing jurisprudence accommodates the pro-competitive potential of such practices, while still safeguarding the rights of trade mark holders. This study offers interpretative clarifications and also considers the harmonised EU rules on misleading advertising.
L’impatto dell’intelligenza artificiale (IA) sul diritto dei marchi ha recentemente attirato l’attenzione di numerosi studiosi che hanno esplorato le sfide poste da questa tecnologia in vari settori, dall’utilizzo da parte degli uffici marchi e nei procedimenti giurisdizionali, alle sue più ampie implicazioni nella pubblicità comportamentale fondata sull’IA e nel Regolamento UE sui servizi digitali (DSA). Il presente studio contribuisce all’analisi, dal punto di vista del diritto dell’Unione Europea, di due specifiche ipotesi: a) l’uso di marchi di terzi da parte di operatori di piattaforme di shopping online (compresi i mercati online) che utilizzano sistemi di IA per consigliare acquisti ai consumatori o effettuare senz’altro detti acquisti per conto dei consumatori medesimi; b) l’uso di marchi di terzi da parte di operatori di mercati online per promuovere o raccomandare agli utenti offerte commerciali di operatori concorrenti. La prima ipotesi mette in discussione la rilevanza del “consumatore medio” come parametro di riferimento per valutare l’esistenza di un rischio di confusione e la responsabilità degli agenti di shopping online. La seconda ipotesi, emersa nella giurisprudenza di diversi sistemi normativi, solleva questioni interpretative che richiedono un esame più approfondito. Esse, in particolare, ruotano attorno al problema di fondo affrontato in giurisprudenza di come si possa, al contempo, fare spazio al potenziale pro-concorrenziale di queste pratiche tutelando gli interessi dei titolari dei marchi protetti dal diritto dei marchi UE (e armonizzato a livello UE). Questo studio propone alcuni affinamenti interpretativi, anche avuto riguardo alle norme armonizzate a livello UE in materia di pubblicità ingannevole.
1. Introduction - 2. Is trade mark law still necessary? - 3. Is trade mark law in need of adaptation? The notion of the “average consumer” - 4. A special case: the use of third parties' trade marks on the platforms and in the advertising of online marketplace operators - NOTE
The impact of artificial intelligence on trade mark law has recently drawn scholarly attention [1]. Some have focused on AI as a tool for prior art searches by IP offices [2] and on computational trade mark infringement and adjudication [3]. Other authors have addressed the intersection between trade mark law, AI-driven behavioral advertising and the transparency requirements set forth by the Digital Services Act (DSA) [4], while others have devoted attention to the more general issue of the impact of AI on the principles and doctrines that lie at the heart of trade mark law [5]. While the latter issue cannot be ignored due to its general relevance, this study aims to contribute to the legal analysis of the use of third parties’ trade marks by operators of online platforms (including online marketplaces) that use artificial intelligence systems to recommend (or directly make) purchases for consumers, and the use of third parties’ trade marks by operators of online marketplaces to promote or recommend competing commercial offers to internet users. The related question of the liability of online platforms arising from such use is also addressed. The subject has been studied [6]. However, as we shall see, there is still room for more insight, particularly in the light of developments in case law.
First, before getting to the heart of the matter, it must be made clear that the legal discourse on the use of trade marks by AI in the online environment cannot take as a privileged point of reference the total substitution of flesh-and-blood consumers by AI in the purchasing process [7]. Admittedly, it is reasonable to assume that the maximum efficiency of AI in identifying the consumption preferences of each individual would imply that trade marks would no longer be necessary to reduce the costs of consumers’ search for satisfactory products, especially if the consumer is completely eliminated from the purchasing process [8]. In this vein one author has articulated the concept of the «death of trademark» [9]. Other authors have highlighted the phenomenon of the removal of the human element from purchases and the irrelevance of trade marks in these purchases, referring to automated purchasing systems that use the Internet of Things (and hence machine-to-machine communication channels) and blockchain as a product-tracking tool [10]. What these authors have in common is that they advocate a comprehensive re-evaluation of trade mark law, starting from the exclusion of humans from the purchase decision process.
However, it is clear that the use of artificial intelligence in automated shopping systems is currently limited to specific scenarios [11]. For example, it is conceivable that the consumer could authorize the fridge to inform the virtual shopping assistant that a particular item (e.g. milk) in the household’s inventory is about to run out. In response, the virtual assistant will order the item in question, selecting from which retailer it should be obtained (based on data processing performed by an AI) [12]. Nevertheless, it is reasonable to assume that the purchase of Veblen goods (e.g. luxury goods) or products with a high unit value (e.g. home furniture) will still be carried out by the end consumer in the future [13]. In addition, even in the context of automated systems that purchase goods on behalf of consumers, if the consumer finds unsatisfactory (or no longer satisfactory) products, she can return the item and, in any case, choose to inform the system of a different preference (for the future). Moreover, even in the latter scenario, trade marks would continue to «serve as a repository for information contributed by third parties (e.g., consumer reviews, word of mouth, etc.)» [14] which may also influence changes in consumer preferences fed back into the automated shopping system. This inherent possibility, even within the shipping-then-shopping paradigm, effectively precludes the complete replacement of human decision-making processes in online shopping contexts [15]. In other words, far from being dethroned, the consumer retains ultimate decision-making authority (at least as a last resort) [16].
It can therefore be argued that purchases will continue to be made predominantly by consumers who make use of targeted recommendations generated by the artificial intelligence used by online platforms and, to a lesser extent, by automated shopping systems. The presence of recommendation systems is, after all, an aspect kept in mind by current legislation and, in particular, by Regulation (EU) 2022/2065 (so called “Digital Services Act”) [17]. Consequently, both systems must be taken into account when interpreting trade mark law.
Indeed, in order to preserve the residual sovereignty of the real consumer, it seems necessary that, even in the case of automated purchasing systems using machine learning (“ML”) or deep learning (“DL”), the consumer should be able to identify, without error, the commercial origin or the commercial affiliation of the (albeit) automatically ordered products [18]. This implies, among other things, the continuing need to protect the origin function performed by trade marks against third party interference [19]. It has been argued that the protection of this function enables the establishment of a reputation and brand image, which in turn is protected by EU (or EU-harmonized) trade mark law [20]. This finding refutes the reductionist approach to trade mark law advocated by some authors, particularly in the United States, and underlines the continuing need to protect the multiple functions of trade marks [21]. More generally, the idea that artificial intelligence can make purchasing decisions (or suggest choices) that best meet consumers’ preferences is far from certain [22]. In fact, there is empirical evidence to suggest the opposite [23]. Thus, artificial intelligence may select one product over another simply because of the way it is presented on the market [24], rendering AI decisions no more efficient than human decisions based on trade marks.
As a result, the relevance and usefulness of EU trade mark law remains undiminished, even in the context of the increasing use of automated shopping systems, although some interpretative adjustments are needed.
In the case of an online service provider recommending (by means of AI systems) the purchase of products infringing the trade mark rights of others, for example because the later sign is likely to be confused with the earlier registered sign or because the exhaustion rules have not been complied with, the paradigm of the “average consumer” as a natural person can be maintained, since it is to the latter that the purchase recommendation is addressed, and the question arises as to the liability regime of the online service provider [25]. Conversely, in the case of artificial intelligence making the purchase, it seems necessary to go beyond the model of the ‘average consumer’, even if the final recipient of the product or service is still a flesh-and-blood person, because in this case she has no role in the individual purchase decision, having authorized the system operator to make purchases on her behalf in an automated way [26]. In this case, the question is whether an artificial intelligence is likely to make a mistake as to the commercial origin of the products sold online and, if so, what the criteria should be for assessing the likelihood of confusion [27]. In addition, it would be crucial to consider the feasibility and limits of judicial review of the factors taken into account by the algorithm in making purchase decisions [28]. Finally, it is crucial to examine the conditions under which the operator of an automated shopping service using artificial intelligence could be held liable (directly or contributorily) for infringements of exclusive trade mark rights that occur during the provision of the service itself [29]. In this respect, it has been rightly pointed out that, in interpreting trade mark law, the case-law takes into account a number of factors attributable to cognitive ‘deficiencies’ peculiar to human beings. However, these factors clearly have no reason to be taken into account when assessing the “behaviour” of software using a form of artificial intelligence. From this point of view, it is reasonable to propose that the concept of the average consumer, as it relates to automated shopping services and targeted recommendations, be extended to include a typified model of the agent. This model would be constructed by examining the characteristics that can reasonably be expected from the proper functioning of artificial intelligence, taking into account the technological evolution at the time of the alleged infringement [30]. Within these limits, it is reasonable to propose a judicial review of the algorithms and factors taken into account by artificial intelligence systems that assist (or substitute for) consumers in their purchasing decisions [31]. This is further strengthened by the current legal framework, in particular following the implementation of Directive (EU) 2016/943 on trade secrets in the legal systems of the EU Member States [32]. In other words, it seems inevitable, at least for certain purposes, to interpretatively construct an artificial agent model that: (1) lacks the (characteristics and) perceptual and cognitive limitations of the average consumer, and (2) is, however, endowed with its own deficiencies – taking into account the technology available at the time – such that, under certain assumptions, there may be a «likelihood of confusion» [33].
