Augmented reality-delivered product information at the point of sale: when information controllability backfires

AR systems provide an indirect view of the physical environment via a digital screen. Different devices—including fixed (interactive screens), mobile (smartphones, tablet computers), and mobile systems (head-mounted displays, AR glasses)—enable AR systems. AR overlays and supplements the real world with computer-generated objects, which appear in real time in the same visual space as the physical world elements, such that they appear to coexist (Azuma et al., 2001; Flavián et al., 2019). Users can even interact with the virtual objects (Zhou et al., 2008). Interactivity and augmentation are, therefore, key elements of AR applications (Javornik, 2016a) that we address with this research. In an information context, we investigate interactivity in terms of controllability and augmentation in terms of information detailedness.

Several characteristic AR features (Javornik, 2016a) provide value to consumers. First, the AR system can augment the self via virtual try-ons (virtual mirrors) for apparel, cosmetics, glasses, etc. (Hilken et al., 2017; Yim et al., 2017). Second, AR can augment the actual environment. Furniture planners, for example, aid consumers in imagining how furniture would look in their rooms (Holdack et al., 2020; Javornik, 2016a). Third, AR can augment the product or shelf at the point of sale (Joerß et al., 2021; van Esch et al., 2019). For example, consumers scan the product to receive further details via a digital information layer. We focus on this information layer because superimposing information onto shopping realities is a promising AR application for brick-and-mortar retailing.

While AR improves consumer experiences in online and offline retailing, it is necessary to augment and enrich fundamentally different aspects in both settings. E-commerce already enables unlimited space for product information, but consumers can struggle to imagine the product or product fit. There is no possibility to touch or try the product. Thus, AR in e-commerce characteristically offers three-dimensional visualizations and virtual try-ons (see the overview in Table 7 in Appendix 1 for AR literature in e-commerce). In offline contexts, the AR’s advantage is vice versa. While consumers can easily touch, taste, or try the actual three-dimensional product in the store, the space for additional product information is often restricted. ARPI overcomes this limitation with additional information layers.

As shown in Table 7 in Appendix 1, a growing body of AR literature focuses on e-commerce, including functions such as virtual try-ons (Kim & Forsythe, 2008; Yim et al., 2017) or furniture planners (Rauschnabel et al., 2019; Rese et al., 2014). Only a few studies consider traditional retail settings (Table 1). These studies have specific foci on anthropomorphism (van Esch et al., 2019), usage intention toward AR glasses (Holdack et al., 2020), or information evaluation (Spreer & Kallweit, 2014). While these studies provide valuable input, they neither manipulated the AR design nor considered purchase-related variables. Only Joerß et al. (2021) manipulated product ratings in the AR to show that the technology can guide consumer decisions. The lab study, however, did not provide field evidence on purchase behavior, nor did it manipulate the AR controllability or other variables relevant for the present research. Given the lack of insight into offline contexts, we also inform our model by AR research in online contexts, which we adapt due to the different AR functions. Further knowledge comes from research streams on related technologies, such as self-services technologies and personal shopping assistants (e.g., Evanschitzky et al., 2015; Giebelhausen et al., 2014; Mende et al., 2019; Meuter et al., 2000).

Studies in e-commerce (Rese et al., 2014) and offline settings (Spreer & Kallweit, 2014) have identified drivers of the intentions to use or reuse AR applications. These studies, among others, build on the technology acceptance model and its extensions (Venkatesh et al., 2003) to explain the adoption of AR technology (Huang & Liao, 2015; Rese et al., 2014, 2017). Other research refers to flow theory (Novak et al., 2003) to explain the motivational factors (Javornik, 2016b). Many studies consider AR’s utilitarian and hedonic benefits (e.g., Rauschnabel et al., 2019), building on insights into consumers’ adoption and digital technologies’ effectiveness to support shopping (Childers et al., 2001). Scholars usually operationalize hedonic benefit as perceived enjoyment, but consider perceived informativeness as a utilitarian benefit (Dacko, 2017; Holdack et al., 2020). Both aspects improve attitudes toward AR technology in e-commerce (Rese et al., 2014; Yim et al., 2017). Holdack et al. (2020) recently provided evidence that this is also true for brick-and-mortar retailing. Synthesizing the literature shows that hedonic benefits are central to the more playful applications in e-commerce (e.g., virtual try-ons). In brick-and-mortar-retailing, users of AR shopping apps may consider the prospect of receiving additional product information (e.g., about product sustainability) as an efficient means to reduce purchase uncertainty, which they deem a unique utilitarian benefit of AR over traditional shopping experiences (Dacko, 2017; Spreer & Kallweit, 2014). In two preliminary studies motivating the present research, we substantiated that AR product overlays predominantly provide utilitarian benefit to consumers in physical stores (see Web-Appendix A). This paper, therefore, shifts the focus to AR-enabled product information at the point of sale.

In retail settings, the AR-enabled controllability of information delivery may have a disadvantage, ignored thus far, that reduces or even cancels out the mentioned benefits. Given that ARPI is primarily helpful for utilitarian consumption decisions (see pre-studies 1 and 2 in Web-Appendix A), its effect on more downstream variables (brand image, purchase intention) will not only be driven by interaction quality. The question of whether consumers feel comfortable with the information collected will be even more important. We suggest that the controllability of information can provide a shortcut to the most relevant pieces of information at the expense of richer, fuller, and broader information.

