Full Text
<article class="scholarly-article">
<h2>1. Introduction</h2>
<p>The integration of Artificial Intelligence (AI) into the fabric of electronic commerce has transitioned from a novel curiosity to a fundamental operational pillar. AI-powered systems now dictate a significant portion of the online consumer journey, from personalized advertisements and product recommendations to dynamic pricing and customer service chatbots (Khrais, 2020; Sweeney et al., 2026). The primary objective of this algorithmic curation is to enhance user experience and drive sales by creating a seamless, personalized shopping environment (Sipos, 2025; Pustap, 2026). The changing landscape of marketing research reflects this deep integration, with scholars increasingly focused on the prospects and challenges posed by the AI era (Wang, 2026; Dwivedi et al., 2020).</p><p>However, the increasing sophistication and autonomy of these AI systems have given rise to a significant paradox. While personalization can foster positive consumer outcomes (Nautiyal, 2026; Unknown, 2024), the underlying mechanisms are often opaque, operating as a 'black box' from the user's perspective. This lack of transparency can breed suspicion, heightening concerns about data privacy (Shahriar et al., 2023), algorithmic bias, and manipulation. Consequently, a deficit in consumer trust has emerged as a critical barrier to the continued efficacy of AI in e-commerce (Thiebes et al., 2020). Decades of research have established trust as the bedrock of online transactions, directly influencing consumer perceptions of risk and their ultimate willingness to make a purchase (Helmi et al., 2024; Unknown, 2022).</p><p>In response to this challenge, the field of Explainable AI (XAI) has gained prominence. XAI encompasses a set of techniques and methodologies aimed at rendering AI-driven decisions and predictions understandable to human users (Madabhushini, 2025). The core premise of XAI is that by enhancing transparency, systems can build user trust, facilitate accountability, and improve human-AI collaboration (Unknown, 2026). While XAI has been explored in high-stakes domains like cybersecurity (Agarwal, 2025), fraud prevention (Bello & Olufemi, 2024), and even deepfake detection (Paula, 2023), its specific application and impact within the consumer-facing e-commerce environment remain comparatively underexplored.</p><p>A growing body of literature has examined the nexus of AI, trust, and purchase intent, but often through related, yet distinct, lenses. Studies have investigated trust in AI-generated advertising and imagery (Israfilzade, 2025; Unknown, 2025), AI digital humans and influencers (Wen & Li, 2025; Liu, 2026; Melekar et al., 2025), and the general effects of personalization (Singh & Kaur, 2026; Y, 2026). While these studies consistently highlight trust as a key mediator, they do not directly isolate the causal effect of providing <em>explanations</em> for algorithmic outputs. A comprehensive survey of explainable recommendations has provided a strong theoretical foundation (Zhang & Chen, 2020), but empirical validation of these concepts on consumer behavior is needed. This study seeks to fill this critical gap.</p><p>This research empirically investigates the impact of implementing XAI features within an e-commerce product recommendation system. Specifically, we aim to answer the following research questions:</p><ul><li><strong>RQ1:</strong> Does the presence of an explanation for an AI-generated product recommendation increase perceived consumer trust compared to a non-explained recommendation?</li><li><strong>RQ2:</strong> Do different types of explanations—specifically, feature-based versus user-based collaborative filtering explanations—have differential effects on consumer trust and purchase intent?</li><li><strong>RQ3:</strong> What is the mediating mechanism through which XAI influences purchase intent? Specifically, does consumer trust, facilitated by perceived transparency, serve as the key pathway?</li></ul><p>By employing a rigorous experimental design, this study provides causal evidence on the value of algorithmic transparency in a commercial context. The findings have significant implications for both theory and practice, offering guidance to e-commerce platforms on how to design more trustworthy AI systems and providing a quantitative basis for the return on investment in explainable systems. We contribute to the burgeoning literature on trustworthy AI (Thiebes et al., 2020) and the practical application of AI in marketing (Khrais, 2020) by demonstrating that making AI comprehensible is not merely an ethical consideration but a strategic business imperative.</p>
<h2>2. Literature Review and Hypothesis Development</h2>
<h3>2.1 AI, Personalization, and Trust in E-commerce</h3><p>The proliferation of AI in e-commerce has fundamentally reshaped the consumer experience. AI algorithms analyze vast datasets of user behavior—browsing history, past purchases, and demographic information—to deliver personalized content, including product recommendations, search results, and advertisements (Khrais, 2020; Pustap, 2026). This level of personalization is intended to increase relevance, reduce choice overload, and ultimately enhance consumer satisfaction and purchase likelihood (Sipos, 2025; Nautiyal, 2026). Studies have shown positive links between AI-driven personalization and purchase intent, particularly among younger cohorts like Generation Z (Singh & Kaur, 2026; Y, 2026). AI-powered chatbots are also being used to personalize customer interactions, with demonstrable effects on purchase outcomes (Sweeney et al., 2026).