Malicious Selling Strategies in E-Commerce Livestream: A Case Study of Alibaba’s Taobao and ByteDance’s TikTok

Livestreaming has been widely used as an engaging way to connect community members (Hamilton et al., 2014a; Lu et al., 2018b). Previous studies have explored engagement motivations, participation demographics, and engagement levels in various livestream communities, including livestreams for entertainment, education (Chen et al., 2019a; Hamilton et al., 2018), skill improvement (Lu et al., 2018a), programming (Faas et al., 2018), cultural practices (Lu et al., 2019), creative activities (Lu et al., 2018b), and online communications (Taylor, 2018). Prior research has explored reasons for livestream participation from both streamer perspectives (Kaytoue et al., 2012; Lottridge et al., 2017; Pellicone and Ahn, 2017) and viewer perspectives (Scheibe et al., 2016; Greenberg, 2016; Hamilton et al., 2014a), how people formed communities between streamers and viewers via livestreams (Hamilton et al., 2014b), and different types of interactions within livestreaming communities (Greenberg, 2016; Lessel et al., 2017; Pires and Simon, 2015; Zhu et al., 2017). For example, streamers on livestream platforms such as Twitch, broadcasted the playing of video games to share gameplay skills and common interests with viewers (Hamilton et al., 2014a; Kempe-Cook et al., 2019; Pellicone and Ahn, 2017; Sher and Su, 2019; Wohn et al., 2018). Prior work has also analyzed viewers’ negative behaviors during livestreams and how livestreaming platforms allowed viewers to report inappropriate content (Wohn, 2019). Livestream platforms have also integrated communication channels to increase viewer engagement (Glickman et al., 2018; Lessel et al., 2017). This high degree of interactivity and engagement results in some viewers being more willing to purchase products created by streamers during a livestream (Lu et al., 2019) or the merchandise designed by the streamers (Sher and Su, 2019). This suggests that there is a great potential for livestreaming to be a new medium for e-commerce.

Livestreaming e-commerce, or livestream shopping, has attracted researchers’ attention in recent years. Some research has focused on viewers’ (i.e., buyers) motivations: why they watch livestreams and buy products during them (Cai et al., 2018; Xu et al., 2020; Li et al., 2020; Cai and Wohn, 2019; Su, 2019). By analyzing self-reported data, scholars found that utilitarian and hedonic values were two of the main values that motivated customers to shop during livestreams. Utilitarian value was related to purchase needs, whereas hedonic value was related to affective commitments to streamers (Cai et al., 2018; Li et al., 2020). Researchers have also found that trust was an intermediate variable that transferred affective commitment to engagement and buying on livestream shopping platforms (Chen et al., 2020b).

A few researchers have also focused on the streamers’ (i.e., sellers) perspectives. Researchers have reported a range of selling strategies that streamers used for increasing sales revenue (Wongkitrungrueng et al., 2020; Chen et al., 2019b). But few of them unpacked how platform designs may mediate streamers’ behaviors and how these streamer practices may impact viewer’s buying experiences. This study aims to explore the questions. In the following, we review selling behaviors in the virtual context (e.g., TV shopping).

Although livestream shopping is emerging, it has similarities to TV shopping and social shopping given all the shopping modalities are happening virtually. To better understand seller behaviors and selling strategies on livestream shopping platforms, we review well-studied selling behaviors in TV shopping and social shopping.

TV shopping is a shopping method dating back to the 1980’s and is still popular in many countries. During a pre-recorded TV shopping program, a TV host presents products in detail, and viewers can order the products by calling the telephone number shown on the screen (Grant et al., 1991). Both livestream shopping and TV shopping show products in a lively way, however livestream shopping is more interactive than TV shopping (Liu, 2021). Many studies have investigated selling strategies in TV shopping. For example, Fritchie and colleagues found that social proof, scarcity, authority, commitment and consistency, liking, and reciprocation were used as selling strategies to simulate selling. These selling strategies were based on persuasion theory (Cialdini, 2009). TV hosts also built personal and emotional connections with viewers to fostered viewer trust  (Kline, 2005; Stephens et al., 1996). For example, hosts asked questions to establish similarities with viewers (Stephens et al., 1996).