While ARPI controllability enables consumers to select the information they value, this also implies filtering out certain pieces of information. Building on the cues-filtered-out perspective (Yadav & Varadarajan, 2005), a high degree of controllability implies that users who selectively choose content miss out on the non-chosen content (i.e., cues are filtered out). If consumers can control the AR-delivered information content, they may consider the seen information as less rich and may even fear that cues have been lost. While this unconsidered information might not be of focal interest, it may still serve as a signal to the amount and type of information that is principally available, shaping the user’s impression about the comprehensiveness of information. In other areas, such as social media, it has been shown that the fear of missing out on information creates stress and affects subsequent behaviors negatively (Tandon et al., 2021, 2022). We assume that these negative aspects also apply to ARPI communication.

Information controllability can create the impression of not being fully aware of all available information. By contrast, if users cannot control the AR-delivered content, they have to browse for the information they deem most relevant. During this search, they notice—consciously or unconsciously—the extent of other product information that they may not deliberately process but which subtly sends signals of being well-informed to them. This peripheral perception with regard to the comprehensiveness of information is lost when the AR device allows users to control the displayed information. Since consumers may not be exposed to all potentially available information chunks, they are less likely to develop a vivid imagination of the extent of information at hand. As a result of this, we propose that the controllability of ARPI decreases consumers’ perceptions of information comprehensiveness.

Since this is relevant to retailers, we discuss next when controllability’s detrimental effects on perceived comprehensiveness spill over to downstream marketing-relevant variables. Rese et al. (2014, 2017) show that in e-commerce, the perceived informativeness of an AR guides the perceived usefulness and, in turn, consumers’ usage intentions. We transfer this to the brick-and-mortar context. Plausibly, the magnitude of the perceived information comprehensiveness is contingent on the amount of information that is readily available and disclosed by the brand.

There is an interaction effect between ARPI controllability and information detailedness on brand image. That is, ARPI controllability weakens consumers’ brand image perception when information detailedness is high; this effect is attenuated when information detailedness is low.

Furthermore, we consider the transactional outcome variable, i.e., purchase intention, which is typically more strongly guided by competence than by the warmth component of the provider (Güntürkün et al., 2020). Useful product information increases clarity and reduces doubts and risks. This enables consumers to make product choices with greater decision comfort, ultimately increasing purchase likelihood (Heller et al., 2019; Hilken et al., 2017).

There is an interaction effect between ARPI controllability and information detailedness on purchase intention. That is, ARPI controllability weakens consumers’ purchase intention when information detailedness is high; this effect is attenuated when information detailedness is low.

Our theory suggests that the backfire effect is due to a loss in perceived comprehensiveness of the information that is accessible through the AR device. We define perceived comprehensiveness as consumers’ sense of being aware of all the accessible details and of being fully informed. Following our reasoning about how information controllability affects the perception of being aware of all the available information, we suggest that reduced perceptions of such information comprehensiveness underlie the patterns proposed with H1. We, therefore, predict the following mediation: Information controllability affects perceived comprehensiveness, which, in turn, affects brand image. Notably, the mediation of controllability via perceived comprehensiveness is conditional on the degree of information detailedness. When information detailedness is high, the information controllability will lead to lower levels of perceived comprehensiveness; as a result, consumers’ perception of not being fully informed will negatively spill over to the brand image.

The effect of ARPI controllability on brand image is mediated by information comprehensiveness when information detailedness is high. This mediation effect is attenuated when information detailedness is low.

Besides the imprint on the rather relational outcome variable brand image, we also expect an effect on the transactional outcome variable purchase intention. The more the consumers have a perception of being fully informed, the better they are able to justify purchasing the product.

The effect of ARPI controllability on purchase intention is mediated by information comprehensiveness when information detailedness is high. This mediation effect is attenuated when information detailedness is low.

While controllability may hamper the ARPI’s effect under the condition of high information detailedness, there are also several alternative mediators of AR controllability that imply positive effects. We need to explore these positive effects to fully understand the backfire effect. The variables tested in AR research are prime candidates for alternative mediators (Table 7 in Appendix 1). These variables include perceived credibility, perceived complexity, user friendliness, presence, novelty, utilitarian benefit, and hedonic benefit. However, past research has focused on e-commerce and often used the adoption intention as dependent variables. These mediators, therefore, have to be tested in the context of ARPI in a brick-and-mortar setting and for more downstream marketing outcomes. The rationale for the potential mediating effects is given in Table 8 in Appendix 2. Still, all these positive effects compete with (and are potentially dominated by) the inhibiting effect of reduced perceived information comprehensiveness.

The effects of ARPI controllability on (a) brand image and (b) purchase intention are mediated by alternative mediators when information detailedness is high. The mediation effect is attenuated when information detailedness is low.

Besides the moderating role of the AR detailedness, the proposed backfire effect of ARPI controllability may also depend on contextual factors involving strong practical relevance, such as rush hours (Irmak et al., 2020). In contrast to the e-commerce settings (e.g., Rese et al., 2014; Yim et al., 2017; Tan et al., 2021; Appendix A), the presence of other people in physical stores strongly influences the shopping experience and particularly consumers’ perceived stress level (Lucia-Palacios et al., 2018; Santini et al., 2020). Building on Lewin’s (1939) field theory, research has shown that particularly task-oriented shoppers feel more stress and display less patronage behavior if retail crowding is high (Baker & Wakefield, 2012). Against this background, we assume that the ARPI backfire effect may be more prominent during busy shopping times. Market reports, such as Bring! Shopper Report Germany (Bring!, 2020, p. 5), imply that retail crowding varies systematically across weekdays and day times. In busy shopping times (which we label “rush hours”), many consumers crowd around the same shelves. This detracts from a calm and relaxed atmosphere that may be required to interact intensively with a digital device for accessing additional product information. Shoppers are often urged to decide quickly and move on to let other consumers reach the shelf. This will determine whether shoppers construe the shopping situation as work or as a pleasant, explorative experience (Babin et al., 1994).

