📄 social_science

From Anthropomorphism to Purchase Intention in AI Shopping Assistant Contexts: A Three-Wave Longitudinal Moderated Mediation Study of Perceived Intelligence and General Self-Efficacy Among College Students

This three-wave longitudinal study of Chinese college students reveals that perceived anthropomorphism of AI shopping assistants prospectively increases purchase intention through enhanced perceived intelligence, with this indirect effect being significantly stronger for individuals possessing higher general self-efficacy.

Original authors: Siyi Wu, Yang Yang, Xiangyu Tan, Jiale Wang

Published 2026-08-07

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Siyi Wu, Yang Yang, Xiangyu Tan, Jiale Wang

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you're walking through a giant, endless digital mall. Instead of a human salesperson, you're guided by a robot friend. This robot isn't just a calculator; it talks, has a name, and maybe even a cute avatar. This is the world of AI shopping assistants. But here's the big question: does making a robot look and act more like a human actually make you want to buy things?

To understand this, we need to know a few simple ideas. First, anthropomorphism is just a fancy word for seeing human traits in non-human things, like thinking a car is "angry" or a robot is "friendly." Second, perceived intelligence is simply how smart you think that robot is. If a robot seems human, do you automatically think it's also super-smart? Third, self-efficacy is your own inner confidence. It's that feeling of, "I can handle this situation," whether it's a tough math test or a tricky shopping decision. Finally, purchase intention is just how likely you are to actually buy something. Researchers care about this because if we can figure out what makes people trust and buy from AI, we can build better tools for everyone, from students to grandmas.

The Robot Shopper Experiment

A team of researchers decided to play detective with 1,416 college students to see how these ideas connect over time. They didn't just ask everyone one question and call it a day; they ran a "three-wave" study. Think of it like checking the weather on three different days to see if a storm is coming, rather than just looking at the sky once.

The Setup:

Time 1 (T1): They asked students how human-like they found their AI shopping assistants (Anthropomorphism) and how confident they felt in general (Self-Efficacy). They also checked what they already planned to buy.
Time 2 (A couple of months later): They asked the students how smart they thought the AI was (Perceived Intelligence) and what they planned to buy now.
Time 3 (A couple more months later): They checked the final shopping plans again.

The Big Discovery:
The study found a clear chain reaction, like a line of dominoes falling.

The Human Look Helps: When students saw the AI as more human-like at the start, they were more likely to think it was smart a few months later. It's like if a robot friend smiles and uses your name, you start to believe it's actually clever enough to help you.
Smartness Leads to Buying: Once the students thought the AI was smart, they were much more likely to say, "Yes, I'll buy what this robot suggests."
The Confidence Booster: Here is the twist. This chain reaction worked best for students who already had high general self-efficacy (high confidence). For students who felt very capable in life, seeing a smart AI made them want to buy even more. For students with lower confidence, the link between "smart AI" and "buying" was still there, but it was weaker.

The Numbers:
The researchers found that the "human-like" feeling at the start predicted how smart the AI seemed later with a statistical strength of 0.319. Then, that feeling of "smartness" predicted the final desire to buy with a strength of 0.344. The "confidence" factor (self-efficacy) made the connection between "smart AI" and "buying" even stronger, with an interaction effect of 0.123. The overall "moderated mediation index" (a fancy way of saying the whole chain reaction) was 0.075, which was statistically significant.

What This Means (and What It Doesn't):
The paper suggests that making an AI look human is a good strategy, but only because it helps convince us the AI is actually smart. It's not the "human-ness" itself that makes us buy; it's the "human-ness" that tricks our brains into thinking, "Wow, this thing is brilliant, I should listen to it."

However, the authors are careful to say this is a suggestion based on observation, not a proven law of physics. They didn't force the robots to change; they just watched what happened naturally. Also, they measured what students intended to buy, not what they actually spent money on. So, while the results are strong and the timeline is solid, it's a map of intentions, not a guarantee of sales.

The Takeaway:
If you are building an AI shopping assistant, don't just slap a human face on it and hope for the best. The study suggests that the human face works because it builds a reputation for intelligence. And if you are a shopper, your own confidence plays a huge role: the more confident you feel in your own abilities, the more likely you are to trust a smart-sounding robot and follow its advice. But remember, this is just a snapshot of college students' minds; the real world might have more surprises!

Technical Summary: From Anthropomorphism to Purchase Intention in AI Shopping Assistant Contexts

Problem Statement
While AI shopping assistants are increasingly utilized as conversational decision aids in online retail, the psychological mechanisms linking their humanlike design features to consumer purchase intention remain under-explored, particularly regarding temporal dynamics. Existing research often relies on cross-sectional data, which may produce biased estimates of mediation processes that unfold over time. Furthermore, it is unclear whether the perception of an AI's "humanlikeness" (anthropomorphism) directly drives purchase behavior or if it operates through intermediate cognitive evaluations, such as perceived intelligence. Additionally, the role of individual psychological resources, specifically general self-efficacy, as a boundary condition for these relationships requires empirical investigation. This study addresses these gaps by examining a temporally ordered moderated mediation model among college students.