Assuming that these assertions are correct, the model in question serves two functions: On the one hand, in the context of targeted recommendation services – and provided that the conduct of the targeted recommendation provider is not itself considered to be infringing – it makes it possible to draw the line beyond which the service provider cannot be held liable for the infringement of others’ trade marks that are not “filtered” due to “defects” in perception that cannot be eliminated by adequate systems; on the other hand, in the case of automated purchases made by the service provider on behalf of the customer, the regulatory model of artificial intelligence chosen appears to be functional to identify innocent counterfeiting behaviour in order to exclude the service provider’s liability for damages [34]. This interpretative approach does not, in any case, affect the subjection to any injunctive measures taken by the judicial authority or any liability arising from the conduct of the service provider after acquiring detailed knowledge of the infringement and, even more so, does not exclude the liability of the service provider arising from failure to comply with injunctions issued by the competent authority.
Irrespective of whether the service of targeted recommendations (or automated purchases) falls within the area of application of the Digital Services Act’s liability exemption provisions or whether the matter is left to the discretion of the national legislator, for the purposes of the liability regime, it is advisable to assume that it would still be necessary to verify that the service provider limits itself to the neutral provision of the service, through a purely technical and automated processing of the data, and that it does not perform any function which could give it knowledge of or control over these data [35].
However, it is true that in many cases online platform operators that function as online marketplace operators for the sale of their own and/or third-party products cannot rely on the exemption from liability provided for in the Digital Services Act (if they are subject thereto) or by national law, if any, because they do not play a neutral role in the design of both the website and the messages addressed to the public through their online marketplaces. Nonetheless the success of certain online marketplaces that use “ML” and “DL” for targeted and micro-targeted (“natural”) recommendations and (“sponsored”) advertising indicates their ability to effectively capture consumer preferences offering a variety of products [36], partly due to the possibility of free returns, which is often guaranteed, along with the possibility of checking reviews of other consumers. In addition, the algorithms often recommend products from lesser-known producers, which serve as alternatives to well-known brands, with equivalent quality and lower prices [37]. This indicates that the operation of such algorithms, which are becoming increasingly sophisticated thanks to the integration of “ML” and “DL”, is a tool for competition in the market to the benefit of consumers [38].
4.1. The use of third parties’ trade marks in the advertising of online marketplace operators: case law
In this respect, it may be appropriate to consider case law (albeit from different jurisdictions) which highlights certain interpretative ambiguities which merit further examination in the context of the interpretation of EU trade mark law and EU-harmonized national trade mark law, in order to assess whether the current jurisprudential guidance promotes or hinders the alleged pro-competitive benefits of the activities of these online marketplace operators, while at the same time safeguarding the protected interests of trade mark owners.
These judgments, in particular, have dealt with: 1) advertisements on the website of a search engine operator (Google or Bing), sponsored by a particular online marketplace operator (Amazon) – triggered by the user’s entering in the search bar a (word corresponding to a) third party’s trade mark – containing a link to the website of the same online marketplace; and/or 2) the use of a third party’s trade mark by a particular online marketplace (Amazon), which displays certain results (natural and sponsored) in response to the user’s entering a particular (word corresponding to a) trade mark in the search bar of the same online marketplace. In both cases, a distinction can be made between: a) the hypothesis that the products bearing the searched trade mark are displayed on the website of the online marketplace alongside other competing products; b) the hypothesis that the website of the online marketplace does not display products bearing the trade mark searched for by the user, but only other competing products. In all these cases the online marketplace operator plays an active role, either as an advertiser on another party’s search engine website or as the designer of its own web page (on which the list of search results is displayed) [39]. In both cases, either the design of the ad or the page is related to the activity of the online marketplace operator and no liability exemption applies.
In this respect, an English judgment seems particularly instructive, in which it was held that there was a likelihood of confusion (in the sense of a likelihood of association) in hypothesis 1) above, where the website of the online marketplace did not present products bearing the trade mark searched for by the user (in this case «Lush» for «bath bombs»), but only other competing products (hypothesis b). In particular, the English court held that: «Amazon assumes the consumer is looking for Lush products, or, at least, intended to search for Lush products and thereafter, without a further indication that such products are not available, offers competing products to the consumer. In these circumstances I do not consider the average consumer would ascertain without difficulty that the products he is shown are not the Lush products of the Claimants» [40].
The standard of the average Internet user used by the English judge to assess the likelihood of confusion is that set out in the Google France and Google judgment of the CJEU [41]. The English judge also added that the advertising and investment functions were impaired as well since «[t]hat quality of attracting custom … is bound to be damaged by the use by Amazon of the Lush mark to attract the attention of consumers to and attempt to sell to them the goods of third parties whilst at the same time making no effort at all to inform the consumer that the goods being offered are not in fact the goods of Lush».
Another interesting decision applying the same average consumer standard as Google France and Google comes from the German Federal Court of Justice. The case is about the acquisition by Amazon of a keyword corresponding to a trade mark (Ortlieb) from the operator of an online search engine (Google). As a result, when the trade mark is entered into the search engine’s search bar, Amazon’s ad appears in the sponsored results: the hypothesis 1) above. If the user clicks on the sponsored link, she is directed to the Amazon’s website where, along with Ortlieb products, competing third parties’ products are displayed in the search results: hypothesis a) above [42]. Nonetheless the advertising contains only the Ortlieb trade mark and no reference to equivalent products of others is made. The latter fact has been deemed decisive in the view of the Court of appeal according to which «the origin-indicating function of the trade mark was adversely affected because the relevant public expected to be showed offers from Ortlieb when clicking on the ads. The design of the ads gives the public no reason to expect to be presented with a list of offers that contains, besides Ortlieb products, corresponding products from other manufacturers without specially identifying them» [43].
The German Federal Court of Justice did not find any error in the reasoning of the Court of Appeal and, as a consequence, confirmed the decision [44]. According to the German Federal Court of Justice, «[t]he fact that the reseller offers for sale not only products of the brand manufacturer but also rival products does not preclude a use of the trade mark in advertising so far as the legitimate interests of the trade mark proprietor are safeguarded. If, however, a trade mark is used misleadingly in advertisements following a Google search as a result of the concrete design, so that customers are (also) directed to the offerings of third-party products through this exploitation of the brand’s advertising impact, the trade mark proprietor may prohibit this use of the trade mark» [45].
Neither of these decisions (the English and the German) prevents the reseller from advertising the resale of the trade mark proprietor’s original goods alongside competing products using the same trade mark «provided that the legitimate interests of the trade mark owner are safeguarded» [46]. This seems to mean in practice that the protected functions of the trade mark must not be impaired by the use of the trade mark by the operator of the online marketplace. This is the case though if the content of the advertisement does not clearly indicate that the products of the trade mark proprietor are present on the online marketplace together with products of other manufacturers and all the more so if the products of the trade mark proprietor are completely absent from the list of results displayed by the online marketplace website and only third-party competing products are presented.
Another judgment of the German Federal Court of Justice dealt with scenario 2) and a) above, where a third party’s trade mark (Ortlieb) was used as a keyword by a particular online marketplace (Amazon) which displayed a list of items in response to the user entering that particular (word corresponding to the) same trade mark in the search bar of the same online marketplace. In particular, «[e]ntry of the search term “Ortlieb” in the internal search engine of the Internet platform “amazon.de” on 9 October 2014 generated a list of 1,129 results. Among the hits were offers of products of the plaintiff and other manufacturers, the latter including defendant 1 as well as third parties» [47]. Again, the question was whether the average internet user could without difficulty distinguish between the trade mark proprietor’s offers and those of third parties. Interestingly, the German Federal Court of Justice overturned the Court of Appeal’s decision and referred the case back to the lower court, giving it some guidance. In particular, the Federal Court of Justice stated that, bearing in mind that the likelihood of confusion (or association) in the online environment depends on the design of the advertising and that, according to the CJEU and national case law, the separation between natural results and sponsored advertising is such as to remove any risk of confusion for the average internet user, it added that «stricter requirements must be met by a pure search engine, considering the expectations Internet users have of such a search engine as regards the separation of search results from advertisements, than by a search engine operating within an online shop or Internet market place, where the average Internet user expects to find alternative offers in the results list. Here the court of appeal will have to consider whether the user may know from experience that results lists generated by site-internal search engines do not always show only the actual relevant hits. If this is the case, the user will have no difficulty in distinguishing between products of the trade mark proprietor and products of third parties. This is at least true «when it is sufficiently clear from the way the individual products are displayed in the results list that they bear a different brand» [48].