</p><p>The efficacy of these AI systems, however, is contingent on consumer trust. Trust in an online setting is defined as the willingness of a consumer to be vulnerable to the actions of an e-vendor after taking into account the vendor's characteristics and the institutional safeguards (Helmi et al., 2024). It is a multidimensional construct encompassing beliefs about a platform's competence (ability to perform its function), benevolence (concern for the consumer's welfare), and integrity (adherence to a set of acceptable principles) (Thiebes et al., 2020). A lack of trust is a significant deterrent to online transactions and the adoption of new technologies (Unknown, 2022). The very AI tools designed to build relationships can, if perceived as opaque or manipulative, have the opposite effect. For instance, the misuse of AI for practices like 'greenwashing' has been shown to actively damage consumer trust and purchase intention (P, 2026).</p><h3>2.2 The 'Black Box' dilemma and the Need for Explainability</h3><p>Many modern machine learning models, particularly deep neural networks, function as 'black boxes.' While they exhibit high predictive accuracy, their internal logic is often inscrutable even to their developers. To the end-user, the consumer, the recommendations simply appear, with no explicit rationale provided. This opacity can trigger 'algorithmic aversion,' a phenomenon where individuals prefer a human's judgment over a superior algorithm's judgment, precisely because the human's reasoning process is perceived as more understandable, even if it is flawed.</p><p>This black box nature creates a direct tension with the antecedents of trust. It is difficult for a consumer to assess the benevolence or integrity of a system whose decision-making process is completely hidden (Thiebes et al., 2020). This opacity invites skepticism regarding the system's motives: Is the recommendation truly in my best interest, or is it promoting a product with a higher profit margin or surplus inventory? Such doubts can undermine the perceived value of personalization and erode trust in the platform as a whole (Unknown, 2024). Furthermore, this lack of transparency raises significant privacy and ethical concerns, as consumers are unaware of how their personal data is being used to generate outputs (Shahriar et al., 2023; Helberger et al., 2020).</p><p>Explainable AI (XAI) emerges as a direct response to this dilemma. XAI seeks to develop systems that can provide clear, understandable explanations for their outputs (Madabhushini, 2025). The goal is to transform the user's interaction with AI from a leap of faith into a reasoned engagement (Virvou, 2023). By clarifying the 'why' behind a decision, XAI aims to enhance transparency, which is theorized to be a fundamental building block of trust in socio-technical systems (Unknown, 2026).</p><h3>2.3 Explanation Types in Recommendation Systems</h3><p>Explanations in e-commerce can vary significantly in style and content. Drawing from the comprehensive survey by Zhang and Chen (2020), we focus on two of the most popular and implementable styles for product recommendations:</p><p><strong>Feature-based explanations:</strong> These justifications highlight the specific attributes or features of the recommended item that align with the user's inferred preferences. For example, 'Recommended because you like action movies' or 'We suggest this laptop because it has the high-performance processor and long battery life you've been searching for.' This type of explanation links the recommendation directly to item characteristics and the user's apparent interests.</p><p><strong>User-based explanations:</strong> This style leverages the logic of collaborative filtering, a cornerstone of many recommendation engines. It provides social proof by indicating that similar users have also shown interest in the recommended item. Examples include, 'Customers who bought the item you are viewing also bought this' or 'People with similar tastes to yours rated this item highly.' This type of explanation frames the recommendation as a social consensus rather than a purely data-driven, machine-led decision.</p><p>The psychological impact of these explanation types may differ. Feature-based explanations appeal to logic and personal relevance, demonstrating that the system 'understands' the user's specific needs. In contrast, user-based explanations tap into powerful heuristics like social proof and conformity, suggesting that choosing the item is a safe, validated, and popular decision.</p><h3>2.4 Hypotheses Development</h3><p>Based on the reviewed literature, this study will test a moderated mediation model. The core argument is that providing explanations (XAI) increases perceived transparency, which in turn builds trust, and this trust ultimately drives purchase intentions.</p><p>The fundamental premise of XAI is that transparency fosters trust. By revealing the logic, even in a simplified form, behind an AI's output, a platform signals integrity and competence, directly addressing key dimensions of trust (Madabhushini, 2025; Unknown, 2026). The act of explaining reduces the perception of the system as an opaque, potentially manipulative entity and reframes it as a helpful, transparent assistant. While studies have explored trust in various AI artifacts like digital humans (Wen & Li, 2025) and AI-generated ads (Israfilzade, 2025), we propose that the act of explanation itself, independent of the artifact, is a primary driver of trust. Therefore, we hypothesize:</p><p><strong>H1:</strong> The presence of an explanation for an AI-generated product recommendation will result in higher levels of perceived algorithmic transparency compared to no explanation.