Social shopping is another form of online shopping, by which product promotions are disseminated through personal social networks on online platforms (e.g., Facebook, Instagram) negating the need to switch to other e-commerce channels (Chen et al., 2020a). The salespeople, or “market intermediaries,” actively share product information with their social connections on social shopping platforms. Some research has studied the behaviors of market intermediaries in social shopping platforms and found that market intermediaries helped customers detect trending products, evaluated products, and provided a good price and discount to customers (Chen et al., 2020a, 2002; Kim and Park, 2013). In addition to providing product-related support, they also built trust with potential customers by generating and sharing content (e.g., short video) (Hajli, 2015) and providing emotional support (Shanmugam et al., 2016). Customers reported that, in addition to trust, culture and social presence were two other factors that impacted market intermediaries’ sales (Ng, 2013).

Among these selling strategies, researchers have also examined the malicious side of selling. The existing literature has extensively discussed the negative side of TV selling strategies. For example, consumers believed TV shopping was extremely risky (Burgess, 2003) as the product quality might be poor and the price was uncertain (Burgess and Drake, 1995). TV shopping hosts were also found to use their knowledge of human behaviors and viewers’ desires to implement deceptive strategies that incited purchasing. For example, TV shopping used compliance-gaining techniques to urge viewers. These techniques included using countdown timers and “sold out” signs (Auter and Moore, 1993). The host of TV shopping were also found to introduce a product with emphasis on financial prudence (e.g., bargain price and versatility) and upward mobility (e.g., items looking expensive) (Cook, 2000). Studies also revealed undisclosed advertising strategies in social shopping (Wu, 2016; Mathur et al., 2018). Undisclosed advertising is a form of advertising where influencers on social media platforms such as YouTube do not explicitly disclose their collaborative relationships with advertisers or products. Undisclosed advertising is prevalent on social media platforms, but it is hard for users to be aware of ambiguous disclosures (Mathur et al., 2018).

Livestream shopping combines salesmanship and parasocial interaction characteristics. During a livestream, a streamer serves as the salesman to display products and build social connections with viewers. For example, a streamer who sells books in her channel may also teach history courses relevant to the promoted books and share her personal stories during her livestream.

On livestream shopping platforms, streamers can use various strategies to attract attention, increase sales, and build customer engagement. For example, Wongkitrungrueng et al. (Wongkitrungrueng et al., 2020) found four sales approaches were used in livestream commerce, i.e., “transaction-based,” “persuasion-based,” “content-based,” and “relationship-based.” Through the transaction-based approach, streamers used simple selling, limit quantity/time/offer, and demonstration to facilitate their selling. Through the persuasion-based approach, streamers relied on game-prize, show, and personal characters to persuade viewers. Through the content-based approach, sellers provided product related or non-product related information to enrich their livestream’s content. Through the relationship-based approach, streamers shared their personal life, feelings or experiences, and organized community activities to strengthen their relationships with viewers.

These characteristics result in livestream shopping differing from TV shopping and social shopping. It signals a need to investigate whether malicious selling strategies are also existing in livestream shopping. To our best knowledge, the only work that studied malicious selling in livestream is (Xu et al., 2019). The study suggested that a viewer’s impulsive buying behavior might be related to “emotional energy,” which refers to customers’ affection with streamers. In this work, we aim to understand what streamers’ malicious selling strategies in livestream shopping are and how livestream platform designs mediate malicious selling strategies. To address the latter question, we review the scholarship on the negative effects of interface designs on users.