Along these lines, in a case of keyword advertising involving the search engine Google, an Italian judgment has stated that although there is a large proportion of Internet users who are not specialists, and despite the fact that they have different levels of education, they have «increasing experience and skills in “surfing” that they have acquired through use» [49].
On the other side of the Atlantic, the federal case law that has addressed the issue (in particular the Ninth and Third Circuits of Appeals) has long held that in the context of online shopping «the default degree of consumer care is becoming more heightened as the novelty of the Internet evaporates and online commerce becomes commonplace» [50] adding that «clear labeling can eliminate the likelihood of initial interest confusion in cases involving Internet search terms» [51]. Interesting in this context is a judgment dealing with a trade mark infringement action brought by a manufacturer of military watches (Multi Time Machine, Inc.) based on the fact that, when the term «mtm special ops» corresponding to the plaintiff’s trade mark is entered in the Amazon search bar, no mtm watch is shown but other military watches from other manufacturers appear in the list of results allegedly creating a likelihood of confusion (in the sense of commercial connection) in the mind of the consumer. The Court rejected the action stating that, «[b]ecause Amazon’s search results page clearly labels the name and manufacturer of each product offered for sale and even includes photographs of the items, no reasonably prudent consumer accustomed to shopping online would likely be confused as to the source of the products» [52].
As said by CJEU in the Interflora judgment, the mere fact that some consumers may be confused does not imply that the inattentive and clueless consumer has to be the standard benchmark in evaluating the risk of confusion [53]. As also stated by Multi Time Machine judgment: «Unreasonable, imprudent and inexperienced web-shoppers are not relevant. … To establish likelihood of confusion, the plaintiff must show that confusion is likely, not just possible» [54].
This particular standard of the average consumer, which both in the EU and US legal systems is specific to the digital environment, is particularly welcome because in the online context ensuring a high level of protection against likelihood of confusion (that is protecting also the unreasonable, imprudent and unexperienced consumer) conflicts with the protection of (all the other) consumers’ interest in receiving information about competing offers [55]. In the context of online shopping, in order for the ML and DL to fulfil their promise of more and better consumer choice, it is necessary to uphold a notion of the “average consumer” with sufficient knowledge of the rules of operation of the tool she is using and can therefore be considered reasonably circumspect (and not clueless) [56]. It is argued that the reference agent model should be able to understand the basic mechanism of how the algorithms of online marketplaces work. These marketplaces typically present consumers with products from the brand they are looking for, as well as other products relevant to their interests. Lists of offers the user sees are based on data processed by machine learning (ML) or deep learning (DL) systems. The imposition of transparency rules by the DSA, in particular in Articles 26 and 27 of the DSA, is an indication of the direction in which the law is moving, namely towards ensuring that consumers are informed about the main parameters used by the online marketplace operator to recommend or promote different purchase alternatives. In line with the regulatory model, it can be argued that the consumer is considered a subject capable of understanding the basic principles underlying the functioning of the algorithms which, when a consumer enters a trade mark in a search engine, produce results that do not consist (exclusively) of goods bearing that trade mark. The interpretation advocated here is also consistent with the model of undistorted competition adhered to by EU law, as well as with the constitutional protection of freedom of (commercial) information and freedom of enterprise, also enshrined in EU law.
4.2. The proposed framework for trade mark infringement analysis of the use of third parties’ trade marks in the advertising of online marketplace operators
Having clarified this, it is possible to return to the examination of cases in which the operator of an online marketplace advertises on the website of a search engine by sponsoring the availability on its platform of products distinguished by the trade mark sought by the user within the search mask of the same search engine [hypothesis 1) above]. Moreover, according to this hypothesis, once the internet user has clicked on the URL of the ad and has been redirected to the marketplace site, she will not find any of the products with the brand she is looking for, but only (equivalent) products from other manufacturers [hypothesis a) above]. At this point, the question arises as to whether, faced with such a case, a normally informed and reasonably attentive and circumspect internet user is likely to be confused as to the commercial origin of the products advertised or recommended in the list of offers presented by the online marketplace, even if only in the sense of believing that they are manufacturers economically linked to the trade mark proprietor whose products the user had originally searched for.
In fact, it seems far from unlikely that, in the case just described – and taking as a reference point the consumer model outlined above – the confusion that initially exists on the basis of the text of the advertisement is destined to disappear at a later stage and, in any event, before the user makes any purchase decision. In the same vein, it has also been argued that the consumer’s attention at the time of purchase is typically higher than the attention paid by the consumer at the time of first contact with the advertisement [57]. Assigning importance only to the presence of the sign in the advertisement, while disregarding the stage in which the internet user’s potential purchase decision is made, gives exclusive and independent importance to what the US federal judiciary would not hesitate to refer to as «initial interest confusion», which, as it is currently stated, is «confusion that creates initial consumer interest, even though no actual sale is finally completed as a result of the confusion» [58]. Part of the English jurisprudence at one point adopted such a theory (although not in the context of infringement in double identity cases), and then changed its orientation (in the context of infringement in double identity cases) instead, believing that the CJEU’s approach takes well into account situations that would also fall under the umbrella term «initial interest confusion» in the Internet environment [59]. Even in other jurisdictions, «initial interest confusion» has not been accepted in trade mark law, except in isolated cases, and has been abandoned since the CJEU took a position on keyword advertising [60]. Indeed, according to the CJEU’s jurisprudence in Google France and Google, Bergspechte, Portakabin and Interflora, where there is an alleged infringement of exclusive trade mark rights due to double identity, «the question whether [the origin] function is adversely affected when internet users are shown, on the basis of a keyword identical with a mark, a third party’s ad depends in particular on the manner in which that ad is presented. The function of indicating the origin of the mark is adversely affected if the ad does not enable normally informed and reasonably attentive internet users, or enables them only with difficulty, to ascertain whether the goods or services referred to by the ad originate from the proprietor of the trade mark or an undertaking economically connected to it or, on the contrary, originate from a third party» [61].
On the basis of the principles reaffirmed by the CJEU, one might think that it is irrelevant for the purposes of the analysis of the existence of detriment to the trade mark’s function of indicating origin that, once the user has entered the sponsored website, she will have no difficulty in realising that the website does not contain the goods sought and that, therefore, the initial confusion is easily eliminated at the time of the decision to purchase. Indeed, in the online environment, starting from (and internalising) a principle of transparency established within the harmonized framework for e-commerce, the CJEU requires, even for the purposes of the existence of an injury to the function of indication of origin in cases of double identity, an additional transparency on the part of sellers in their online advertisements, in the sense that if the advertising is so vague that doubts remain or cannot be easily dispelled as to the existence of an economic link between the trade mark proprietor and the third party, an injury of the trade mark proprietor’s distinctive function and, therefore, a likelihood of confusion (in the sense of a likelihood of association) on the part of the public may be claimed [62]. However, it does not seem to be in line with the case law of the CJEU to assume that, even in the absence of a likelihood of confusion in the above sense, (at least) with regard to the existence of an economic link between the trade mark owner and the owner of the goods marked with the different trade mark and offered for sale on the online marketplace, the function of indicating origin can be impaired, especially since the CJEU has never expressly confirmed the exclusive relevance of confusion at the time of the first contact with the advertisement for the purpose of the impairment of the distinctive function of the trade mark. In other words, it has never explicitly attributed relevance, even in the context of double identity, only to the confusion that «creates the initial interest of the consumer». If anything, the existence of the likelihood of confusion at the time of the first contact is an element to be assessed, and in some cases may even be one of the most important elements, but still in the context of an overall analysis aimed at establishing the existence of a likelihood of confusion (also in the sense of a likelihood of association) at the point of sale [63].