We, therefore, expect that retail crowding and time pressure govern whether customers value having control over the AR-enabled information presentation. Although controlling the information may sometimes help save time, it may increase the probability that consumers miss out on relevant information in stressful situations. Therefore, we investigate whether consumers respond less favorably to detailed information presented in a controllable manner during rush hours. Building on the rationale that retail crowding evokes stress that impairs attention and working memory, Gelbrich and Sattler (2014) demonstrate that technology anxiety hampers the usage of self-service technologies more strongly when retail crowding is high. Transferred to our context, we assume that controllable and detailed AR-delivered product information may unfold its unfavorable effect particularly when many consumers shop during rush hours (e.g., after work).

The backfire effect of ARPI more likely occurs in busy rush hours but is less likely in more relaxed shopping times.

The task-media fit hypothesis (Mennecke et al., 2000), an extension of the media richness theory, argues that the medium needs to be adjusted to the task for better interaction effectiveness. The controllability effect and its interplay with other aspects (detailedness) should also depend on the medium through which the information is transported. As such, the proposed backfire effect should primarily be an issue for digital devices. In principle, ARPI increases media richness when combining the product’s image with virtual elements, thereby providing more cues (Dennis & Kinney, 1998). Yet, the controllability filters out cues, which reduces media richness (Yadav & Varadarajan, 2005). Controllability enables consumers to jump directly to the exact piece of information they are interested in while omitting supposedly irrelevant information. Since this happens digitally, users do not gain experience with the selected route that led to this detail, nor are they exposed to the structure by which the information is organized or the extent of other potentially available details. The backfire effect of controllability should be particularly prominent here. In nondigital environments, such as print media (e.g., books, booklets, newspapers, flyers), tools like content overviews, page numbers, tabs, and registers enable a certain sort of adaptability too. These tools allow easier retrieval of the piece of information the user is interested in. Although booklet users do not read all other information when skipping to the relevant detail, they still gain an impression of the wealth of available information. For example, when selecting a tab or opening the target page, users are still exposed to the other pages and can feel the booklet’s thickness, giving them a sense of the available information. Consequently, we assume no or a much weaker backfire effect in nondigital media.

The backfire effect of ARPI more likely occurs when information is delivered in augmented realities but is less likely when information is delivered in physical realities.

The ARPI and its content were specifically developed for this study. This allowed us to standardize the information delivered by the AR device. We chose the cereal domain because the relevant information can be systematically varied for this product type. We focus on information about sustainability, as this is a product attribute that appears to be highly relevant, marked by a high level of uncertainty and ambiguity,² and which has been tested in a retailing context with AR (Joerß et al., 2021; van Esch et al., 2019). According to media richness theory, richer media can effectively help overcome the uncertainty in such contexts (Daft & Lengel, 1986). This main study included a specific brand to rule out distortions from brand preferences and reduce nuisance variance.

We developed the application with the Vuforia Framework in the unity engine. The participants received the app on a tablet computer (Android, 10.1 in., 1920 × 1200 pixels, camera with 13 megapixels) at the point of sale in a hypermarket. When pointing the tablet camera at the shelf, the real world (e.g., the product and shelf) was displayed, giving the impression of looking through a glass pane. If participants focused on a specific product, an overlay with product information appeared, hovering half a centimeter in front of the product package. We pretested and optimized the app’s usability with consumers, company managers, and business researchers. Users could interact with the overlay, browse through the product information, and open further information. Figures 6, 7, Table 10 in Appendix 4 shows the ARPI in use.

The target product was a major cereal brand, which we selected because cereal products vary substantially in sustainability depending on the composition and production methods. Three flavors (chocolate, fruit, and honey nut) account for individual taste preferences. We developed the sustainability information’s content and structure from several pre-tests and discussions with the company’s product and marketing managers. For each product, the managers developed three statements about ecological, social, and regional aspects and provided three arguments for each statement. We reworked the statements and details in intensive discussions with the company to ensure that they were comparable in text length, comprehensibility, and tonality. This process resulted in systematic and standardized sustainability content for each of the three flavors (Figs. 6, 7, Table 10 in Appendix 4). In a two-factorial design, we independently manipulated the controllability and detailedness of the AR-enabled information.

In the controllable condition, the users received the initial information regarding whether the products are ecologically, socially, and regionally produced. When interested in obtaining further information, they could click on a particular category to see three corresponding statements and, eventually, further details. Participants in the uncontrollable condition received the full information and did not have the opportunity to tailor the presentation (extend or hide details); they had to scroll to find and access the information they were interested in.

We split the delivered information into general statements and supporting details (Figs. 6, 7, Table 10 in Appendix 4). The complete information in the app consisted of 1337 words (10,517 characters). The statements made up about one-fourth of the text (in sum 378 words, 2858 characters). Participants in the nondetailed condition saw these statements, while the detailed condition also included three details per statement (in sum 959 words, 7659 characters). The statements and details, on average, consisted of 12.4 words (SD = 2.1) and 97.4 characters (SD = 10.8). Participants in the detailed conditions had access to all statements and details, whereas those in nondetailed cells had access to statements, but not to the arguments at the lower level. In the uncontrollable-detailed cell, they saw a long text and needed to scroll to search the information. If they were in the controllable-detailed cell, they could click on the statements they were interested in to see the three corresponding details.

To differentiate ARPI from the more traditional means of informing consumers at the point of sale, we designed four versions of a paper booklet, mirroring the four app manipulations. In the uncontrollable-detailed version, participants received a booklet with statements and the corresponding details. In the uncontrollable-nondetailed version, we provided only the statements, not the details. In the controllable versions, the same content was provided, but we added a content overview and several tabs on the right-hand side of the booklet to provide quick access to the relevant information.