Methodology
The study employed a three-wave longitudinal survey design to separate the predictor, mediator, and outcome variables temporally.

Participants: The final analytic sample consisted of 1,416 Chinese college students (aged 18+) recruited via cluster sampling from eight universities across Central, North, and South China.
Procedure: Data were collected online over three waves:
- T1 (Dec 2025): Measured perceived anthropomorphism, general self-efficacy, and baseline purchase intention.
- T2 (Feb 2026): Measured perceived intelligence and purchase intention.
- T3 (Apr 2026): Measured purchase intention as the final outcome.
Measures:
- Perceived Anthropomorphism: Adapted from the Godspeed Anthropomorphism Scale (5 items).
- Perceived Intelligence: Adapted from the Godspeed Perceived Intelligence Scale (5 items).
- Purchase Intention: Adapted from classic willingness-to-buy measures (3 items).
- General Self-Efficacy: Measured using the 10-item General Self-Efficacy Scale.
Analysis: Data analysis utilized SPSS 26.0 and Hayes' PROCESS macro (Model 14) with 5,000 bootstrap samples. The study conducted confirmatory factor analysis (CFA), reliability testing (Cronbach's alpha, CR, AVE), discriminant validity checks (HTMT), and common method bias diagnostics. The primary analysis tested a longitudinal moderated mediation model controlling for baseline purchase intention and demographic covariates.

Key Results

Model Fit: The four-factor model (anthropomorphism, self-efficacy, intelligence, purchase intention) demonstrated good fit ( $\chi^2/df = 2.440$ , CFI = 0.962, RMSEA = 0.041).
Direct and Indirect Effects:
- T1 perceived anthropomorphism positively predicted T3 purchase intention ( $\beta = .187, p < .001$ ), supporting the direct association (H1).
- T1 perceived anthropomorphism positively predicted T2 perceived intelligence ( $\beta = .319, p < .001$ ).
- T2 perceived intelligence positively predicted T3 purchase intention ( $\beta = .344, p < .001$ ).
- Perceived intelligence statistically mediated the longitudinal association between anthropomorphism and purchase intention (H2).
Moderation: T1 general self-efficacy significantly moderated the relationship between T2 perceived intelligence and T3 purchase intention ( $\beta = .123, p < .001$ ). The positive association between perceived intelligence and purchase intention was stronger for students with higher general self-efficacy (H3).
Conditional Indirect Effect: The index of moderated mediation was significant (Index = .075, 95% CI [.045, .106]). The indirect effect of anthropomorphism on purchase intention via perceived intelligence increased as general self-efficacy levels increased (H4).
Robustness: A robustness model controlling for T2 purchase intention confirmed that the conditional process remained significant, though the magnitude of the effects was reduced.

Key Contributions

Contextual: The study specifically focuses on AI shopping assistants rather than general chatbots or broad recommendation systems, isolating the decision-support context.
Conceptual: It integrates anthropomorphism, perceived intelligence, purchase intention, and general self-efficacy into a single conditional process model, clarifying that humanlike cues influence purchase intention primarily through cognitive evaluations of competence.
Methodological: By utilizing a three-wave design with baseline controls, the study provides stronger temporal evidence for mediation and moderation than cross-sectional studies, addressing concerns regarding biased estimates in longitudinal processes.

Significance and Claims
The authors claim that the study provides temporal evidence for a cognitive pathway linking humanlike AI cues to purchase-related intention. The findings suggest that perceived anthropomorphism is not merely a direct persuasion tool; rather, its impact is largely mediated by the user's subsequent judgment that the AI is intelligent and competent. Furthermore, the study identifies general self-efficacy as a critical psychological boundary condition: students with higher self-efficacy are more likely to translate their perception of an AI's intelligence into actual purchase intention.

The paper maintains a modest stance regarding causality, explicitly stating that while the design offers stronger temporal inference than cross-sectional data, causal claims should remain cautious. The results are interpreted as longitudinal associations rather than definitive causal mechanisms. The authors emphasize that these findings help explain why anthropomorphic design may yield inconsistent effects across contexts, suggesting that humanlike features must be coupled with perceived competence and supported by the user's own confidence in decision-making to effectively drive purchase intention.

The Robot Shopper Experiment

Technical Summary: From Anthropomorphism to Purchase Intention in AI Shopping Assistant Contexts

More like this