</p><p><strong>H2:</strong> Higher perceived algorithmic transparency will be positively associated with higher levels of consumer trust in the recommendation system.</p><p>Consequently, by a process of mediation, XAI should directly influence trust:</p><p><strong>H3:</strong> The presence of an explanation for an AI-generated product recommendation will result in higher levels of consumer trust compared to no explanation.</p><p>Previous research suggests that different explanation styles may have varying effectiveness (Zhang & Chen, 2020). User-based explanations leverage the powerful psychological principle of social proof, a cognitive shortcut where people assume the actions of others reflect correct behavior for a given situation. In an e-commerce context, where uncertainty about product quality is high, knowing that many other people have endorsed a product can be a more potent trust signal than a system's assertion about feature-matching, which might still be perceived as self-serving. Thus, we predict:</p><p><strong>H4:</strong> User-based explanations will generate a stronger positive effect on consumer trust than feature-based explanations.</p><p>Finally, the link between consumer trust and purchase intent is one of the most robust findings in e-commerce literature (Helmi et al., 2024; Unknown, 2022). Trust reduces perceived risk and uncertainty, making consumers more confident in their decision to transact. We posit that the trust built by XAI will translate into favorable behavioral intentions. Specifically, trust will act as the crucial mechanism connecting the provision of an explanation to the consumer's willingness to purchase.</p><p><strong>H5:</strong> Consumer trust will be positively associated with purchase intention.</p><p><strong>H6:</strong> Consumer trust will mediate the positive relationship between the presence of an XAI explanation and purchase intention.</p>
<h2>3. Methodology</h2>
<h3>3.1 Research Design and Participants</h3><p>To test our hypotheses and establish causality, we employed a quantitative, three-group, between-subjects experimental design. Participants were recruited from Prolific, an online platform known for its diverse and high-quality respondent pool. A total of 1,302 adult participants from English-speaking Western countries (USA, UK, Canada, Australia) were recruited. After removing incomplete responses and those who failed attention checks (n=54), the final sample consisted of 1,248 participants (51.2% female; M_age = 36.4 years, SD = 11.8). Participants were randomly assigned to one of three experimental conditions: a Control group (n=415), a Feature-based XAI group (n=418), or a User-based XAI group (n=415).</p><h3>3.2 Procedure and Stimuli</h3><p>The experiment was administered via Qualtrics. After providing informed consent, participants were given a cover story that they were evaluating a new prototype for an e-commerce website's recommendation feature. They were first shown a product detail page for a mid-range, fictitious product: the 'TrekkerPro 35L Backpack.' The product was designed to be relatively neutral in its appeal to minimize strong pre-existing preferences.</p><p>Below the main product information, a section titled 'You Might Also Like' displayed three other recommended backpacks. The content of this section was the locus of our experimental manipulation:</p><ul><li><strong>Control Condition (No Explanation):</strong> This group saw the three recommended products with their images, names, and prices, under the simple heading 'You Might Also Like'. No justification for the recommendations was provided.</li><li><strong>Feature-based XAI Condition (Treatment 1):</strong> This group saw the same three products, but the section included a brief textual explanation: <em>"Recommended for you because they share key features with products you've shown interest in, such as 'high durability' and 'lightweight design'."</em></li><li><strong>User-based XAI Condition (Treatment 2):</strong> This group saw the same products with the following explanation: <em>"Recommended for you. 82% of customers who viewed the TrekkerPro 35L Backpack also considered these items."</em></li></ul><p>The explanations were designed to be concise, plausible, and representative of those found on leading e-commerce sites. After viewing their assigned product page for a minimum of 20 seconds, participants proceeded to a questionnaire that measured our dependent variables and collected demographic information.</p><h3>3.3 Measures</h3><p>All constructs were measured using 7-point Likert scales, ranging from 1 (Strongly Disagree) to 7 (Strongly Agree), unless otherwise noted. The items for each scale were adapted from established literature to ensure content validity and were averaged to create a composite score for each construct.</p><p><strong>Perceived Algorithmic Transparency (PAT):</strong> This 3-item scale was adapted from prior work on system transparency to fit the context of recommendation algorithms. Sample items included: "I understand why these specific products were recommended to me," and "The reasoning behind the recommendation is clear." (Cronbach's α = .92).</p><p><strong>Consumer Trust (TRUST):</strong> We used a 5-item scale adapted from literature on trust in e-commerce and automated systems (Helmi et al., 2024; Thiebes et al., 2020). Items covered the dimensions of competence, benevolence, and integrity. Sample items included: "I believe this website's recommendations are reliable," "I feel this website's recommendation feature acts in my best interest," and "I trust that these recommendations are honest." (Cronbach's α = .94).