We ran the experiment in a hypermarket offering more than 80,000 articles, and located in a shopping mall of a medium-sized city. The experiment took place during four weeks. A planned schedule helped systematically vary the data collection across different times (9:00 a.m. to 12:30 p.m., 12:30 p.m. to 4:00 p.m., and 4:30 p.m. to 7:30 p.m.) and week days (Monday to Saturday). The hypermarket was closed on Sundays. Each slot used all four treatments (controllability × detailedness) and participants were randomly assigned.

Trained interviewers approached shoppers entering the hypermarket. Every time the interviewers were not busy explaining the app function to one of the participants, they asked the next shopper entering the hypermarket whether he or she was willing to participate in the study. There was no exclusion criterion (e.g., in terms of age, gender, etc.). Approximately 10 % of the approached shoppers agreed to participate. The interviewers briefly introduced the participants to the app’s basic functionality for approximately 30 s. The interviewers asked the participants to use the app for as long as they wanted to learn more about the cereals. They did not urge the participants to buy the cereals. The participants were instructed to walk unaccompanied to the cereals shelf where we provided a tablet with the ARPI (about five meters from the entrance on the main floor, see the map of the hypermarket in Web-Appendix B). The ARPI’s instructions and usage were, therefore, separated in time and space. The shoppers registered in the AR app with an anonymous code, ensuring that participants used the app only once. They logged in with their code and then used the ARPI to inform themselves about the target products in the absence of an interviewer. This ensured that they could do their shopping and use the ARPI at their own pace. After completing their shopping, the interviewer asked them at the hypermarket’s exit to complete a questionnaire that was matched with the ARPI app data (e.g., time taken to consider the information) and the interviewers’ observations (e.g., how full their shopping cart was). The shoppers received a cereal bar for participating in the study.

We applied random assignment to the experimental condition, as well as multiple data sources to reduce common method variance (Table 2). We included several measures to reduce or control for the noise that comes with experiments in field settings.

We designed the app to track the usage time. On average, the participants used the ARPI for 59.07 s (max = 277.52 s).³ Table 11 in Appendix 5 presents the wording and the sources of all scales. We measured purchase intention with one item (“I intend to buy cereals of XY in the future,” M = 2.80) adapted from Dodds et al. (1991) and brand image with three items adapted from Hoffmann and Mueller (2009) and Mai et al. (2014) on five-point scales (M = 3.66, Cronbach’s α = .89). To operationalize the mediating mechanisms of the information’s comprehensiveness (M = 3.73, α = .91), we created a four-item scale on the basis of Rese et al. (2014, 2017) and Mai et al. (2014). We included several variables to control for and rule out alternative mediating processes (for two-item scales, we show Spearman and Brown’s ρ, Eisinga et al., 2013). The participants indicated perceived credibility on a two-item rating scale (M = 3.82, ρ = .90), perceived complexity on a two-item scale (M = 2.14, ρ = .76) adapted from Geissler et al. (2001), and perceived user friendliness with three items (M = 3.78, α = .74) adapted from Srinivasan et al. (2002). As controls, we measured the self-reported product category knowledge in the food domain with two five-point items (M = 3.81, ρ = .80) adapted from Chang (2004) and sustainability attitude with two nine-point items (M = 6.84, ρ = .84) based on Whitmarsh and O'Neill (2010). Confirmatory factor analysis (AMOS 25.0) with all multi-item constructs shows a good model fit (χ²₍₁₁₅₎ = 198.61, χ²/d.f. = 1.73; CFI = .98; RMSEA = .04). The analysis confirms discriminant validity, because each construct’s average variance extracted exceeds the maximum of the squared correlations with all latent variables (Fornell & Larcker, 1981). To track purchases, consumers indicated in the questionnaire whether they had bought a box (or more) of the focal brand.⁴ Trained interviewers estimated and documented further details on five-point scales. First, they coded the participants’ apparent stress. We dichotomized this variable to relaxed (59.2%) and stressed (40.8%). Second, they tracked how full the shopping carts were, because consumers are potentially more likely to purchase cereals if they purchase more items overall.

To gain more managerial insights, we created a rush hour index. Following previous field studies (e.g., Irmak et al., 2020), we use daytime to capture rush hour. We distinguish no rush hour (Monday–Friday morning and Saturday afternoon) vs. rush hours (Monday–Friday afternoon and Saturday morning).⁵ This distinction is in line with the Bring! Shopper Report (Bring!, 2020, p. 5) that presents the daily distributions of shopping trips for Germany, which are left-skewed for weekdays (Monday to Friday), with a peak at about 5 p.m. On Saturdays, the distribution is right-skewed, with a peak at 11 a.m. We conclude that the rush hour is characteristically in the afternoon on weekdays and in the morning on Saturdays. We validate the rush hour shopping based on the interviewers’ observation of the shoppers’ stress level (0 = relaxed, 1 = stressed). We found that on weekdays (Monday to Friday), the stress level was rather low before 2:00 p.m. and higher in the afternoon (M_morning = .29 vs. M_afternoon = .48), while on Saturdays the pattern is flipped (M_morning = .39 vs. M_afternoon = .13).

We gathered data of 271 shoppers who used the ARPI. For contrasting the ARPI to traditional means of informing consumers, we asked 153 additional shoppers to use a booklet providing the same information. Within both groups, we randomly assigned participants to one of the four treatments.⁶ The exclusion of shoppers who used the ARPI app for less than three seconds resulted in 250 ARPI users.⁷ The excluded participants do not differ from the sample in terms of gender (χ²(1) = 1.28, p = .258) or age (t(415) = .24, p = .814). The age of the participants ranged from 16 to 82 years, with a mean of 39.6 years (SD = 17.8); 39.9% of the participants were men.