Livestream shopping is a new type of online shopping experience, thus streamers’ and viewers’ experience are empowered or bounded by a platform’s UI designs. In the HCI community, one avenue of research that focuses on the negative effects of interface designs on users is “dark patterns.” As a concept referring to marketing deceptive practices, nudging public policy and others, dark pattern does not yet have a consistent definition. Brignull et al. initially defined it as “a user interface that has been carefully crafted to trick users into doing things” (Brignull et al., 2015). Later, Brignull used a new term “deceptive design” to emphasize the deceptive attribute in the definition and make the term more inclusive (Brignull, 2018). We bring this change to the front and use the new term in the following⁴⁴4We follow the ACM’s inclusive terminology policy https://www.acm.org/diversity-inclusion/words-matter and transition to the new term “deceptive design” in this work regardless of what prior literature used. We also urge researchers to use the new term in future work.. Narayanan et al. then added the covert manipulation and coercion aspects and defined that deceptive designs “deceive users while others covertly manipulate or coerce them into choices that are not in their best interests” (Narayanan et al., 2020). These definitions emphasize that deceptive designs undermine individual decision-making and harm their interests (Mathur et al., 2021).

Researchers have developed taxonomies to classify deceptive designs. Brignull proposed a deceptive design taxonomy to categorize the various designs (Brignull et al., 2015; Brignull, 2018). The taxonomy consists of 12 categories, e.g., “Bait and Switch,” “Disguised Ad”  (Brignull, 2018). Gray and colleagues criticized that the taxonomy mixed up context, strategy, and outcome. Thus they simplified the taxonomy into five categories: “nagging,” “obstruction,” “sneaking,” “interface interference,” and “forced action” (Gray et al., 2018). Mathur and colleagues built upon this research and developed a deceptive design taxonomy specifically for online shopping websites (Mathur et al., 2019). The taxonomy included seven deceptive design categories. Compared to Bringnull’s and Gray’s taxonomies (Gray et al., 2018), Mathur’s version added “Urgency,” “Social Proof,” and “Scarcity” for online shopping context, and they combined multiple categories such as “Confirmshaming” and “Trick Questions” into a single “Misdirection” category. Further, Mathur and colleagues summarized five characteristics of deceptive designs, “Asymmetric,” “Covert,” “Deceptive,” “Hides Information,” and “Restrictive” (Mathur et al., 2019). Deceptive designs need to meet at least one of the five characteristics. Gray et al. stated that deceptive designs should be “obnoxious, coercive, or deceitful” (Gray et al., 2020). These properties serve to differentiate deceptive designs from non-deceptive designs. Deceptive designs have been reported in various scenarios including e-shopping websites (Mathur et al., 2019; Moser et al., 2019), mobile applications (Di Geronimo et al., 2020), gaming (Deterding et al., 2020; Zagal et al., 2013), robot (Lacey and Caudwell, 2019), etc. Studies that were not particularly looking at deceptive designs also reported “bad” design practices that could potentially harm users, such as malicious interface techniques (Conti and Sobiesk, 2010), asshole designers (Gray et al., 2020), privacy paradoxes (Waldman, 2020), and impulsive buying (Moser et al., 2019). Conti and Sobiesk defined malicious interface techniques as the intention to “deliberately sacrifice the user experience” to achieve designers’ goals (Conti and Sobiesk, 2010). Gray et al. found that even without the presence of deceptive designs, asshole designs could still deceive users, such as restricting users’ tasks, setting traps that were hard to avoid, and misrepresenting information, which highlighted designers’ unethical intentions and motivations (Gray et al., 2020). Moser and colleagues (Moser et al., 2019) studied impulse buying in e-shopping websites. Some impulsive buying features were different from deceptive designs. For example, an impulse buying feature “Offering Returns” is not a deceptive design because it does not directly benefit e-shopping websites (Brignull et al., 2015; Mathur et al., 2019).