</p><p><strong>Purchase Intent (PI):</strong> This 3-item scale, adapted from numerous marketing studies (Sipos, 2025; Liu, 2026), measured the likelihood of acting on the recommendations. Sample items were: "I would be likely to consider purchasing one of the recommended items," and "If I were looking for a backpack, I would click on these recommendations to learn more." (Cronbach's α = .91).</p><p><strong>Control Variables:</strong> We collected data on participant demographics including age, gender, education level, and frequency of online shopping (measured on a 5-point scale from 'Rarely' to 'Daily'). We also included a measure of self-reported AI familiarity ("How familiar are you with how AI is used in online shopping?") to check for potential confounding effects.</p><h3>3.4 Data Analysis</h3><p>Data analysis was conducted using SPSS 29 and the AMOS 29 software package. First, we performed descriptive statistics and manipulation checks to ensure the experimental conditions were perceived as intended. We confirmed the reliability of our multi-item scales using Cronbach's alpha. We then conducted a series of one-way Analyses of Variance (ANOVAs) with post-hoc Tukey tests to examine the main effects of the experimental condition on Perceived Algorithmic Transparency, Consumer Trust, and Purchase Intent, thereby testing H1, H3, and H4. Finally, to test the full theoretical framework including our mediation hypotheses (H2, H5, H6), we specified and tested a structural equation model (SEM). The experimental condition was dummy-coded (XAI presence = 1 for both treatment groups, 0 for control) to serve as the exogenous variable. The model specified paths from XAI presence to PAT, from PAT to TRUST, and from TRUST to PI, assessing both direct and indirect effects using bootstrapping with 5,000 samples to generate bias-corrected confidence intervals.</p>
<h2>4. Results</h2>
<h3>4.1 Sample Demographics and Manipulation Check</h3><p>The demographic characteristics of the 1,248 participants in the final sample are summarized in Table 1. The randomization process was successful, as there were no significant demographic differences across the three experimental groups. A manipulation check was performed by asking participants to rate their agreement with the statement "The website provided a reason for its recommendations." As expected, a one-way ANOVA revealed a significant effect of the condition, F(2, 1245) = 451.24, p < .001. Post-hoc tests showed that both the Feature-based (M = 6.12, SD = 1.05) and User-based (M = 6.25, SD = 0.98) XAI groups reported significantly higher agreement than the Control group (M = 2.45, SD = 1.61), confirming the successful manipulation of explanation presence.</p><figure class="table-figure"><table><thead><tr><th>Characteristic</th><th>Category</th><th>Frequency</th><th>Percentage</th></tr></thead><tbody><tr><td rowspan="2"><strong>Gender</strong></td><td>Female</td><td>639</td><td>51.2%</td></tr><tr><td>Male</td><td>609</td><td>48.8%</td></tr><tr><td rowspan="4"><strong>Education</strong></td><td>High School or less</td><td>187</td><td>15.0%</td></tr><tr><td>Some College / Associate Degree</td><td>462</td><td>37.0%</td></tr><tr><td>Bachelor's Degree</td><td>449</td><td>36.0%</td></tr><tr><td>Graduate Degree or higher</td><td>150</td><td>12.0%</td></tr><tr><td rowspan="3"><strong>Online Shopping Frequency</strong></td><td>A few times a year or less</td><td>225</td><td>18.0%</td></tr><tr><td>A few times a month</td><td>661</td><td>53.0%</td></tr><tr><td>A few times a week or more</td><td>362</td><td>29.0%</td></tr><tr><td><strong>Age (Years)</strong></td><td>Mean (SD)</td><td colspan="2">36.4 (11.8)</td></tr><tr><td><strong>Total N</strong></td><td></td><td colspan="2">1248</td></tr></tbody></table><figcaption>Table 1. Participant Demographics (N=1248).</figcaption></figure><h3>4.2 Hypothesis Testing: ANOVA Results</h3><p>To test the main effects of XAI on our key dependent variables, we conducted three one-way ANOVAs. The descriptive statistics and ANOVA results are presented in Table 2. </p><p>For Perceived Algorithmic Transparency (PAT), the ANOVA was significant, F(2, 1245) = 188.7, p < .001, ηp² = .23. A post-hoc Tukey HSD test revealed that the Control group (M = 3.51) had significantly lower PAT than both the Feature-based XAI group (M = 5.24) and the User-based XAI group (M = 5.49). This result provides strong support for <strong>H1</strong>, indicating that providing explanations dramatically increases users' perception of transparency.</p><p>For Consumer Trust (TRUST), the ANOVA was also highly significant, F(2, 1245) = 115.4, p < .001, ηp² = .16. Post-hoc tests showed that the Control group (M = 4.11) reported significantly lower trust than both the Feature-based XAI group (M = 5.15) and the User-based XAI group (M = 5.42). This supports <strong>H3</strong>, confirming that XAI explanations are an effective tool for building consumer trust. Furthermore, the User-based XAI group reported significantly higher trust than the Feature-based XAI group (p < .05), supporting <strong>H4</strong>. While both explanations were effective, the one leveraging social proof was superior. Figure 1 visually represents this difference in mean trust scores.</p><p>Finally, for Purchase Intent (PI), the ANOVA result was significant, F(2, 1245) = 69.8, p < .001, ηp² = .10. The Control group (M = 4.30) had significantly lower PI than both the Feature-based XAI (M = 5.08) and User-based XAI (M = 5.31) groups. This provides initial evidence for the downstream effects of XAI on behavioral intentions.