We now seek to understand better when controllability impairs perceived comprehensiveness, and ultimately brand image and purchase intentions (referring to H2). We spotlight different degrees of ARPI controllability. In the main study, users could access multiple paths of information (category-statements-details) and they could decide when to exit the ARPI. Consumers may react differently if they can choose only one path of information (category-statements-details) or can follow multiple paths with an exit option. For this reason, we manipulated the degree of ARPI controllability. We focus on detailed information because the backfire effects were particularly strong in this condition in the main study. For generalizability, we use another brand and chocolate cereal bars as the product category.

We ran a between-subjects experiment with three treatment groups and a control group. Similar to recent AR research (Barhorst et al., 2021), we created videos showing a shopping scene from a first-person perspective. Participants were instructed to imagine that they were this person. In the treatment groups, they were also instructed to imagine using an AR tablet that enables them to see additional product information as soon as they fixate a product. The video shows a supermarket shelf containing several boxes of cereal bars (Web-Appendix C for screenshots), and the person takes one box of chocolate cereal bars. The short video ended with the person holding the box in the hand. The degree of ARPI controllability was manipulated as follows. [1] Control group without ARPI (N = 70): The participants saw the box of cereal bars being taken from the shelf, but did not receive any additional information. [2] Controllability, one round (80): The participants saw three virtual text fields appear with the categories ecological, social, and regional hovering on the cereal bars box. They could choose one of the fields, and three related statements appeared. They could then pick a statement to which they received three details. [3] Controllability, multiple rounds (90): This condition consisted of three rounds. Each round was identical to the condition [2] (Controllability, one round), but at the end of rounds 1 and 2, the participants were referred to the starting screen to choose a category again. [4] Controllability, multiple rounds with exit option (70): This condition also includes three rounds, but in each step of the interaction the participants could indicate having received enough information and leave the ARPI. All product details were identical to the statements and details in the main study.

Next, the participants completed a short questionnaire. We measured brand image with three items as in the main study (M = 5.57, α = .91), and purchase intention with three items (M = 4.77, α = .95). As a more behavioral variable, we presented our focal brand and a competing brand. Box sizes were identical; the packaging design (colors, pictures) was very similar. The participants indicated on a seven-point scale whether they would rather buy the competing brand (1) or the brand presented in the ARPI (7) (M = 4.21). To operationalize the mediating mechanisms, the participants indicated their perceptions of the information’s comprehensiveness on the same scale as in the main study (M = 5.66, α = .91). As controls, we again measured product category knowledge (M = 4.08, ρ = .88) and sustainability attitude (M = 4.75, ρ = .89). Table 11 in Appendix 5 provides the sources, wording of items, and psychometric properties.

We gathered data of 310 U.S. consumers via Prolific. Their age ranged from 18 to 82 years, with a mean of 36.8 years (SD = 12.7); 41.6% are male, 58.1% female, and 0.3% diverse.

We ran OLS regressions with dummy variables for the groups. Condition [4] (controllability, multiple rounds with exit option) imitates the main study’s controllable-detailed condition; therefore, we use this condition as contrast. The analyses show that all groups are assessed more positively than group [4]. This occurred for perceived comprehensiveness (group [1]: β = .425, t = 6.374, p < .001; [2]: β = .155, t = 2.290, p = .023; [3]: β = .245, t = 3.596, p < .001), brand image ([1]: β = .214, t = 3.069, p = .002; [2]: β = .140, t = 1.977, p = .049; [3]: β = .199, t = 2.783, p = .006), purchase intention ([1]: β = .239, t = 3.458, p < .001; [2]: β = .279, t = 3.989, p < .001; [3]: β = .179, t = 2.528, p = .012), and choice ([1]: β = .138, t = 1.962, p = .051; [2]: β = .158, t = 2.221, p = .027; [3]: β = .152, t = 2.114, p = .035). When entering the controls (sex, age, education, sustainability attitude, product category knowledge), the pattern of results remains (Table 6). Mediation analyses (Web Appendix C) confirm that perceived comprehensiveness mediates how the different types of the treatment shape brand image and purchase intention.

Follow-up study 1 confirms that the backfire effect of AR controllability arises due to reduced perceptions of information comprehensiveness. This negative effect, however, occurs mostly for higher degrees of controllability: when controllability is spread across multiple rounds and users have an exit option. Under these specific conditions, consumers realize that the cues of the richer medium are filtered out, and there was more information available such that they lack part of this information.

This study elaborates the downstream consequences when applying lower degrees of information controllability (as established in follow-up study 1) and scrutinizes which positive effects on interaction quality inhibit or offset the backfire effect via perceived comprehensiveness. To advance our answers to H2, we include one round (category-statements-details) without an exit option as a low degree of controllability. Referring to H3, we expanded the interaction quality’s set of indicators as competing mediating processes. We test for variables that are relevant in AR settings, such as presence (Hilken et al., 2017; Javornik, 2016b), novelty (Yim et al., 2017), as well as utilitarian and hedonic benefit (Holdack et al., 2020; Rese et al., 2014). Moreover, this study compares the effect to that of a control group without ARPI. As another variation to increase the external validity, we use a video with AR glasses (Holdack et al., 2020; Rauschnabel et al., 2018) instead of AR tablets.