Studies have started to understand users’ perceptions about deceptive designs, i.e., whether users can be aware of deceptive designs (Di Geronimo et al., 2020; Maier and Harr, 2020; Waldman, 2020) and how deceptive designs influence user behavior (Luguri and Strahilevitz, 2019). Studies indicated that users tended to be insensitive to deceptive designs even with hints (Di Geronimo et al., 2020; Maier and Harr, 2020). For example, Di Geronimo and colleagues explored user’s perception of UI deceptive designs in mobile applications. Their online user experiment illustrated that the presence of malicious designs was not easy for users to notice (Di Geronimo et al., 2020). If users were informed of the existence of deceptive designs, however, they were found to be better aware of them (Waldman, 2020). Maier and Harr (Maier and Harr, 2020) found that deceptive techniques were perceived as less malicious by users and some of them were regarded as just sneaky and dishonest. Luguri and Strahilevitz found that well-educated users were less susceptible to deceptive designs than a less-educated group. The deceptive designs “hidden information,” “trick questions,” and “obstruction” were more likely to be manipulative than others such as “nagging,” “sneaking,” “forced action” (Luguri and Strahilevitz, 2019). Mathur et al. discussed the origins of deceptive designs from the disciplines of psychology and behavioral economics and proposed possibilities to evaluate the impacts of deceptive designs on individuals such as financial loss, violation of privacy and cognitive burden (Mathur et al., 2021). These findings indicate the need to better understand users’ perceptions of, and reactions to, deceptive designs with the purpose of mitigating harm.

In this present research, with the reported malicious selling in livestream shopping, there may be some designs that impair viewers’ decision-making and purchase behaviors due to the ways that streamers use these designs for malicious selling purposes. Thus, we investigated what malicious selling strategies were used by streamers on livestream shopping platforms and how these designs might fluidly turn into deceptive designs during livestream shopping scenarios.

To investigate streamer selling strategies, a group of 40 publicly available livestream sessions were sampled and archived as video recordings. These 40 sessions were taken from two predominant livestream shopping platforms in China, Alibaba’s Taobao and ByteDance’s TikTok. To identify malicious selling strategies, inductive coding was performed with the recorded videos.

To understand viewer perceptions of these malicious selling strategies, we designed and conducted a semi-structured interview study to gather viewers’ awareness, attitudes, and reactions to these selling strategies. This study was approved by the IRB of the fourth author’s university.

First, as the purpose of this research was not to identify individual livestream sellers who were applying malicious selling strategies, we de-identified the streamers’ channel names. Note that most of the studied streamers’ selling strategies and tactics were not malicious. Readers should not be misled or biased into thinking negatively about all platforms or steamers. To avoid general marketing tactics that mistakenly were identified as malicious selling strategies, the researcher team conducted several rounds of discussion to refine the definitions of malicious selling strategies in the taxonomy.

Moreover, a critical perspective was used to analyze livestream selling behaviors. We think it is critical to pay attention to any selling strategies or livestream design that undermine viewers’ decision-making. Thus, we reported all livestream selling strategies either identified by researchers or participants, no matter whether these strategies are legal or prevalent.

The content analysis of the livestream shopping video recordings revealed 15 malicious selling strategies. Participants were aware of most malicious strategies except for the Pressured Selling strategy. Moreover, three participants mentioned a new malicious selling strategy, Forced Wholesale. Notably, if a participant mentioned a livestream shopping experience related to a malicious selling strategy but did not perceive it as malicious, we did not count the participant for the malicious strategy. We categorized the 16 malicious selling strategies into four types—Restrictive, Deceptive, Covert, and Asymmetric—which were developed by (Mathur et al., 2019).

Table 3 lists the four categories with 16 malicious selling strategies, the explanations, and the participants who were aware of the malicious strategies.

Figure 1 demonstrates the interfaces of a TikTok livestream channel. Table 4 shows 15 platform designs that could mediate the malicious selling strategies. The 15 designs were presented in both TikTok and Taobao platforms. The video recordings demonstrated that Restrictive selling strategies (e.g., Forced Endorsement, Forced Subscription) and Deceptive selling strategies (e.g., Fake Scarcity, Fake Social Proof, Fake Exclusive Pricing, and Fake Urgency) were more likely to be facilitated by platform designs than Covert and Asymmetric strategies. Covert and Asymmetric strategies relied more on sellers’ sale skills or styles so they were less influenced by platform designs. Nine specific strategies were supported by design features or functions within the two livestream platforms. Below, we reflect on how livestreaming platform designs mediated the four categories of malicious selling strategies.