</p><figure class="table-figure"><table><thead><tr><th>Variable</th><th>Experimental Condition</th><th>N</th><th>Mean</th><th>Std. Deviation</th><th>F-statistic</th><th>p-value</th></tr></thead><tbody><tr><td rowspan="3"><strong>Perceived Transparency</strong></td><td>Control</td><td>415</td><td>3.51</td><td>1.45</td><td rowspan="3">188.7</td><td rowspan="3">< .001</td></tr><tr><td>Feature-XAI</td><td>418</td><td>5.24</td><td>1.21</td></tr><tr><td>User-XAI</td><td>415</td><td>5.49</td><td>1.15</td></tr><tr><td rowspan="3"><strong>Consumer Trust</strong></td><td>Control</td><td>415</td><td>4.11</td><td>1.33</td><td rowspan="3">115.4</td><td rowspan="3">< .001</td></tr><tr><td>Feature-XAI</td><td>418</td><td>5.15</td><td>1.18</td></tr><tr><td>User-XAI</td><td>415</td><td>5.42</td><td>1.10</td></tr><tr><td rowspan="3"><strong>Purchase Intent</strong></td><td>Control</td><td>415</td><td>4.30</td><td>1.38</td><td rowspan="3">69.8</td><td rowspan="3">< .001</td></tr><tr><td>Feature-XAI</td><td>418</td><td>5.08</td><td>1.19</td></tr><tr><td>User-XAI</td><td>415</td><td>5.31</td><td>1.12</td></tr></tbody></table><figcaption>Table 2. Descriptive Statistics and ANOVA Results for Key Variables across Experimental Conditions.</figcaption></figure><p></p><figure class="article-figure"><img src="https://smnxsewcdnayrztrrghn.supabase.co/storage/v1/object/public/journal-assets/scholarly/opening-the-black-box-the-causal-effect-of-explainable-ai-xai-on-consumer-trust-and-purchase-intenti-d9mzk/figure-1-1778090534216.png" alt="bar chart of mean trust scores by experimental condition (Control, Feature-XAI, User-XAI)" loading="lazy" style="max-width:100%;height:auto;"><figcaption>Figure 1. bar chart of mean trust scores by experimental condition (Control, Feature-XAI, User-XAI)</figcaption></figure><p></p><h3>4.3 Hypothesis Testing: Structural Equation Modeling (SEM)</h3><p>To test the proposed mediation framework (H2, H5, H6), we constructed a structural equation model. The experimental condition was dummy coded (1 = XAI present, 0 = Control) and entered as the exogenous variable. The model fit was excellent: χ²(1, N=1248) = 2.15, p = .142; CFI = .999; TLI = .995; RMSEA = .029; SRMR = .011. The standardized path coefficients are presented in Table 3 and visualized in Figure 2.</p><p></p><figure class="article-figure"><img src="https://smnxsewcdnayrztrrghn.supabase.co/storage/v1/object/public/journal-assets/scholarly/opening-the-black-box-the-causal-effect-of-explainable-ai-xai-on-consumer-trust-and-purchase-intenti-d9mzk/figure-2-1778090542320.png" alt="path diagram of the structural equation model showing relationships between XAI condition, transparency, trust, and purchase intent" loading="lazy" style="max-width:100%;height:auto;"><figcaption>Figure 2. path diagram of the structural equation model showing relationships between XAI condition, transparency, trust, and purchase intent</figcaption></figure><p></p><p>As predicted, the presence of an XAI explanation (XAI Presence) had a strong, positive effect on Perceived Algorithmic Transparency (PAT) (β = .48, p < .001). In turn, PAT had a strong positive effect on Consumer Trust (TRUST) (β = .61, p < .001), providing support for <strong>H2</strong>. Finally, TRUST was a significant predictor of Purchase Intent (PI) (β = .47, p < .001), supporting <strong>H5</strong>. Notably, the direct path from PAT to PI was not significant (β = .05, p = .180), suggesting that transparency's effect on purchase intent is fully channeled through trust.</p><p>Most importantly, we tested the mediation hypothesis (H6) using bootstrapping. The indirect effect of XAI Presence on Purchase Intent via the path through PAT and TRUST was significant and substantial (Indirect Effect = β = 0.14, 95% CI = [0.11, 0.17], p < .001). We also tested the simpler mediation proposed in H6 (XAI -> Trust -> PI). The indirect effect here was also strong and significant (Indirect Effect = β = 0.29, 95% CI = [0.24, 0.34], p < .001). These results robustly support H6 and the overall theoretical model: XAI fosters transparency, which builds trust, and this trust is the crucial ingredient that positively influences a consumer's intention to purchase.</p><figure class="table-figure"><table><thead><tr><th>Path</th><th>Standardized Coefficient (β)</th><th>Standard Error</th><th>p-value</th><th>Hypothesis</th><th>Result</th></tr></thead><tbody><tr><td>XAI Presence → Perceived Transparency</td><td>0.48</td><td>0.03</td><td>< .001</td><td>H1</td><td>Supported</td></tr><tr><td>Perceived Transparency → Trust</td><td>0.61</td><td>0.02</td><td>< .001</td><td>H2</td><td>Supported</td></tr><tr><td>XAI Presence → Trust (Direct)</td><td>0.24</td><td>0.04</td><td>< .001</td><td>-</td><td>Significant</td></tr><tr><td>Trust → Purchase Intent</td><td>0.47</td><td>0.03</td><td>< .001</td><td>H5</td><td>Supported</td></tr><tr><td>Perceived Transparency → Purchase Intent</td><td>0.05</td><td>0.04</td><td>.180</td><td>-</td><td>Not Supported</td></tr><tr><td colspan="6"><strong>Indirect Effects (Bootstrapping, 5000 samples)</strong></td></tr><tr><td>XAI → PAT → TRUST → PI</td><td>0.14</td><td>0.02</td><td>< .001</td><td>Full Model</td><td>Supported</td></tr><tr><td>XAI → TRUST → PI</td><td>0.29</td><td>0.03</td><td>< .001</td><td>H6</td><td>Supported</td></tr></tbody></table><figcaption>Table 3. Structural Equation Modeling Path Coefficients and Mediation Analysis.</figcaption></figure>
<h2>5. Discussion</h2>