We ran a 2 (controllable vs. uncontrollable) × 2 (detailed vs. nondetailed) between-subjects online experiment, adding a control group without an AR condition. We created videos for all five conditions, showing from a first-person perspective how a person takes one box of chocolate cereals from a shelf. Participants were instructed to imagine that they wear augmented reality glasses, which enable them to see additional product information as soon as they fixate a product. The experimental treatments were manipulated as follows. [1] Uncontrollable-nondetailed (N = 66): Three fields (ecological, social, regional) hovered on the cereals box. When participants clicked to proceed, they saw all three ecological statements; after another click, the ecological statements disappeared and three social statements appeared, followed by three regional statements. [2] Uncontrollable-detailed (86): This is the same as condition [1], except that the participants also saw three details for each statement. [3] Controllable-nondetailed (92): The participants could choose either the ecological, social, or regional field; afterward, a corresponding statement appeared. [4] Controllable-detailed (91): As in the third condition, the participants could choose one of the fields (ecological, social, or regional) to see the corresponding statement. Afterward, the participants could pick a statement to which they received a corresponding detail. [5] The control group (85) saw the video with the cereals box, but did not receive any additional information. The procedure is detailed in Web-Appendix D. The brand and all product details are identical to those used in the main study.

We again measured brand image (M = 5.40, α = .89), purchase intention (M = 4.29, α = .95), and as a more behavioral variable, choice of the focal brand over a competitor brand (M = 4.56). For the mediating mechanisms, the participants again indicated their perceptions of the information’s comprehensiveness (M = 5.16, α = .94), perceived credibility (M = 5.25, ρ = .92), perceived complexity (M = 2.15, ρ = .85), and perceived user friendliness (M = 4.75, α = .77). Extending the main study’s mediators, we added scales for presence (M = 4.43, α = .87) adopted from Huang and Liao (2015), novelty (M = 5.26, α = .82), utilitarian benefits (M = 4.42, α = .80) and hedonic benefit (M = 4.23, α = .92) taken from Voss et al. (2003). As controls, we again measured the self-reported product category knowledge (M = 4.75, ρ = .84) and sustainability attitude (M = 5.10, ρ = .81). Usage time was tracked (M = 67.47 s). Table 11 in Appendix 5 provides the sources, wording of items, and psychometric properties. We gathered data of 420 consumers via Prolific (M_age = 30.0 years, SD = 9.9; 57.9% of the participants are men, 41.2% female, and 1.0% diverse; 52.6% have a university degree).

First, we contrasted the four experimental conditions against the control group without a treatment (group [5]). We ran OLS regressions, which show that all experimental groups produced a more positive brand image (group [1]: β = .201, t = 3.429, p < .001; [2]: β = .178, t = 2.936, p = .004; [3]: β = .143, t = 2.334, p = .020; [4]: β = .244, t = 3.990, p < .001). Moreover, purchase intentions ([1]: β = .122, t = 2.048, p = .041; [2]: β = .091, t = 1.473, p = .142; [3]: β = .059, t = .954, p = .341; [4]: β = .160, t = 2.591, p = .010) and preferences ([1]: β = .069, t = 1.167, p = .244; [2]: β = .130, t = 2.110, p = .035; [3]: β = .051, t = .817, p = .414; [4]: β = .144, t = 2.326, p = .020) were higher for half of the experimental groups compared to the control group.

In a second step, an OLS regression confirms the ARPI controllability’s influence on perceived comprehensiveness (β = −.397, t = −7.892, p < .001). The effect remains stable (β = −.388, t = −7.700, p < .001) when entering the control variables gender (β = −.006, t = −1.259, p = .209), age (β = −.060, t = −1.173, p = .242), education (β = −.098, t = −1.686, p = .093), product category knowledge (β = .042, t = .738, p = .481), and sustainability attitude (β = −.036, t = −.624, p = .533). For a full report, please consult the Web-Appendix D.

Finally, we ran mediation analyses (Table 12 in Appendix 6) that confirm the negative mediating influence of perceived comprehensiveness. Controllability decreases perceived comprehensiveness, which, in turn, influences the brand image and purchase intention. This effect can be found for both detailed and nondetailed information. For nondetailed information, there is also a mediating effect through perceived credibility. Adding to the main study, in the detailed conditions, there are further mediating effects of perceived complexity, perceived user friendliness, utilitarian benefits, and hedonic benefits.

This follow-up study spotlighted lower degrees of controllability. Although the study confirms, again, information controllability’s negative influence on perceived comprehensiveness, in the condition of detailed information, there are also positive mediating effects via interaction quality, including (reduced) perceived complexity, user friendliness, utilitarian benefit, and hedonic benefit. Note that in contrast to the main study, we applied an online experiment to abstract the research model from the environmental setting of the point of sale. Presumably, due to this abstraction and the lower level of controllability, the positive mediators (known from online studies) are more pronounced. Still, our main finding that controllability reduces the perceptions of information comprehensiveness, which implies negative (countervailing) consequences for the marketing-relevant downstream variables, is replicated in this setting. Notably, contrasting ARPI against the group without ARPI supports the generally positive effect of the ARPI.

In the final stage of our research, we explored, in two follow-up studies (3 and 4), the moderating role of rush hour (H4) because this boundary condition carries strong managerial implications for retailers. We ran another study in a field setting that employed a more objective criterion (the number of customers purchasing in a given period). The second objective is to contrast ARPI users with regular shoppers based on hard data taken from check-out scanners. Third, our previous studies focused on only one brand in one product category. This study considered another product category with several product types and brands. We did not manipulate the ARPI design; instead, we focused on the critical controllable-detailed information condition. In the main study, we found the backfire effect to occur in this condition in busy shopping times.