Designs using subscription, reminder tags, sharing, liking, gifting or chat channels enabled or supported Restrictive malicious selling strategies such as Forced Subscription and Forced Endorsement. It is common for e-shopping websites and livestream channels to apply these designs to require customers’ or viewers’ subscription, sharing, or gifting, however, we found that the backend filtering tool, which was orally claimed by streamers, was the key function to enable Restrictive malicious selling strategies. For example, before starting a lucky draw, the VT6 streamer told viewers that his assistant would check the viewers’ subscription. If viewers did not subscribe to his channel, the streamer would cancel the viewers’ prize. The VT7 streamer required viewers to click the “like” button and send a required message “888” (stands for fortune in Chinese pronunciation) to the public chat channel. She said: “Everyone who wants a coupon sends 888.” However, it is not clear whether and how the filtering tool ran. Viewers could only follow the streamer’s requirement to obtain the coupon. Since Restrictive strategies were enabled or supported by platform designs, this suggests the need to modify or remove such design features to constrain malicious selling strategies.

The designs that enabled or supported Deceptive strategies presented misleading information (i.e., notification messages, product pages, countdown timers, back-ordered warnings, and chat channels) and inductive discounts (i.e., red pockets, coupons, or flash sales). These designs helped streamers create fake scarcity, social proof, exclusive pricing and urgency to induce viewers to buy products. For example, both Taobao and TikTok have a notification message feature. Once a viewer adds an item to his/her shopping cart, there will be a public notification message on the screen saying “User X is ordering the item”. TikTok also shows the total number of buyers on the screen. The public notification messages can be used as evidence of high demand. Using this evidence, streamers can urge viewers to place an order. In one case, the streamer claimed, “This is really the last one”, however, the streamer asked her assistant to restock the product later. Although Deceptive strategies were enabled or supported by specific designs, it took time for viewers to detect or verify fake information compared to the Restrictive strategies. It is thus necessary to provide tools to help viewers or customers to identify deceptive malicious selling.

Designs supporting beautify camera filters enabled and supported Covert strategies. For example, the visual misrepresentation enabled by the beautify camera filter facilitated malicious selling. The streamer of VD11 used the beautify camera filter provided by TikTok to hide the flaws of her apparel such as the bad texture and workmanship.

The results also revealed that the chat channel supported three types of malicious selling strategies. It was also the only design used by viewers to counter malicious selling strategies. This was not surprising because the chat channel was the only means for viewers to provide negative feedback. For example, some streamers enforced Egoistic Norms by asking viewers to send certain positive terms in the chat channel. Viewers sent comments such as “poor quality products” to fight against malicious selling via the chat channel.

Participants reported multiple ways to cope with malicious selling strategies. The most frequent mechanism was to select professional streamers to watch due to their high reputation. A professional streamer is a salesperson or agent that acts as a liaison between product manufacturers and customers and takes a commission on the sale (Financing China, 2019). Participants noted that choosing brands and stores with a good reputation was a useful means to circumvent malicious selling strategies. In addition, participants suggested “shopping around” or comparing products in different livestream channels before buying. Participants also leveraged the chat channel to push back against malicious selling strategies. For example, they reported low-quality products, poor return service, or astroturfers in the chat to alert other viewers.

Participants reported four ways to deal with malicious selling strategies, however, not all participants considered these effective for some malicious strategies. For example, P5 planned to buy a pearl necklace for her mom. She watched different pearl livestream shopping channels and acquired the pearl knowledge introduced by the streamers. After about a week, she gained a sense of which necklaces were high-quality and found that some streamers had fake testimonials and did not tell the truth. It was also very time-consuming for her to acquire her knowledge. This suggests a need to understand the challenges inherent in defending against malicious selling strategies for viewers.

The interview-based data provided insights into the challenges of coping with malicious selling. These challenges were due to highly personalized stream procedures, malicious designs that impaired viewers, and the visual limitations of livestream for special products.