The principles laid down by the CJEU regarding the detriment to the distinctive function of the trade mark in the context of keyword advertising must therefore be understood in relation to the specific cases brought before the CJEU and to which the CJEU referred when laying down these principles. In this sense, the abovementioned judgments of the CJEU can therefore be read as meaning that, in the context of keyword advertising, if the abovementioned conditions are met, and in particular having regard to the degree of clarity (or opacity) of the advertising in question, there may be a risk that the average, normally informed reasonably observant and circumspect consumer, at the time of purchase, will believe (or at least will not be able easily to dispel any doubt) that there are economic links between the trade mark proprietor and the third party. Such an assessment is, of course, a matter for the national court. In that sense, it is by no means a foregone conclusion that the national court will, in the light of the circumstances of the particular case, find that there is such a likelihood of confusion (or, rather, association). As the Interflora judgement shows, the English court in that case found that there was a likelihood of association because of the specific characteristics of the trade mark proprietor’s business, which, as is well known, consists of operating a network of florists, and since the third party advertiser had also advertised his flower delivery business, the Court considered that there was a reasonable likelihood of confusion on the part of the public as to the third party’s membership of the trade mark proprietor’s network and, in that factual context, held that there had been an impairment of the trade mark’s distinctive function. Conversely, it can be argued that the average internet user, having entered a keyword corresponding to a trade mark, can expect to receive commercial offers from third parties in addition to those from the trade mark proprietor, either among the sponsored results or the “natural” results, if this is how the marketplace operates. As long as it is clear from the context that these are commercial alternatives to the trade mark holder’s products, there is no likelihood of confusion, even in the broad sense [64].
In this respect, it is reasonable to assume that, applying the standard of the average consumer adopted by the CJEU, there is no likelihood of confusion (or association) if, once inside the marketplace website, the user is able, thanks to the presentation of appropriate information, to understand without difficulty that the commercial offers come from third parties and the trade marks and manufacturers of the offers advertised (or otherwise presented) on the online marketplace are expressly indicated [65]. In this hypothesis, a situation of confusion seems all the less likely the more recognisable the brand name of the product initially sought by the consumer. Indeed, since the brand is well known to the relevant public, it seems unlikely that the seller (who offers her products for sale on the marketplace) would not communicate (or otherwise make easily recognisable) the existence of an economic link with the manufacturer of the brand well known to the public. Indeed, such a link would in itself constitute an incentive for the public to buy. If, on the other hand, such a link is not adequately publicized (and, in fact, we are dealing with marks which are completely unknown or even difficult to pronounce [66]), the reasonably well-informed and reasonably observant and circumspect consumer will expect that the commercial offer comes from a person who has no economic link with the owner of the well-known mark and that, therefore, the commercial offer comes from a third party producer (which the consumer will choose to buy only because it is recommended by the marketplace operator or, which is essentially the same thing, because it ranks high in the search results or, again, because it has a particularly good price or because there are positive reviews or for all of the above reasons together). Such an expectation on the part of a normally informed and reasonably attentive and circumspect user seems all the more reasonable given that the success of many online marketplaces is based on offering a variety of equivalent products that, according to algorithms based on behavioural analysis, may be of interest to the user. In fact, it can be said that this is one of the most distinguishing features of such online marketplaces and is particularly appreciated by consumers as it increases their options. It should also be remembered that the consumer’s attention in the context of purchasing (or, if preferred, at the point of sale) is certainly higher than the attention paid by the same consumer at the time of first contact with the commercial offer through advertising.
Of course, it is still possible, as some national courts have decided in the light of the case law of the CJEU, that in cases of double identity (and, according to BergSpechte, also outside cases of double identity) there is an injury to the distinctive function if there is a failure to comply with the appropriate standards of transparency (as defined from time to time by the case law) that may be expected of advertisers, and that a likelihood of confusion for consumers may be inferred from this failure. This is all the more the case when it comes to: misleading advertising, which announces the presence on the marketplace of commercial offers from a particular manufacturer, when on the other hand such offers are completely absent on the same marketplace; or when the product sought, which at first sight appears to be available, actually turns out to be out of stock, without this being clearly indicated. In other words, in such cases, since the rules of e-commerce require that the relevant information be managed in a clear manner by advertisers, sellers and managers of online marketplaces, if a given advertisement announces the presence of commercial offers with a given brand name on a given online marketplace, the consumer is entitled to expect that the list of commercial offers displayed on that marketplace is composed (at least in part) of products that are economically linked to the brand owner. It follows that, in such cases, the user may actually notice that this is not the case with some difficulty. Moreover, this is a distortion, albeit temporary, of the search process which, if it were the result of generalized behaviour, would make the experience of searching for commercial offers online tedious, if not frustrating. This would have a negative impact on the competitiveness of e-commerce as a whole and, consequently, on the free movement of goods within the EU single market. It is perhaps worth noting, however, that in order to address such a distortion through a harmonized regulatory solution at EU level, it is not necessary to resort to the concept of likelihood of confusion (or detriment to distinctiveness), which is specific to trade mark law (with the risk of overstepping its bounds).
4.3. Bait-and-switch as a form of misleading advertising
In fact, on closer examination, the Directive on misleading and comparative advertising seems to be an appropriate instrument for combating this type of commercial practice [67]. Indeed, on the basis of this legislation, it can also be said that national unfair competition law (although not formally harmonized) has to some extent undergone a kind of indirect harmonisation of the advertising cases covered by this Directive, thus providing injured competitors with an effective and dissuasive private enforcement tool against misleading advertisers. Of course, online marketplace operators who advertise branded products that in fact are not actually available on their platforms are not outside the scope of this discussion. Indeed, it can be argued that in such cases there is a bait-and-switch effect, which is a particularly insidious form of advertising that is also prohibited in the US as false advertising [68]. Such conduct appears particularly insidious to competitors when it is perpetrated by an online marketplace which, thanks to AI-driven behavioural analysis, is able to recommend products to the consumer (who, deceived by the advertising, ends up on its platform) that may nevertheless be of interest to her. In such a case, strictly speaking, there can be no damage to the consumer, who, although a victim of temporary confusion, once “landed” on the marketplace’s website, easily realized that the products proposed there were not the ones originally sought. On the other hand, it is the competitor (the trade mark proprietor) who suffers the damage, since it is likely to lose customers as a result of the third party’s deception. Since the means used by the operator of the online marketplace are undoubtedly unfair in that they are contrary to the most fundamental principles of the proper functioning of markets, the existence of misleading advertising can be affirmed both for the purposes of public and private enforcement under the EU Directive on Misleading and Comparative Advertising and for the purposes of private enforcement under national unfair competition law. For the same reasons, and also in the light of Directive (EU) 2015/2436 and Regulation (EU) 2017/1001, which provide for the protection of trade marks with a reputation even in the case of double identity, it can be said that the advantage that the third party derives from the use of the mark in advertising constitutes an unfair advantage, because although such use does not create confusion in the mind of the public at the point of sale, it does enable it to draw the attention of consumers to its market platform, thereby facilitating the possibility of the user finding commercial offers of interest to her [69].
However, the most delicate issue concerns the case where the operator of an online marketplace advertises on a search engine by sponsoring the availability on its site of products distinguished by the brand name sought by the user in the search mask (hypothesis 1 above). After clicking on the URL of the advertisement and being redirected to the website of the marketplace, the internet user finds the products with the brand name she is looking for (in the quantities and varieties that seem to be legitimate for an online marketplace of the type in question), together with equivalent products from other manufacturers (hypothesis b above). At this point, the question arises as to whether, according to the CJEU’s jurisprudence, there is an infringement of the trade mark’s function as an indication of origin, as held by the German Federal Court, in the case where only the trade mark owner’s products are advertised in the advertisement, while the marketplace website also offers equivalent products from third parties.
In this hypothesis, the commercial offers of the operator of the online marketplace are not presented in a misleading manner. Certainly, in this case, the advertisement does not advertise the presence of other manufacturers’ products, even though they are present in the marketplace’s list of results (sponsored and unsponsored). In other words, it is an omission. However, this omission is not misleading simply because third-party products are also offered within the marketplace. In fact, this is an inherent feature of the marketplace that any consumers with a basic understanding of the shopping environment will recognise. In that hypothesis, it cannot be said that the distinctive function of the mark is infringed, since the advertised goods are actually present on the market and the commercial alternatives are clearly identifiable as such, for the reasons set out above. If this is the case, there also does not appear to be any impairment of the other protected functions of the mark. The advertising function does not appear to be infringed, since such use does not prevent the trade mark proprietor from using the trade mark to attract customers, and the fact that an online marketplace also offers equivalent third-party goods appears to be in line with the normal functioning of competition [70]. Nor is the investment function infringed, since such use of the trade mark does not, except in special cases, damage the commercial image of the trade mark proprietor.