We ran the study in the same hypermarket as in the main study. We used an ARPI app that we designed exclusively for this study to give information about beer products. The participants could scan the product’s front or entire beer boxes to access the ARPI. The ARPI recommends other beer products that fit the customers’ preferences (see Web-Appendix E for a description). The hypermarket in which we ran the study offers 316 different items in the beer category (different types, brands, and containers, such as bottles and boxes), which we included in the app (we excluded cans and barrels). The price per bottle ranged from €0.29 to €2.79 for craft beer. Bottle sizes ranged from 0.2 to 1.0 l. The study covered a wide range of beer types, including non-alcoholic beer and beer-mixed drinks.

We collected the data over three weeks. As in the main study, we balanced the treatments across daytime slots and weekdays. Interviewers approached shoppers in the hypermarket’s beer section (opening claim: “Find out more about your beer!”). Most shoppers agreed to participate, and the interviewers briefly showed them the app’s basic functionality; shoppers used the app without an interviewer present to ensure that the shoppers made their choice at their own pace.

To determine the ARPI effect, we contrasted the ARPI users with the scanner data of the entire population, i.e., all other consumers who bought beer during the time that we ran the study, as well as about one week before and two weeks later (n = 10,657). We received usable data from 51 subjects who used the AR. The instructors guessed and coded the participants’ age (≤ 30 years: 41.9%, 31–50 years: 32.5%, > 50 years: 25.6%) and gender (male: 25.6%, female: 39.5%, male/ female couples: 34.9%). We applied the app’s tracking data and check-out scanner data. The AR tracked the time shoppers used the device. The subjects used the app for 86.1 s (SD = 90.7). Excluding the time for reading the instruction screen, the net usage time was 66.9 s (SD = 41.0). Having finished their shopping, we asked the shoppers to hand in their shopping receipts.

We coded the rush hours and more relaxed shopping hours as in the main study. As an additional measure, we coded the crowding in the retail area from the scanner data. For this proxy, we determined the number of customers in the hypermarket who purchased within 30 min (M = 17.77, SD = 9.70), ranging from one to a maximum of 60 purchasing customers.

More than half of the ARPI users (53.2%) bought the beer they initially planned (beer type and brand), meaning that 46.8% deviated from their first choice. As another indication of ARPI’s potential to shift consumption patterns, 61.7% bought the recommended brand, although they may have switched the beer type (note that several users bought more than one brand or item). Likewise, 63.8% bought the recommended type, although they might have changed the brand. In sum, 55.3% bought exactly the beer (beer type and brand) the ARPI recommended.

To assess whether the ARPI resulted in more and different purchases, we compared the scanner data for the ARPI users with the scanner data for the population who purchased beer in the same period. The ARPI led to a notable shift and extension in consumers’ choice sets. The most popular and commonly purchased beer type sold in the hypermarket is pilsner. The share of consumers who bought only pilsner decreased from 29.1% to 23.5%, while the share of those who combined pilsner with other beer types when buying different types at the same time increased from 10.0% to 21.6% (χ²₍₁₎ = 6807.325, p < .001). The average number of bottles of other beer types (not pilsner) per shopper almost doubled (15.7 vs. 7.9, + 98.6%, t_(10,706) = 4.150, p < .001, Cohen’s d = .583).

As shown in Fig. 5 and supporting the main study’s findings regarding H4, the ARPI usage’s effect in shifting consumption greatly varies across rush hours and more relaxed shopping hours. There is a large ARPI influence on the number of bottles of other beer types (not pilsner) that consumers buy (ANOVA: main effect ARPI: F_(1,10,704) = 24.315, p < .001, main effect rush hour F_(1,10,704) = 13.182, p < .001, interaction: F_(1,10,704) = 10.011, p = .002). The ARPI only helped increase the variety of bought beer types during relaxed shopping times. Note that there is no effect on the number of pilsner bottles bought during rush hours (main effect ARPI: F_(1,10,704) = .027, p = .870, main effect rush hour F_(1,10,704) = .381, p = .537, interaction: F_(1,10,704) = .548, p = .459).⁹ This interaction effect also occurs when applying the more objective operationalization of rush hour, namely crowding (interaction: B = −.496, t = 2.320, p = .020). The positive ARPI-effect only occurs when fewer (9) or average (16) numbers of customers are in the market (both ps < .001), but turns insignificant when many (27) customers were shopping in the 30 min period (p = .272).

As a robustness check, we also conducted propensity score matching on a large set of shopping characteristics detailed for certain product categories (e.g., amount of and spending on purchased food products; Web-Appendix E). These checks support the managerial relevance of the AR technology and the importance of consumer stress in busy shopping times.

With hard scanner data, this study substantiated that ARPI can influence purchase behaviors at the point of sale. The ARPI was found to shape consumer decisions, as more than half of the ARPI users bought another beer type and/or brand than initially planned. Most importantly, the study further supported H4 and the adverse effects in busy shopping times.

The results of the main study imply that in rush hours, the backfire effect is particularly prominent, the more detailed the accessible information is. Thus far, we explored this critical contextual factor through proxies, such as the daytime (main study) or the number of purchasers in a given period (follow-up study 3). To gain further insights, we manipulated this variable in a two-level (rush hour vs. no rush hour) experiment in a fashion lab store of a French business school (133 students, 20.03 years, 53% females), where participants used an AR device to gain access to additional (sustainability-related) product information. Participants searched for a fashion item as a present for a friend. The focal product’s brand provides additional information on the item (a sweater) and the labels. This information was, however, shown in another language (German, in this case). To access the additional product information, participants used a tablet with an AR app that allows translations in real time. As for the ARPI in the main experiment, the AR function thus showed the additional (translated) information directly on the product. Participants used the AR device in both experimental conditions for the same time, but with vs. without being rushed. In the rush hour condition, the participants were informed that they are “in a hurry because the next customer wants to look at the product too” and pop-up messages reinforced this aspect (see Web-Appendix F for a more detailed description of the study design). This manipulation was successful, as participants experienced greater stress in this condition (one-tailed testing, M_{no rush} = 2.61; M_rush = 4.42, t = 5.715, p < .001). This study zooms in on whether rush hour favors more nondetailed information; therefore, we assessed how comprehensive the participants perceive the information to be. Since the study instructions ask the participants to imagine that they intend to purchase the sweater, we used their willingness to pay as a purchase-relevant variable in this study.