The recording coding process referred to existing typologies of deceptive designs. Although malicious selling strategies and deceptive designs are two different concepts, bad intentions or strategies behind selling strategies and deceptive designs shared some similarities. We borrowed codes such as “urgency,” “scarcity,” “social proof,” “forced subscription,” and “nagging” from the existing typologies (Mathur et al., 2019; Gray et al., 2018; Crabtree and Miller, 1992). More importantly, our study contributes new findings that are different from existing studies in two regards.

First, we found that participants were more aware of Restrictive and Deceptive selling strategies than Covert and Asymmetric selling strategies. One reason could be that the frequency of these strategies were different. On a deeper level, the implementation of Covert and Asymmetric selling strategies relied more on sellers than platform design features or functions. Here, it seems that platform designs serve as checkpoints to verify if selling strategies were malicious or not. Without these checkpoints, malicious selling strategies might be more concealed. Thus, it is worth investigating if the absence of design features and functions will make the awareness of malicious selling strategies more difficult.

We also found that more than half of the malicious selling strategies were enhanced by platform designs. The chat channel supported more malicious selling strategies than other designs however, it was the only design used by viewers to fight against malicious selling. Restrictive selling strategies (e.g., Forced Endorsement, Forced Subscription) and Deceptive selling strategies (e.g., Fake Scarcity, Fake Social Proof, Fake Exclusive Pricing, and Fake Urgency) were more likely to be amplified by designs than others. Moreover, the Forced Endorsement and Forced Subscription strategies were amplified by multiple engagement designs such as subscriptions, reminder tags, sharing, liking, and gifting. These features were initially designed to engage viewers, however, they were maliciously used by streamers.

Second, compared to prior studies about malicious selling strategies during TV shopping and social shopping, our study identified new malicious strategies: Forced Subscription, Forced Endorsement, Playacting, Fuzzy Targeting, Sophistry, Retaining Customers, Egoistic Norms, and Disgracing Others. These strategies were enabled by the frequent and various interactions between a streamer and their viewers, such as the streamer’s verbal communication and viewers’ comments in the chat channel, sending gifts, and likes. This demonstrates how more attention needs to be paid to malicious selling strategies when the interaction between the seller and customers is more frequent and diverse.

One might argue that some malicious selling strategies identified in this work seem like traditional marketing strategies. In this work, strategies such as Retaining Customers and Playacting were similar to the content-based approach in Wongkitrungrueng’s work (Wongkitrungrueng et al., 2020) and were not malicious. However, this work emphasizes that these livestream selling tactics were supported by platform functions or features and thus could manipulate viewers’ decision-making more easily than offline marketing tactics. As Narayanan et al. argued, deceptive designs could be more aggressive than offline marketing strategies (Narayanan et al., 2020). Because designers can apply A/B tests to pick out the most manipulative design and make great profits for companies. As a result, “even seemingly trivial changes to design elements can result in substantial differences in behavior” (Narayanan et al., 2020). It is thus important to evaluate the malicious side of selling strategies happening on livestream platforms.

We studied two popular livestream shopping platforms in China, where the regulation and cultural context of China might impact malicious selling and platform designs. To provide a better context for readers, we discuss how regulation and cultural context in China impacts livestream commerce, especially malicious selling strategies and deceptive designs.

First, from the regulation perspective, livestream e-commerce is regulated by state government departments in China. As previously mentioned, with the ambition of developing “digital China,” the state government has provided support to the internet economy, including livestream e-commerce. The government also encourages small businesses to leverage livestream to expand their business. Thus, many small business owners become streamers and sell their products via livestream. This leads to a variety of streamers on livestream platforms, ranging from professional streamers to amateur streamers from different industries (e.g., students, farmers). Thus, there are versatile and various selling strategies, including malicious selling strategies.