Therefore, if the issue is examined from the point of view of the harmonized framework prohibiting misleading advertising and the national framework on unfair competition, there are, in principle, no grounds for claiming the existence of misleading advertising or unfair competition, except as discussed below. An Italian decision concerning the offline environment seems instructive in this respect. In that case, it was held that «neither misleading advertising nor unfair competition is constituted by the distribution by a hypermarket of an advertising leaflet depicting a certain product, allegedly of high quality and distinguished by a renowned brand, when the actual offer for sale on the counters of the hypermarket concerned not only the product depicted, but also another product, distinguished by the same brand, but allegedly of lower quality» [71]. In this case, the judge considered that «there was no deception or unfairness […] because even the high-quality T-shirt was actually present in the sales outlets, with the result that the consumer, who may have been specifically attracted by the prospect of buying the “top” model (and therefore not the low-quality one), would still have been able to recognize it by examining the label, even if the products of different quality were very similar in appearance» [72]. Relevant to the online environment, an English decision on a case of double identity bait-and-switch, concerning the referential use of the trade marks of a manufacturer and former supplier (Pliteq) by a former non-exclusive distributor (iKoustic), which sold various hearing aid products of the former supplier on its e-commerce site. After the termination of the distribution agreement, the now former distributor continued to sell stock of products bearing the former supplier’s trade marks («GenieClip» and «GenieMat») and had at some point also started to sell competing products under its own trade marks («MuteClip» and «MuteMat») [73]. As long as the former supplier’s branded products were available for sale online, the former distributor advertised the availability of the former supplier’s branded products on its e-commerce site on Google (through Google’s dynamic advertising service) [74]. The former distributor also communicated on its website the availability (or lack of availability) of the former supplier’s products that it actually offered for sale on its e-commerce site [75]. In this case, the judge found that there was no impairment of the protected functions of the trade mark. In particular, the distinctive function was not affected because the users of the website were aware of the different origin of the products sold by the former distributor, including through specific notices on the e-commerce site [76]. According to the same decision, there was no detriment to the advertising function because it had not been shown that the use by the former distributor affected the suitability of the former supplier’s mark as an advertising tool or business strategy [77]. Furthermore, again according to that decision, there is no injury to the investment function, since some users may have turned to the former distributor’s products under conditions of fair competition [78].
However, if the above is accepted, it should be borne in mind that it is quite possible that in the online marketplace: (a) the products available under the trade mark sought by the user are of a different type than those sought, and instead products of the type sought by the user but from other manufacturers are displayed; or (b) the branded products sought by the user are not of the variety or availability that would legitimately be expected on the basis of the advertising and normal operation of this type of website, and instead numerous offers of competing products (perhaps recommended or highlighted by the same marketplace) are displayed, or in any case (c) it is possible that the offers of the proprietor of the trade mark sought by the user are few in number, even though the advertising only promoted the offers of such products, and at the same time the proprietor of the marketplace presents many offers of third parties (perhaps recommended or highlighted by the same marketplace). In such cases, the prohibition of misleading advertising is even more clearly called into question. In particular, in such cases, there is no infringement of the function of the trade mark as an indication of origin, since the clear traceability of the products on the online marketplace to the different producers through the different trade marks of each product is not at issue. However, there is an instrumental use of the presence of the advertised products on the market in order to induce the consumer to choose (or be interested in) the other products presented as alternatives (or, at any rate, of possible interest to the consumer). However, in the case of hypotheses (a) and (b) in particular, it must first be ascertained whether the advertising in question is not of the bait-and-switch type, as defined in Annex 1 no. 6 of the Unfair Commercial Practices Directive (2005/29/EC as amended), where the advertiser «mak[es] an invitation to purchase products at a specified price and then: (a) refus[es] to show the advertised item to the consumer; or (b) refus[es] to take orders for it or deliver it within a reasonable time […] with the intention of promoting a different product». However, it is quite possible that some of the elements that the Directive considers to be misleading «in all circumstances» are missing. For example, in the case of advertising, the promotion of a certain price may be absent and yet, on the basis of the circumstances of the specific case, the existence of a misleading commercial practice to the detriment of the consumer may be affirmed. In this respect, the element of «intention» required by the rules may also result from the circumstances of the concrete case. More generally, in the hypotheses under (a), (b) and (c), the diversion of customers by unfair means should, as a general rule, not be seen as an infringement of the protected functions of the trade mark under harmonized EU trade mark law, but from the perspective of unfair advantage under the protection of trade marks with a reputation as well as from the perspective of the prohibition of misleading advertising under Directive 2005/29/EC or, again, under Directive 2006/114/EC and the national framework for the enforcement of unfair competition.
[1] This work was carried out as part of a research project funded under the PRIN PNRR programme P2022ZZSY7, Promises and Perils of NFT, DLT and AI in managing IPRs. Questo lavoro è stato realizzato nell’ambito di una ricerca finanziata con fondi PRIN PNRR P2022ZZSY7, Promises and Perils of NFT, DLT and AI in managing IPRs.
Not so much when I wrote an article on the topic (in Italian), see D. Arcidiacono, Gli atti di sfruttamento dei marchi da parte delle intelligenze artificiali. Prime riflessioni, in AIDA, 2018, 150. Some of the remarks made in this contribution are also present in that first article. An exception at the time was represented by L. Curtis, R. Platts, AI is coming and it will change trade mark law, in ManagingIP.com, (accessed on 8th December 2017), and in hgf.com (accessed on 8th April 2025), available at www.hgf.com.
[2] D.S. Gangjee, A quotidian revolution: artificial intelligence and trade mark law, in Research Handbook on Intellectual Property and Artificial Intelligence, edited by R. Abbot, Cheltenham UK-Northampton Massachusetts, Elgar Publ., 2022, 325; D.S. Gangjee, Eye, robot: artificial intelligence and trade mark registers, in Transition and Coherence in Intellectual Property Law, edited by N. Bruun, G. B. Dinwoodie, M. Levine, A. Ohly, Cambridge-New York-Port Melbourne-New Delhi-Singapore, Cambridge Univ. Press, 2021, 174; A. Moerland, C. Freitas, Artificial intelligence and trade mark assessment, in Artificial intelligence and intellectual property, edited by J.-A. Lee, R. Hilty, K.-C. Liu, Oxford, Oxford Univ. Press, 2021, 266; S. Katyal, A. Kesari, Trademark Search, Artificial Intelligence, and the Role of the Private Sector, 35 Berkley Tech. L. J., 2020, 501. Interesting empirical studies has been conducted by S. Adarsh, E. Ash, S. Bechtold, B. Beebe, J. Fromer, Automating Abercrombie: Machine-learning trademark distinctiveness, in J. of Empirical Legal Studies, 2024, at onlinelibrary.wiley.com and J. Goodhue, Y. Wei, Classification of Trademark Distinctiveness using OpenAI GPT 3.5 model (February 8, 2023), at ssrn.com (accessed on 22 September 2025).
[3] See D. Lim, Computational Trademark Infringement and Adjudication, in Research Handbook on Intellectual Property and Artificial Intelligence, edited by R. Abbot, Elgar Publ., 2022, 259; R. Batty, Trade mark infringement and artificial intelligence, available at ssrn.com (accessed on 8th April 2025).
[4] See M. Senftleben, Trademark law, AI-driven behavioral advertising, and the Digital Services Act: toward source and parameter transparency for consumers, brand owners, and competitors, in Research Handbook on Intellectual Property and Artificial Intelligence, edited by R. Abbot, Cheltenham UK-Northampton Massachusetts, Elgar Publ., 2022, 309.
[5] Particularly so L. Curtis, R. Platts, (nt. 1); A. Moerland, C. Kafrouni, Online shopping with artificial intelligence: what role for trade marks?, in Research Handbook on Intellectual Property and Artificial Intelligence, edited by R. Abbot, Cheltenham UK-Northampton Massachusetts, Elgar Publ., 2022, 290. A possible starting point could be represented by EU Regulation 2024/1689 comprising harmonized rules on artificial intelligence (“ArtificiaI Intelligence Act”). Nonetheless at a closer look the Artificial Intelligence Act does not seem to introduce any new regulatory elements in this respect. In the context of recommendation systems or targeted advertising, artificial intelligences are not considered high risk and, consequently, do not appear to be affected by the AI Regulation. However, there is a possibility that they may pose risks to competition in downstream markets if, for example, recommendation systems discriminate unreasonably and non-transparently between operators. In such an event, this would be a competition issue for which Recital 45 of the AI Act emphasizes that the Regulation is without prejudice to the application of the rules of EU competition law and consideration can be given to the Digital Markets Act. In any case the issue does not directly concern the use of trade marks by recommendation systems or automated purchasing systems made possible by the use of artificial intelligence. Section 2(5) of the IA Act states that it «shall be without prejudice to the provisions relating to the liability of intermediary lenders set out in Regulation (EU) 2022/2065 of the European Parliament and of the Council». A different set of issues arises when we consider the impact of online behavioural advertising on fundamental rights. In this respect, see, also for further appropriate citations, E. Izyumenko, M. Senftleben, N. Schutte, E. G. Smit, G. van Noort, L. van Velzen, Online behavioural advertising, consumer empowerment and fair competition: Are the DSA transparency obligations the right answer?, available at ssrn.com (accessed on 8th April 2025).