Regression analysis (one-tailed) shows a significant interaction effect of rush hour and information comprehensiveness on willingness to pay (B = −19.34, t = −1.698, p = .046). AR use while being rushed had no implications when participants perceive the accessed information to be less comprehensive (B_-1SD = 6.89€, t = .431, p = .333) or average (B_mean = −12.45€, t = −1.101, p = .137). However, AR use while being rushed reduces willingness to pay by around €32 when participants perceive this information to be more comprehensive (B_+1SD = −31.80€, t = −1.974, p = .025). To ensure that this shift is not driven by outliers, we reran the test with ranks, or when eliminating extreme willingness to pay, with the same results. Furthermore, we assessed brand image for which we find a similar negative shift (interaction: B = −.34, t = −2.316, p = .011).

This final experiment confirmed the key role of rush hour as a context factor for ARPI. Confirming our main study and follow-up study 3, the use of AR while being rushed hampers marketing outcomes (e.g., willingness to pay), especially when consumers perceive the accessible product information as being overly comprehensive. This liability in rush hours does not arise if consumers perceive the accessed information to be less comprehensive.

ARPI can augment the limited physical space with digital supplements and create virtual space that provides almost unlimited information compared to traditional means of communication. This research has demonstrated that ARPI offers the potential to shape marketing outcomes, such as brand image and purchase intention. As the pre-studies show, AR technology outperforms other marker-based sources of information, such as QR codes, because it creates a stronger tie between product and information in the consumer’s mind. ARPI is, therefore, the preferable tool to provide information at the point of sale when the physical space is limited. However, marketers and retailers need to be aware of the backfire effect when the AR is controllable and lots of detailed information is principally accessible. This is important to stress, because this drawback is ARPI specific. In traditional media (e.g., booklets), the body in which the information chunks are embedded serves as a peripheral cue. In the digital world, consumers arguably have greater difficulty gathering knowledge (consciously or unconsciously) about the realm of accessible information. The amount of information that is accessible via ARPI is less salient to them. While the use of AR is generally recommended, its superiority over more traditional communication methods should thus be checked before implementing the technology.

Using AR in retailing provides several functionalities including informing, placing, and trying-on. These functionalities offer different benefits for different retailing channels. Contrasting the present research with the extant AR studies in e-commerce (Table 7 in Appendix 1) pinpoints that the ARPI functionality of informing in physical stores differs from the characteristic AR functionalities of placing and trying-on in e-commerce. ARPI designers, marketers, and retailers should be well aware that the hedonic benefit is more relevant for AR applications in e-commerce. However, our pre-studies revealed that ARPI users are guided by AR’s utilitarian benefit instead. In physical retailing settings, marketers and retailers should therefore ensure that the AR technology provides relevant details and facts, which support the consumption decision, instead of trying to entertain the shopper.

This research has demonstrated that ARPI will only be effective if marketers and retailers design their applications carefully. The risk that ARPI controllability backfires is contingent on other design elements, such as the detailedness of information. Our dual framework and the empirical investigations emphasize the existence of two countervailing mechanisms that the ARPI’s controllability can activate. Especially for fairly nondetailed information, the controllability stimulates positive inferences about information credibility, with even favorable consequences for buying intentions and brand image. If retailers seek to augment concise information, they may thus use a controllable approach. Conversely, when providing detailed information, AR controllability more likely provokes perceptions that relevant cues are filtered out. ARPI designers should therefore include elements that offer relevant cues to the available information’s comprehensiveness. Alternatively, retailers may limit the possibility of filtering out too many details to avoid harming perceived information comprehensiveness. Not allowing consumers to exit at any time could help as well. Although not tested within our research, certain design elements, such as content overviews and visualizations of the content structure, may help.

In times of retail crowding and when shoppers are more stressed (Monday–Friday afternoons and Saturday mornings), retailers are well advised to avoid delivering detailed information in a controllable fashion. For consumers in a hurry, concise information via AR is helpful, while more detailed information is not helpful if the AR is controllable. In more relaxed shopping situations, consumers are more open to consider additional details, which ultimately feeds into their decisions. Retailers may consequently adjust the AR application to the specific situation or might take measures to nudge a certain shopping mood in rush hours before equipping consumers with ARPI. Various ambient measures, such as slow music or relaxing lighting, could help consumers slow down (Biswas et al., 2017, 2019). Beyond adjusting ARPI according to daytimes and weekdays, setting the ideal default for controllability and detailedness may be best to improve ARPI effectiveness. Goldstein et al. (2008) propose a decision tree for different defaults that can be adopted for the ARPI design. If individual preferences are known, the AR app could use the customer’s ideal setting as default (persistent default). If information is available for similar customers only (target group), a smart default adjusting to the target group preferences can be applied. If no customer information is available, adjusting the design concerning daytime and weekdays could be a benign default. In the long run, integrating ARPI usage data and scanner sales for machine learning should help continuously optimize the design.