The state government tries to seek a balance between economic development and regulation. Currently, the regulations for livestream e-commerce in China are government policies rather than laws (Ning and Yang, 2021). In 2020, the Cyberspace Administration of China announced that livestream platforms have to build a credit assessment list for livestreamers, where any livestreamer on the “blacklist” will be prohibited from using all livestreaming platforms (Yu and Goh, 2020). In 2021, the State Administration of Market Regulation said in a statement that all livestream platforms should “quickly conduct self-control and comprehensive inspections” on product quality, and should punish streamers who sell poor quality products or misleading advertising. These regulations demonstrate that livestream platforms take the main responsibility to monitor and regulate streamers’ illegal or inappropriate behaviors. But there are no specific policies or laws to govern deceptive designs on livestream platforms. In the U.S., the DETOUR Act states that it is unlawful “to design, modify, or manipulate a user interface with the purpose or substantial effect of obscuring, subverting, or impairing user autonomy, decision-making, or choice to obtain consent or user data” (Warner and Fischer, 2019). Future research could further study the legality of deceptive designs in different online scenarios.

Second, from the cultural perspective, users in China prefer to use one app for many different functions (Liu, 2021). For example, WeChat has combined social media, e-commerce, payments, and many other functions into a self-contained ecosystem (Davis and Xiao, 2021). Such integrated platforms are rarely seen in the U.S. (e.g., Facebook or Apple). Similarly, with livestream shopping, platforms try to include many features and functions in their interfaces. However, in the U.S. and other countries, the interface design for livestream shopping is much simpler. For example, Amazon has only four main features in its livestream interface: the streaming video, a list of products being sold, a chat area, and a reaction button where viewers can send “likes” to the streamer (Amazon Live, 2021). Platforms and streamers benefit from a greater degree of stability and opportunity in China and also experience a greater precarity in policy and regulation (Cunningham et al., 2019).

In China, the ecosystem of livestream commerce is becoming mature. It mainly consists of three roles: product and service providers (e.g., manufacturers, supply chains), marketers and retailers (e.g., livestream commerce platforms, streamers, online stores), and consumers (KPMG and Ali Research Institute, 2020). It might be fruitful in the future to investigate the roles of different stakeholders in this ecosystem, and to understand how they influence platform designs and regulation and how they perceive and cope with streamers’ malicious selling strategies.

Our research identified many negative behaviors from the streamer perspective. Although some of the streamers’ malicious selling strategies were highly enhanced by system designs, we also identified seven malicious strategies that were purely implemented by streamers. These problems revealed standardized product stream procedures have not yet been established in livestream e-commerce. On the surface level, participants expressed that streamers were skilled in introducing products and persuading customers to buy. When we asked participants if they thought the ways or the strategies streamers used were malicious, they were not sure. Participants expressed that it was hard to differentiate exaggeration and deception. They reported cases where customers were controlled by streamers or designs. We found that most malicious strategies were directly or indirectly related to this. Compared to other e-shopping channels, customers were in a much more inferior position to streamers while livestream shopping. Streamers had a high level of freedom to decide how they sold products in livestream channels.

Prior work has analyzed viewers’ negative behaviors during livestreams and how livestreaming platforms allow viewers to report inappropriate content (Wohn, 2019). For example, viewers reported trolling and being talkative in livestream channels. Given the unbalanced power between viewers and streamers, the platforms need to provide more ways for customers to combat malicious selling, such as collective reporting.

In the interview study, the participants mentioned several types of professional streamers, i.e., professional streamers, celebrity streamers, and influencer streamers. Some participants watched livestream e-commerce hosted by celebrities. Participants also watched influencers’ livestreams and shopped in their channels. Before starting livestream selling, these influencers shared self-generated content (short videos) on TikTok and drew a large base of fans. Participants’ trust and preference towards these professional streamers was due to their professionalism in choosing high-quality products and introducing products, and their authenticity within or outside of platforms. To fight against malicious selling, more professional streamers are needed.