[6] The subject has been examined, among the others, by A. Moerland, C. Kafrouni, (nt. 5); V. N. Sevastianova, Trademarks in the Age of Automated Commerce: Consumer Choice and Autonomy, in IIC, 2023, 1561-1589; J. Randakevičiūtė-Alpman, The Role of Trademarks on Online Retail Platforms: An EU Trademark Law Perspective, in GRUR Int., 2021, 633-643.
[7] Differently see M. Grynberg AI and the “Death of Trademarks”, in Kentucky L. J., 2019, 199 and A. Roy, A. Marsoof, Removing the Human from Trademark Law, in IIC, 2024, 727-761.
[8] M. Grynberg, (nt. 7), 200 ff.; see also D. Lim, (nt. 3), 261 who remarks that «AI diminishes the role of trademark’s function of simplifying information»; V. N. Sevastianova, (nt. 6); A. Moerland, C. Kafrouni, (nt. 5), 297; J. Randakevičiūtė-Alpman, (nt. 6), 641; C. H. Farley, Trademarks in an Algorithmic World, in Washington L. Rev., 2023, 1123; lastly, with particular reference to the functioning of Amazon’s online marketplace, see also J. C. Fromer, M. P. McKenna, Amazon’s Quiet Overhaul of the Trademark System, available at ssrn.com (accessed on 8th April 2025), where the authors remark that «[i]f Amazon’s algorithm and consumer reviews are as good or better at conveying information about products, then the informational value of brands is less important».
[9] M. Grynberg, (nt.7); see also J. C. Fromer, M. P. McKenna, (nt. 8).
[10] A. Roy, A. Marsoof, (nt. 7), 727 ff. On the interrelation between AI, IoT and blockchain see e.g. Internet of Things, Artificial Intelligence and Blockchain Technology, edited by R. Lakshmana Kumar, Y. Wang, T. Poongodi, A.L. Imoize, Springer, 2021.
[11] Such as Amazon’s Dash Replenish service, see A. Moerland, C. Kafrouni, (nt. 5), 295.
[12] See R. Hilty, Big Data: Ownership and Use in the Digital Age, in Intellectual Property and Digital Trade in the Age of Artificial Intelligence, Global Perspectives for the Intellectual Property System, edited by X. Seuba, C. Geiger, J. Pénin, CEIPI-ICTSD, Geneva/Strasbourg, num. 5, 2018, 87, highlighting the possibility of «fridges automatically purchasing missing food and beverages»; the same example is used by A. Moerland, C. Kafrouni, (nt. 5), 295.
[13] See V. N. Sevastianova, (nt. 6).
[14] See, among the others, M. Grynberg, The consumer duty of care in trademark law, in Research handbook on trademark law reform, edited by G.B. Dinwoodie, M.D. Janis, Cheltenham UK-Northampton Massachusetts, Elgar Publ., 2021, 329.
[15] See A. Moerland, C. Kafrouni, (nt. 5), 298 who remark that «while trade marks may no longer serve the purpose of simplifying information and thereby making the purchasing process more effective, they still convey important information to the consumer when she consumes the product. Consumer experiences feed back into the next purchasing process, thereby maintaining the relevance of human decisions for future purchases».
[16] The protection of consumer choice is traditionally held to be one of the fundamental elements of the undistorted competition model enshrined in primary EU law. See, lastly, among the others, A. Kur, N. Lee, A. Tischner, Fairness in Intellectual Property Law, Cheltenham UK-Northampton Massachusetts, Elgar Publ., 2024, 112 ff.
[17] See, in particular, Section 27 DSA.
[18] On further considerations, under the perspective of the preservation of consumer autonomy, see V. N. Sevastianova, (nt. 6); the constitutional relevance of consumer autonomy under EU and ECHR law is highlighted by E. Izyumenko, M. Senftleben, N. Schutte, E.G. Smit, G. van Noort, L. van Velzen, (nt. 5), where some remedies to «burst the bubble» are proposed and examined.
[19] Along the same lines, see A. Moerland, C. Kafrouni, (nt. 5), 298.
[20] See M. Senftleben, Trade Mark Protection: A Black Hole in the Intellectual Property Galaxy, in IIC, 2011, 383; A. Kur, M. Senftleben, European Trade Mark Law. A Commentary, Oxford, Oxford Univ. Press, 2017, 339; on the continuing need to protect the ancillary functions of trade marks in the context of AI-assisted purchases see A. Moerland, C. Kafrouni, (nt. 5), 299 ff.
[21] As evidenced by the research of A. Moerland, C. Kafrouni, (nt. 5), 296 ff.
[22] Let alone the fact that a choice (or suggestion) made by artificial intelligence may be based on uncertain and unverifiable data. In the case of homogeneous products in particular, choices depend on differences that are not objectively relevant, but rather on other contingent factors unrelated to the quality, reliability or safety of a given product.
[23] See M. Wan, J. Ni, R. Misra, J. McAuley, Addressing Marketing Bias in Product Recommendations, 2019, available at arxiv.org (accessed on 8th July 2025).
[24] M. Wan, J. Ni, R. Misra, J. McAuley, (nt. 23), where suggestions on how to address the problem are made.
[25] See D. Arcidiacono, (nt. 1), 166. See, on this point, A. Moerland, C. Kafrouni, (nt. 5), 306 who, for less sophisticated AI mechanisms with high consumer involvement, suggest that «the concept of the average consumer as used by CJEU in its keyword advertising judgments may still be relevant». See also, among the others, P. Neidinger, J. Wildgans, Liability of providers of intermediary services, in New Digital Services Act: A Practitioner’s Guide, edited by K. Torsten, Baden-Baden: Nomos/Hart Publ., 2024, 36, para. 19, who underline that – according to CJEU, 23 March 2010, Google France SARL and Google Inc. v Louis Vuitton Malletier SA et al., Joined cases C-236/08 and C-238/08 – «Google’s referencing service “AdWords” constituted an information society service within the meaning of Art. 2(a) of E-Commerce Directive in conjunction with Art. 1(2) of Directive 98/34/EC, which in principle could also benefit from the liability exemption of Art. 14 of the E-Commerce Directive if the conduct of that service provider [is] limited to that of an “intermediary services provider” within the meaning intended by the legislature in the context of Section 4 of [the E-Commerce Directive]».
[26] See D. Arcidiacono, (nt. 1), 166 ff.
[27] See D. Arcidiacono, (nt. 1), 167.
[28] See D. Arcidiacono, (nt. 1), 167. See also D. Lim, (nt. 3).
[29] See D. Arcidiacono, (nt. 1), 167.
[30] See D. Arcidiacono, (nt. 1), 167. See also A. Moerland, C. Kafrouni, (nt. 5), 305.
[31] See D. Arcidiacono, (nt. 1), 167. See also D. Lim, (nt. 3).
[32] This implementation eliminates the risk of disclosure of proprietary information. This is based on the premise that the aforementioned secrets consist of the methods and criteria used by artificial intelligence systems to determine the recommendation of certain goods or services to retail customers or the direct procurement of goods or services.
[33] See D. Arcidiacono, (nt. 1), 167 ff. For example, systems for visual trade mark searches (i.e. the acquisition, reception and processing of images corresponding to trade marks) still have some difficulties (see e.g. L. Knock, M. Lux, Confusing Similarity between Visual Trademarks. A Dataset Based on USTTAB Examinations, 92, Proceedings of the OAGM&ARW Joint Workshop 2017, in diglib.tugraz.at, (30 May 2018 10:13 AM EDT), diglib.tugraz.at. In addition, several attempts have been made to perfect systems capable of perceiving conceptual similarities between trade marks, see e.g. F. M. Anuar, R. Setchi, Y.-K. Lai, A Conceptual Model of Trademark Retrieval based on Conceptual Similarity, 450-459, 23 Procedia Comp. Sc. (2013), available at www.sciencedirect.com, (accessed on 30th May 2018). Clearly, these and other limitations must be taken into account in the construction of the agent model discussed in these pages. Similarly, it must be borne in mind that artificial intelligence is structurally incapable of receiving suggestive messages and that it would therefore make no sense to speak of damage to the distinctiveness or reputation of marks with a reputation or unfair advantage where the purchase decision is attributable to the “decision” of an artificial intelligence. If anything, the “reputation” of a mark can be established indirectly, in that the fact that, in certain sectors or for certain goods, users usually prefer goods bearing a certain mark with a “reputation”. That constitutes data which, as such, can be analytically processed and taken into account (or not) by the artificial intelligence for the purposes of the “decision” to carry out the targeted recommendation or the automated purchase.