<h3>5.1 Summary and Interpretation of Findings</h3><p>This study set out to investigate the causal impact of Explainable AI (XAI) on consumer trust and purchase intent in an e-commerce context. Our findings provide compelling empirical evidence that 'opening the black box' of AI-driven recommendations is a highly effective strategy for enhancing consumer perceptions and driving favorable business outcomes. The results can be summarized in three key points.</p><p>First, the presence of an explanation for an AI recommendation—regardless of its specific style—dramatically increases perceived algorithmic transparency and, consequently, consumer trust. Participants who were told *why* a product was recommended reported significantly higher trust in the system compared to those who received an unexplained recommendation. This directly supports our core thesis and aligns with the foundational principles of trustworthy AI, which posit that transparency is a prerequisite for trust (Thiebes et al., 2020; Unknown, 2026). The effect sizes were substantial, suggesting that even a simple, concise explanation can have a powerful psychological impact.</p><p>Second, the style of explanation matters. While both feature-based and user-based explanations were effective, our results showed that user-based explanations, which leverage social proof, generated a significantly higher level of trust. This finding is particularly insightful. It suggests that in the uncertain environment of online shopping, consumers find more assurance in the collective wisdom of their peers ('other people bought this') than in a system's analytical claims about product attributes ('the AI thinks you'll like this feature'). This resonates with classic social psychology theories on conformity and persuasion and highlights the importance of human-centric design in AI explanations (Virvou, 2023).</p><p>Third, our structural equation model elucidates the precise mechanism through which XAI operates. The results robustly confirmed that consumer trust is the critical mediator between XAI and purchase intent. Providing an explanation boosts transparency, which builds trust, and it is this earned trust that translates into a greater willingness to consider purchasing the recommended products. The non-significant direct path from transparency to purchase intent underscores that transparency is not an end in itself; its value lies in its ability to serve as a foundation for a trusting consumer-platform relationship. This finding empirically substantiates the long-held belief in marketing that trust is the ultimate currency in consumer relationships (Helmi et al., 2024; Unknown, 2022) and extends it to the domain of human-AI interaction.</p><h3>5.2 Theoretical Implications</h3><p>This research makes several important contributions to the academic literature. First, it extends theories like the Technology Acceptance Model (TAM) by introducing perceived transparency as a key antecedent to perceived usefulness and trust in the context of AI-driven systems. Our work provides a more nuanced understanding of how system characteristics influence user attitudes. Second, we contribute to the nascent but rapidly growing literature on consumer-facing XAI (Zhang & Chen, 2020). While much XAI research is technical and focused on model-centric interpretability (Singh et al., 2025; Paula, 2023), our study provides crucial user-centric, empirical evidence on the behavioral consequences of explainability, responding to calls for such work (Dwivedi et al., 2020).</p><p>Furthermore, this study validates the conceptual framework of trustworthy AI (Thiebes et al., 2020) in a practical e-commerce setting. We empirically demonstrate that the principles of transparency and interpretability have a direct and measurable impact on the trust construct. Finally, our comparison of explanation styles contributes a new dimension to the study of recommendation systems. We show that the framing of an explanation (analytical vs. social) can differentially impact its effectiveness, adding a layer of psychological depth to what has often been treated as a purely informational problem.</p><h3>5.3 Managerial Implications</h3><p>The practical takeaways for e-commerce managers and digital marketers are direct and actionable. First and foremost, investing in XAI functionality for recommendation systems is not just an ethical 'nice-to-have' but carries a clear return on investment. The observed increases in trust and purchase intent provide a strong business case for allocating development resources to implement explanation features. Given that simple, text-based explanations were effective in our experiment, the barrier to entry may be lower than many firms assume.</p><p>Second, the design of explanations should be a strategic consideration. Our finding that user-based (social proof) explanations outperformed feature-based ones suggests that managers should prioritize and A/B test different explanation styles. Leveraging the power of the collective ('Customers also bought...') can be a simple yet highly potent way to build confidence and reduce purchase anxiety. This reinforces the idea that the best AI experiences are often those that feel transparently human.</p><p>Third, this research offers a pathway to building more resilient and long-term customer relationships. In an era of increasing consumer skepticism about data use and AI (Shahriar et al., 2023), proactively demonstrating transparency can be a powerful brand differentiator. By explaining their reasoning, firms can shift the perception of their AI from a covert persuader to a trusted advisor, fostering a level of brand loyalty that goes beyond the transient benefits of a single personalized recommendation (Sipos, 2025; Unknown, 2024).