These findings suggest that viewers’ behaviors are constrained by the current designs of livestream platforms. In the livestream shopping setting, nine malicious selling strategies were enhanced by design features in livestream shopping channels. For instance, there were multiple notification features (e.g., purchase notification, viewing notification, sold-out warning notification) that allowed streamers to create fake scarcity or urgency. The feeling of scarcity led to many impulse buying behaviors. Thus, streamers used such features and capitalized on the behaviors they resulted in. The chat channel, which was the main way for viewers to interact with the streamer and other viewers, was also manipulated by streamers to create fake scarcity impressions. Previous research has showed how synchronous chat facilitated knowledge sharing among viewers (Wu et al., 2018). However, as the streamer asked their viewers to post required content, such as their desired products in the chat channel, a viewer might only see thousands of “wanting” messages in the channel instead of product related knowledge shared by other viewers. Additionally, the chat channel did not support viewer-centric functions such as search, filter, and reply. The missing functions further constrained viewers’ power to defend malicious selling behaviors. The designs of livestream platforms had a negative side when they were maliciously used by streamers. Gray and colleagues suggested that the intentions and motivations behind designs should be examined even if the designs were not deceptive designs (Gray et al., 2020). As our research suggests, engagement and communication designs should be studied with human behaviors (both streamers and viewers). An isolated investigation might overlook the potential negative side of the designs. Our study only investigated the features and functions level, however, in the future, an interesting direction could be to explore what properties of these features and functions can amplify malicious selling strategies.

Based on these findings, various design implications can be proposed to protect users from these malicious selling strategies and designs. First, it might be helpful to increase the transparency and viewers’ autonomy in designs. For instance, the backend filtering tool was the key function for Restrictive selling strategies. Viewers need to know what filter mechanisms are available and used if they comply with such forced actions. The chat channel also needs to be improved to enhance viewers’ autonomy. Additional functions can be added to the chat channel, such as filter, search, or reply. Viewers should also be able to switch off message features (e.g., purchase message, viewing message, sold out message) on the screen so that they can focus on the product presentation and avoid being affected by others decisions. Second, it might be helpful to implement crowdsourced features to validate the information provided in livestream channels. For example, to counter the Fake Exclusive Pricing selling strategy, viewers can check prices across different e-commerce platforms or check other viewers’ ratings or reviews to verify whether it was true or a false claimed exclusive pricing. We acknowledge such designs are unlikely to be adopted by the platform itself as it may be against its own business interests, however, third party developers may implement such designs as system add-ons to support viewers.

One general finding is that streamer’s selling behaviors were loosely regulated. Even if they had bragged or promised something during the real-time streaming session, it was very hard for buyers to double-check whether or not it was true in the moment, and it was even more difficult to go back to what the streamer said after the session ended. For example, a streamer may promise an unconditional return policy during the livestream session, but the actual return policy written on the product information page was conditional. Since malicious selling strategies relied more on streamers’ sale skills and styles rather than platform features and functions, the platform and regulators should make and enforce policies to regulate streamers’ selling practices similar to how they would regulate an offline salesperson. For example, it is important to determine if forced actions are legal or necessary for customers to purchase products during livestream shopping. Platforms should also implement machine learning-based solutions to automatically transcribe and record what the streamer is saying during the livestream session to extract keywords from the transcripts and cross-reference the product information page (e.g., return policy, brand, and manufacture info, etc.). If there is a mismatch, the system may pop up a real-time alert to the viewers. Viewers can also use recorded transcripts as evidence in post-sale customer service.

Our study only examined two livestream shopping platforms—Taobao and TikTok and the interview participants were mainly users of these two platforms. The findings of malicious selling strategies may thus not be generalizable to other platforms given their different platform policies and designs.

The current work is also descriptive. The qualitative analysis and the interview study only identified the presence of malicious selling strategies. Future work could use the taxonomy developed in this study to investigate the prevalence of various malicious selling strategies or run experiments to evaluate if malicious selling strategies could lead to financial loss of customers.

Additionally, we only recorded 40 livestream sessions over a short period. In the future, we plan to conduct a longitudinal analysis, that is, observing several livestream shopping channels as a real viewer over a long term. We will also conduct an interview study for streamers to understand their motivations and attitudes.