[34] See D. Arcidiacono, (nt. 1), 168.
[35] See D. Arcidiacono, (nt. 1), 168. CJEU, 23 March 2010, Google France SARL and Google Inc. v Louis Vuitton Malletier SA et al., Joined cases C-236/08 and C-238/08, para. 106, in examining Google’s Adwords under the EC Directive 2000/31 legal framework, has established the conditions for the applicability of the exemption regime to search engines, qualifying keyword advertising in web search results as hosting. On the condition of “neutrality” see M. Husovec, Principles of the Digital Services Act, 101 ff., 139 ff. (Oxford Univ. Press, 2024). According to the author a line must be drawn between «advertising results that are clearly covered as hosting following Google France ruling» on the one side, and «search suggestions delivered via autocomplete that are not because they are content produced by search engines themselves», the most delicate issue being if, and at what conditions, natural search results indexing has to be qualified as hosting service.
[36] See Cosmetic Warriors and Lush v. Amazon.co.uk and Amazon EU [2014] EWHC 181 (Ch), paras 21-23. See also Multi Time Machine, Inc. v. Amazon.com, Inc., 804 F.3d 930 (3rd Cir. 2015): «By going beyond exactly matching a user’s query to text describing a product, Amazon’s search function can provide consumers with relevant results that would otherwise be overlooked».
[37] Partially doubtful on this J. Randakevičiūtė-Alpman, (nt. 6).
[38] This of course is true if the algorithm effectively displays to the consumer the best options available at the lesser price. However, on this assertion see J. Randakevičiūtė-Alpman, (nt. 6); see also M. Wan, J. Ni, R. Misra, J. McAuley, (nt. 23).
[39] See on this Cosmetic Warriors and Lush v Amazon.co.uk and Amazon EU [2014] EWHC 181 (Ch).
[40] Cosmetic Warriors and Lush v Amazon.co.uk and Amazon EU [2014] EWHC 181 (Ch), para. 68.
[41] According to CJEU, 23 March 2010, Google France SARL and Google Inc. v Louis Vuitton Malletier SA et al., Joined cases C-236/08 and C-238/08: «the trade mark’s function of indicating origin is adversely affected if it is not possible for normally informed and reasonably attentive internet users, or only with difficulty, to ascertain whether the goods or services referred to by the ad originate from the proprietor of the trade mark or an undertaking economically connected to it or, on the contrary, originate from a third party».
[42] BGH, 25 July 2019, Case no. I ZR 29/18, Ortlieb v. Amazon (hereinafter also “Ortlieb II”), in GRUR Int., 2020, 399-404.
[43] The quoted passage is in BGH, 25 July 2019, Case no. I ZR 29/18, Ortlieb II, para. 29, 399, 402 69(4) GRUR Int. (2020) referring to the Court of appeal’s legal reasoning. For the original Court of Appeal passage see OLG München, 11 Jan. 2018, 29 U 486/17, paras 49 and 50 where the Court affirmed: «In the present case, the main function of the trade mark, namely its function of indicating origin, is impaired by the use of the sign in the ad, which was linked to a list also containing products from other manufacturers. […] The relevant public expects that when they click on the ads in question, they will be shown offers for the products advertised in the ads […]. In view of the design of the ads, the public has no reason to assume that when they click on the ad, they will be presented with an overview of offers which, without any special indication, includes offers from other manufacturers on an equal footing with Ortlieb products» (own translation).
[44] It is well established in law that «statements on the perception of the public, essentially to be concluded by the fact-finding court, may only be reviewed in terms of whether the court in its assessment violated rules of logic or experience or left fundamental circumstances unconsidered» see BGH, 25 July 2019, Case no. I ZR 29/18, Ortlieb II, in GRUR Int., 2020, 399-402. That brought also to a finding of a legitimate reason for the trade mark owner to oppose to the further commercialization of the items sold on the Amazon’s platform.
[45] BGH, 25 July 2019, Case no. I ZR 29/18, Ortlieb II, in GRUR Int., 2020, 399, official headnote.
[46] In this vein see BGH, 25 Jul. 2019, Case no. I ZR 29/18, Ortlieb II, in GRUR Int., 2020, 399 403, official headnote and para. 35.
[47] BGH, 15 Feb. 2018, Case no. I ZR 1138/16, Ortlieb I, in in IIC, 2018, 1121, 1122, para. 3.
[48] BGH, 15 Feb. 2018, Case no. I ZR 1138/16, Ortlieb I, in IIC, 2018, 1121, 1131, para. 56. Italics added.
[49] Milan Court of Appeal, 19 Apr. 2019, in Riv. dir. ind., II, 2019, 3 ff., with comment of Loconsole,7 ff.
[50] See Network Automation, Inc. v. Advanced Systems Concept, Inc., 638 F.3d 1137, 1152 (9th Cir. 2011). Along these lines, regarding keyword advertising, Lerner & Rowe, PC v. Brown Engstrand & Shely LLC, 119 F.4th 711 (9th Cir. 2024): «those accustomed to online shopping are typically savvy enough to differentiate between search engine results».
[51] See e.g. Multi Time Machine v. Amazon, 804 F.3d 930, 937 (3rd Cir. 2015); Network Automation, Inc. v. Advanced Systems Concept, Inc., 638 F.3d 1137, 1154 (9th Cir. 2011): «clear labeling “might eliminate the likelihood of initial interest confusion that exists in this case”»; Lerner & Rowe, PC v. Brown Engstrand & Shely LLC, 119, 722 F.4th 711 (9th Cir. 2024).
[52] Multi Time Machine v. Amazon, 804 F.3d 930, 938 (3rd Cir. 2015). The Superior Court of Justice (Superior Tribunal de Justiça) in Brazil adopted a different perspective in Case No. 2.012.895–SP on 8 August 2023, as reported in IIC, 2024, 1721 ss., 1726, specifically stating that «there is confusion for the average person, who does not pay much attention to the details of a website» (para. 41). This is consistent with the assertion of a «heightened vulnerability of consumers in the digital environment» (para. 41). Consequently, it is not surprising that, according to this judgment, keyword advertising is prohibited by unfair competition law because it engenders a likelihood of confusion for the public (as well as being considered parasitic behaviour).
[53] See CJEU, 22 September 2011, Interflora, Inc. v. Marks & Spencer, plc, Case C-323/09, ECLI:EU:C:2011:604, para. 50.
[54] Multi Time Machine v. Amazon, 804 F.3d 930, 938 (3rd Cir. 2015).
[55] See M. Grynberg, (nt. 14), 334-336. See, more generally, on the adverse effects of permitting trade mark holders to control the use of their trade marks in third parties’ advertising, R. G. Burrell, The Exceptional Nature of the Right to Control Use in Advertising, in Charting Limitations on Trademark Rights, edited by H. Sun, B. Beebe, Oxford, Oxford Univ. Press, 2023, 186 ff.
[56] See also L. Giordani, Consumer Attention in the Digital Era: Modernizing the Likelihood of Confusion Standard, in GRUR Int., 2025, 724 ff., 731: «Regardless of the product itself, modern consumers often demonstrate fairly high levels of attention, particularly due to their active use of digital tools during the purchasing process. …In particular, for experience and credence goods, it is more likely – based on practical reasoning – that consumers will engage with digital tools to process information, thereby exhibiting higher levels of attentiveness». The author contends that the average consumer, relevant for the likelihood of confusion test, should not be defined primarily by the intrinsic characteristics of the product – such as its price – but rather by acknowledging that contemporary consumers, shaped by their constant use of digital tools in both online and offline purchasing environments, tend to display a heightened level of attentiveness.
[57] I. Fhima, D. S. Gangjee, The confusion test in European trade mark law, Oxford Univ. Press, 2019, 218. According to the authors the link between the attention that can be expected from the consumer and the «transactional context» emerged clearly in CJEU, 12 January 2006, Ruiz-Picasso v. OHIM, Case C-361/04 P, ECLI:EU:C:2006:25.