</p><h3>5.4 Limitations and Directions for Future Research</h3><p>Despite its robust findings, this study has several limitations that open avenues for future research. First, our use of an online experiment with hypothetical purchase scenarios enhances internal validity but may limit external validity. Future research should seek to replicate these findings using field experiments on live e-commerce platforms to measure effects on actual click-through rates and conversion. Second, we tested two common but simple explanation types. Future studies could explore more complex, interactive, or personalized explanations. For instance, would allowing users to click on an explanation to get more detail further enhance trust? Or could explanations themselves be personalized based on a user's technical literacy?</p><p>Third, our study focused on a single product category (backpacks), a mid-involvement good. The impact of XAI might differ for high-involvement, high-risk products (e.g., financial services, electronics) versus low-involvement, habitual purchases (e.g., groceries). Investigating these boundary conditions is a logical next step. Furthermore, future work could explore the 'dark side' of explainability, such as the potential for poor or nonsensical explanations to harm trust more than no explanation at all, or the ethics of using explanations to justify biased or unfair recommendations.</p><p>Finally, the growing ecosystem of AI in commerce presents further research opportunities. How does XAI impact trust when applied to AI influencers (Liu, 2026), AI-generated ad copy (Unknown, 2025), or pricing decisions? Understanding the role of explainability across the entire AI-driven marketing landscape will be crucial as these technologies continue to evolve (Wang, 2026).</p>
<h2>6. Conclusion</h2>
<p>As artificial intelligence becomes ever more woven into the tapestry of our digital lives, ensuring that these systems are not only intelligent but also intelligible is paramount. This research sought to move beyond the theoretical discourse on trustworthy AI by providing concrete, causal evidence of the benefits of explainability in a real-world business context. Our findings are unequivocal: opening the 'black box' of AI recommendations through simple, clear explanations significantly enhances consumer trust and, through that trust, positively influences purchase intentions.</p><p>The study contributes to business research and practice by demonstrating that algorithmic transparency is not a cost center but a value driver. We have shown that the mechanism is psychological: explanations foster a perception of transparency, which nourishes trust, the essential lubricant of commerce. Moreover, we have provided practical guidance on this process, revealing that explanations grounded in social proof can be particularly potent. For managers and developers, the message is clear: the path to more effective AI is not just through more complex algorithms, but through more considerate and transparent human-computer interaction (Virvou, 2023).</p><p>In conclusion, the challenge for the next generation of e-commerce is not simply to deploy more AI, but to deploy AI that consumers can understand and trust. By embracing explainability, firms can build more sustainable, ethical, and ultimately more profitable relationships with their customers. This research represents a significant step in quantifying the value of that transparency, paving the way for a future where technology serves not only to personalize our experiences, but also to empower our choices.</p>
<h2>References</h2>
<ol class="references">
<li>Unknown (2025). Synthesizing Desire: An Investigation into Consumer Trust and Purchase Intent Towards AI-Generated Product Imaginary and Ad Copy. <em>European Economic Letters</em>. https://doi.org/10.52783/eel.v15i4.4070</li>
<li>Jiaming Liu, J. L. (2026). The Influence of AI Influencer Characteristics in E-commerce on Consumer Trust and Purchase Intention. <em>Korea International Trade Research Institute</em>, <em>22</em>(1), 1-19. https://doi.org/10.16980/jitc.22.1.202602.1</li>
<li>Melekar, M. A., Dalvi, M. A., Ambavade, M. U. (2025). Impact of AI-Generated Influencers on Consumer Trust and Purchase Intent. <em>IARJSET</em>, <em>12</em>(12). https://doi.org/10.17148/iarjset.2025.121231</li>
<li>Madabhushini, I. (2025). Explainable AI (XAI) in Business Intelligence: Enhancing Trust and Transparency in Enterprise Analytics. <em>The American Journal of Engineering and Technology</em>, <em>7</em>(08), 9-20. https://doi.org/10.37547/tajet/volume07issue08-02</li>
<li>Sipos, D. (2025). The Effects of AI-Powered Personalization on Consumer Trust, Satisfaction, and Purchase Intent. <em>European Journal of Applied Science, Engineering and Technology</em>, <em>3</em>(2), 14-24. https://doi.org/10.59324/ejaset.2025.3(2).02</li>
<li>Arya, A., Goel, P., Verma, S. K., Jain, K. (2025). AI’s impact on consumer purchase intent through influencer marketing – A determinant of purchase behaviour in e-commerce. <em>Multidisciplinary Reviews</em>, <em>8</em>(11), 2025349. https://doi.org/10.31893/multirev.2025349</li>
<li>Israfilzade, K. (2025). AI-generated versus human-created advertising: Effects on consumer trust and purchase intent. <em>Equilibrium. Quarterly Journal of Economics and Economic Policy</em>, <em>20</em>(4), 1301-1337. https://doi.org/10.24136/eq.4038</li>
<li>P, S. (2026). AI-Driven Greenwashing: Effects on Consumer Trust and Purchase Intention. <em>International Journal of Science and Research (IJSR)</em>, 624-626. https://doi.org/10.21275/sr26406143445</li>
<li>Y, S. (2026). AI-Driven Personalization in Sustainable Marketing: Effects on Purchase Intent Among Generation Z Consumers in Indian E-Commerce. <em>International Journal of Research Publication and Reviews</em>, <em>7</em>(4), 1114-1122. https://doi.org/10.55248/gengpi.07.0426.10807</li>
<li>Helmi, M., Alharthi, S., Habib, S. (2024). Online Trust Determinants, Consumer Perception, and Purchase Intent in Saudi E-Commerce: Exploring Determinants and Evidence. <em>Scientific Journal of King Faisal University: Humanities and Management Sciences</em>, 100-107. https://doi.org/10.37575/h/mng/240001</li>
<li>Paula, M. d. O. P. (2023). Explainable AI (XAI) na Detecção de Deepfakes: Transparência e Interpretação em Modelos de Visão Computacional. <em>RCMOS - Revista Científica Multidisciplinar O Saber</em>, <em>1</em>(1). https://doi.org/10.51473/rcmos.v1i1.2023.1867</li>
<li>Abhilash Nautiyal, D. (2026). IMPACT OF AI POWERED PERSONALIZATION ON CONSUMER PURCHASE DECISIONS IN E COMMERCE PLATFORMS. <em>JOURNAL OF ADVANCE AND FUTURE RESEARCH</em>, <em>4</em>(1). https://doi.org/10.56975/jaafr.v4i1.502588</li>
<li>Singh, H. K., Kaur, D. S. (2026). Factors Influencing AI-Driven Personalization on Consumer Trust and Purchase Intent among Gen-z. <em>INTERNATIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT</em>, <em>10</em>(04), 1-9. https://doi.org/10.55041/ijsrem59689</li>
<li>Unknown (2022). The Impact of Risk-aware Consumer Trust on CB E-commerce Platforms on Purchase Intention. <em>Journal of Global Information Management</em>, <em>30</em>(3), 0-0. https://doi.org/10.4018/jgim.20220801oa03</li>
<li>Wen, J., Li, X. (2025). AI Digital Human Responsiveness and Consumer Purchase Intention: The Mediating Role of Trust. <em>Journal of Theoretical and Applied Electronic Commerce Research</em>, <em>20</em>(3), 246. https://doi.org/10.3390/jtaer20030246</li>
<li>Unknown (2026). Enhancing Transparency and Trust in Artificial Intelligence Systems Using Explainable AI (XAI) Techniques. <em>International Journal of Innovative Research in Technology</em>, <em>12</em>(7). https://doi.org/10.64643/ijirtv12i7-191723-459</li>
<li>Unknown (2024). "Exploring the Role of Personalization in E-commerce: Impacts on Consumer Trust and Purchase Intentions". <em>European Economic Letters</em>. https://doi.org/10.52783/eel.v14i3.1845</li>
<li>Sweeney, J. C., Smith, L., Naveed, Q. N. H. (2026). The Impact of AI Chatbot Personalization on Consumer Purchase Intent. <em>International Journal of Engineering and Computational Applications</em>, <em>2</em>(1), 10-12. https://doi.org/10.54660/.ijeca.2026.2.1.10-12</li>
<li>El Shaddai Sandhy Pustap (2026). AI personalization in marketing management: Consumer trust and purchase behavior in social commerce. <em>International Journal of Economic and Business Research</em>, <em>1</em>(2), 20-27. https://doi.org/10.65310/rsyee128</li>
<li>Singh, S., Ravindran, V. K., Patil, S. (2025). Explainable AI (XAI) for Cloud Resource Forecasting in E-Commerce Environments. <em>International Journal For Multidisciplinary Research</em>, <em>7</em>(4). https://doi.org/10.36948/ijfmr.2025.v07i04.51796</li>
<li>Giriraj Agarwal (2025). Explainable AI (XAI) for Cyber Defense: Enhancing Transparency and Trust in AI-Driven Security Solutions. <em>International Journal of Advanced Research in Science, Communication and Technology</em>, 132-138. https://doi.org/10.48175/ijarsct-23624</li>
<li>Dwivedi, Y. K., Ismagilova, E., Hughes, D. L., Carlson, J., Filieri, R., Jacobson, J. (2020). Setting the future of digital and social media marketing research: Perspectives and research propositions. <em>International Journal of Information Management</em>, <em>59</em>, 102168-102168. https://doi.org/10.1016/j.ijinfomgt.2020.102168</li>
<li>Zhang, Y., Chen, X. (2020). Explainable Recommendation: A Survey and New Perspectives. <em>Foundations and Trends® in Information Retrieval</em>, <em>14</em>(1), 1-101. https://doi.org/10.1561/1500000066</li>
<li>Thiebes, S., Lins, S., Sunyaev, A. (2020). Trustworthy artificial intelligence. <em>Electronic Markets</em>, <em>31</em>(2), 447-464. https://doi.org/10.1007/s12525-020-00441-4</li>
<li>Khrais, L. T. (2020). Role of Artificial Intelligence in Shaping Consumer Demand in E-Commerce. <em>Future Internet</em>, <em>12</em>(12), 226-226. https://doi.org/10.3390/fi12120226</li>
<li>Wang, C. L. (2026). Editorial: The changing landscape of marketing research in the AI era: prospects and challenges. <em>Journal of Research in Interactive Marketing</em>, <em>20</em>(1), 1-10. https://doi.org/10.1108/jrim-02-2026-766</li>
<li>Bello, O. A., Olufemi, K. (2024). Artificial intelligence in fraud prevention: Exploring techniques and applications challenges and opportunities. <em>Computer Science & IT Research Journal</em>, <em>5</em>(6), 1505-1520. https://doi.org/10.51594/csitrj.v5i6.1252</li>
<li>Shahriar, S., Allana, S., Hazratifard, S. M., Dara, R. (2023). A Survey of Privacy Risks and Mitigation Strategies in the Artificial Intelligence Life Cycle. <em>IEEE Access</em>, <em>11</em>, 61829-61854. https://doi.org/10.1109/access.2023.3287195</li>
<li>Helberger, N., Huh, J., Milne, G. R., Strycharz, J., Sundaram, H. (2020). Macro and Exogenous Factors in Computational Advertising: Key Issues and New Research Directions. <em>Journal of Advertising</em>, <em>49</em>(4), 377-393. https://doi.org/10.1080/00913367.2020.1811179</li>
<li>Virvou, M. (2023). Artificial Intelligence and User Experience in reciprocity: Contributions and state of the art. <em>Intelligent Decision Technologies</em>, <em>17</em>(1), 73-125. https://doi.org/10.3233/idt-230092</li>
</ol>
</article>