[58] See, recently, Jim S. Adler, P.C. v. McNeil Consultants, L.L.C., 10 F.4th 422, 424 (5th Cir. Tex. 10 August 2021). See also B. Beebe, Trademark Law. An Open-Access Casebook, 2024, 529, available at BeebeTMLaw-v11-digital-edition-2024.pdf (accessed on 9th April 2025): «Initial interest confusion remains a highly controversial basis for a finding of infringement, one which courts typically resort to only in a limited set of contexts. Courts appear to be more likely to find initial interest confusion if the defendant has engaged in patently bad faith “bait and switch” sales practices or in conduct akin to intentional cybersquatting, if the relevant consumers are unsophisticated, or if the defendant competes directly with the plaintiff».
[59] For an explicit acceptance of the theory of initial interest confusion, see Och-Ziff Management Europe & Anor v OCH Capital LLP & Ors [2010] EWHC 2599 (Ch). However, for a rejection of the initial interest confusion theory, see Interflora Inc & Anor v Mark and Spencer plc [2014] EWCA Civ 1403, paras. 154-155. For further references, see also I. Fhima, D. S. Gangjee, (nt. 57), 205 ff.
[60] For the Italian system in the context of online sales, and in particular for the case where a domain name corresponding to a reputed trade mark is adopted, see Napoli Court of first instance, 7 July 2005, in DeJure database.
[61] See CJEU, 23 March 2010, Google France SARL and Google Inc. v Louis Vuitton Malletier SA et al., Joined cases C-236/08 and C-238/08, paras 83-84; CJEU, 25 March 2010, Die BergSpechte Outdoor Reisen und Alpinschule Edi Koblmüller GmbH v Günther Guni and trekking.at Reisen GmbH, Case C-278/08, para. 35; CJEU, 8 July 2010, Portakabin Ltd and Portakabin BV v. Primakabin BV, Case C-558/08, para. 34; a partially different formula is used by CJEU, 22 September 2011, Interflora Inc., Interflora British Unit v Marks & Spencer plc, Flowers Direct Online Ltd, Case C-323/09, para. 44 where a reference is made to «reasonably well-informed and reasonably observant internet users». «Similarly, in the case where the ad, while not suggesting the existence of an economic link, is vague to such an extent on the origin of the goods or services at issue that normally informed and reasonably attentive internet users are unable to determine, on the basis of the advertising link and the commercial message attached thereto, whether the advertiser is a third party vis-à-vis the proprietor of the trade mark or, on the contrary, economically linked to that proprietor, the conclusion must be that there is an adverse effect on that function of the trade mark»: CJEU, 23 March 2010, Google France SARL and Google Inc. v Louis Vuitton Malletier SA et al., Joint Cases C-236/08 and C-238/08, paras 89-90; CJEU, 25 March 2010, Die BergSpechte Outdoor Reisen und Alpinschule Edi Koblmüller GmbH v Günther Guni and trekking.at Reisen GmbH, Case C-278/08, para. 36; CJEU, 8 July 2010, Portakabin Ltd and Portakabin BV v. Primakabin BV, Case C-558/08, para. 35; here again CJEU, 22 September 2011, Interflora Inc., Interflora British Unit v Marks & Spencer plc, Flowers Direct Online Ltd, Case C-323/09, para. 45, uses the different formula of the «reasonably well-informed and reasonably observant internet users».
[62] Critical of such a jurisprudential development, encouraged by the Google France and Google and Berspechte CJEU judgments, see M. Senftleben, Adapting EU trademark law to new technologies: Back to Basics?, in Constructing European Intellectual Property: Achievements and New Perspectives, edited by C. Geiger, Elgar Publ., 2013, 137, 162: «the CJEU imposed new obligations on advertisers with regard to the prevention of consumer confusion», registering in this sense «a shift from proof of likely confusion by the trademark owner to an obligation on all third parties to secure market transparency when using keyword advertising services».
[63] That being said, for the relevance of post-sale confusion in the English case law see Iconix Luxembourg Holdings SARL v Dream Pairs Europe Inc & Anor [2024] EWCA Civ 29.
[64] For an indication of this, see also recent Norwegian case law, Supreme Court of Norway, HR-2021-2479-A (Bank Norwegian), in GRUR Int., 2022, 835 ff., 837: «The advertisements on Google are marked with the word “advertisement” or “Ad” in bold on the first line of text preceding the link to the URL. The text of the advertisement is underneath. The advertisement may contain search words that were used during the search, but this is not always the case. Bank Norwegian’s advertisements do not contain the disputed paid search words. After a change of practice in 2019, Bank Norwegian’s name is always stated in the heading of the advertisement’s text».
[65] This is to some extent implied by Iconix Luxembourg Holdings SARL v Dream Pairs Europe Inc & Anor [2024] EWCA Civ 29 which takes into account the context of use of the allegedly infringing sign as relevant when assessing the likelihood of confusion at the point of sale. The images and information provided on the Amazon website were deemed sufficient to dispel any risk of confusion, particularly because, in the purchasing context, the differences between the conflicting marks are evident. As a result, the relevant issue was whether, in a post-sale context, a likelihood of confusion could still be affirmed (footwear trade marks involved).
[66] This is often the case with products on the Amazon marketplace, see J.C. Fromer, M.P. McKenna, (nt. 8).
[67] See I. Fhima, D.S. Gangjee, (nt. 57), 225.
[68] See, among the others, D.A. Friedman, Explaining “Bait-and-Switch” Regulation, in William & Mary Bus. L. Rev., 2013, 575 ff.
[69] On this point, see I. Fhima, D.S. Gangjee, (nt. 57), 225-226.
[70] On this point, see in particular A. Kur, Trademarks Function, Don’t They? CJEU Jurisprudence and Unfair Competition Principles, in IIC, 2014, 434, 442. According to the author the ancillary functions of trade marks are infringed only in «exceptional cases» where the interests of the trade mark proprietor are «seriously jeopardized by a conduct that is incompatible with fair and sound competition» (italics in the original text).
[71] Venice Court of first instance, 22 Jan. 2010, in Giur. ann. dir. ind., 2010, no. 5524/1.
[72] Comment to Venice Court of first instance, 22 Jan. 2010, in Giur. ann. dir. ind., 2010, no. 5524/1.
[73] Pliteq Inc v Ikoustic Ltd [2020] EWHC 2564 (IPEC), para. 23: «from a certain date and for one year, the sale of both products took place without the consent of the former licensor».
[74] Pliteq Inc v Ikoustic Ltd [2020] EWHC 2564 (IPEC), para. 54 ff. Google’s dynamic ads service is, as the applicants state, «a fundamentally different process to the generation of a sponsored ad as a result of the use of a keyword. Instead of selecting a Mark as a keyword, dynamic ads are generated through Google’s analysis of the content of the iKoustic website and other parameters, including the searching history of the person searching which […] could affect the wording of the ad Google generates» (Pliteq Inc v Ikoustic Ltd [2020] EWHC 2564 (IPEC), para. 55). The judge found that, instead, when Pliteq products were sold out, the former distributor removed the web page containing the commercial offer. The judge also found that «[g]iven the Defendants’ prompt removal of the pages from the website, it does not seem to me that the failure to use the negative matching facility shows that the Defendants’ real motivation was to continue to attract customers for their goods by using the Marks on those pages, as opposed to inadvertence or lack of full understanding of the failure to negatively match those pages»: Pliteq Inc v Ikoustic Ltd [2020] EWHC 2564 (IPEC), para. 74.
[75] Pliteq Inc v Ikoustic Ltd [2020] EWHC 2564 (IPEC). If there were low stocks of a particular product, the former supplier’s website would highlight the situation, see Pliteq Inc v Ikoustic Ltd [2020] EWHC 2564 (IPEC), paras 70-76: «where stocks of a particular product ran low, the message was displayed on the relevant website stating that the product could not be sold online».
[76] Pliteq Inc v Ikoustic Ltd [2020] EWHC 2564 (IPEC), para. 88.
[77] Pliteq Inc v Ikoustic Ltd [2020] EWHC 2564 (IPEC), para. 91.
[78] Pliteq Inc v Ikoustic Ltd [2020] EWHC 2564 (IPEC), para. 91. The only exception was a web page where the former distributor had claimed that the former supplier’s products were out of stock and contained a link to the MuteClip branded product, described as iKoustic’s «new alternative insulation clip» (Pliteq Inc v Ikoustic Ltd [2020] EWHC 2564 (IPEC), para. 30). According to the Court, such conduct does not constitute fair competition because it is a use of the proprietor’s trade mark to promote its own products, since the design of the page was such as to encourage consumers to look for alternatives to the proprietor’s products, see Pliteq Inc v Ikoustic Ltd [2020] EWHC 2564 (IPEC), paras 91, 97.