An AI shopping assistant can make a persuasive recommendation even when it has misunderstood what a shopper wants. The interface needs to give consumers a way to spot that mismatch and correct it. It should also help them distinguish AI-generated guidance from information the seller provides.
Consumer product designer Zhengyang Yang focuses on helping shoppers understand AI recommendations before they act on them. He helped develop the design language and key visual patterns for TikTok’s first AI shopping assistant across search and chatbot experiences. Yang’s work focused on helping consumers understand recommendations and decide how to act on them. In conventional online shopping, people communicate their preferences through search terms and filters. AI assistants let them explain what they want in their own words, then interpret those intentions to suggest products. That gives UX designers greater responsibility for how the interaction unfolds.
Yang argues that shoppers need opportunities to refine their preferences and consider alternatives throughout the conversation.
His work also includes short-video commerce and store experiences on TikTok Shop. On its store-tiering and Official Shop projects, he worked on making seller classifications understandable and communicating relevant consumer rights. Yang says that a completed transaction reveals less about the quality of a recommendation than whether the purchase meets the shopper’s expectations.
An earlier background in architecture shaped Yang’s approach to design, particularly how people navigate complex environments. At BioMatrix, a consumer platform combining biometric identity with AI and Web3 technologies, he applied that perspective to products whose underlying systems many users did not understand. Yang also co-authored research presented at CSCW 2025 examining Black American users’ experiences with AI-supported writing tools.
In an exclusive Q&A, he explains how those experiences inform a design philosophy that gives consumers more context for their decisions, and why products earn trust through interactions people can understand.
Yang speaks solely in a personal capacity. The views he shares in this interview are his own and do not represent those of his current or former employers. Responses have been lightly edited for length and clarity.
You helped design TikTok’s first AI shopping assistant across search and conversational interfaces. How does AI change a UX designer’s role, and how do you help consumers make informed decisions while keeping them in charge?
AI shopping assistants should provide context rather than control. Traditional e-commerce relies heavily on users comparing products as they navigate categories and apply filters to communicate what they are looking for. Instead of requiring users to translate their intentions into predefined search terms or filters, an AI shopping assistant can allow them to express preferences directly, including their needs, priorities, budgets, and personal circumstances. By interpreting this information and asking clarifying questions when necessary, the system can develop a more accurate understanding of shopping intent and make recommendations that are better aligned with what the user actually wants.
This changes the role of UX designers because we are no longer simply organizing information and optimizing navigation. We are designing how users communicate their intentions, how AI interprets those intentions, and also shaping how the system presents recommendations to support decision-making. When I worked on the design language and key visual patterns for TikTok’s first AI shopping assistant across search and conversational experiences, the broader challenge was translating AI capabilities into an experience that felt intuitive and actionable.
However, an AI system can identify patterns and suggest products, but users still bring personal priorities that the available information may not fully capture. Good UX should therefore make recommendations understandable and allow users to adjust their preferences. The objective for AI should be to help consumers express their needs more clearly and make informed decisions with less effort.
How do you help shoppers evaluate and question a convincing AI recommendation when its underlying information may be incomplete? When should the system step back, and can shopping become too streamlined?
One of the biggest challenges in designing AI shopping experiences is that users may not always distinguish between information that AI generates and information that comes directly from another source, such as a seller or the platform. As AI enters search results, recommendations, product descriptions, and conversational interfaces, I believe products need a consistent and recognizable AI design language. This includes visual patterns and interaction conventions that help users identify when they are interacting with AI or viewing AI-generated content. Clear indicators should reinforce that recognition.
Recognizability is only the first step. Users also need sufficient information to evaluate a recommendation and meaningful opportunities to question or refine it. For example, if an assistant recommends a product based on price and popularity, but the user actually prioritizes durability or a particular technical requirement, the interface should make it easy to introduce that preference and adjust the recommendation. Designers need to consider which information is essential at each stage and when an explanation is necessary. They also need to consider when additional controls would be useful. The goal is to provide transparency without overwhelming users with implementation details that do not contribute to their decisions.
Removing unnecessary steps improves efficiency, but removing every opportunity for reflection can encourage users to accept recommendations without evaluating them. For low-risk purchases, a streamlined experience may be appropriate. For more consequential decisions, users may benefit from additional comparisons and explanations. They may also need opportunities to confirm their priorities. A consistent AI design language and the ability to exercise independent judgment are both essential to achieving that balance.
You also worked on TikTok Shop’s store-tiering and Official Shop experiences. How do you make a trust badge meaningful, and give shoppers useful information about sellers and their rights?
One challenge with introducing concepts such as store tiers or Official Shop status is that designers cannot assume every consumer immediately understands what those concepts mean. When I worked on TikTok Shop’s store-tiering and Official Shop experiences, an important design consideration was how to communicate seller identity and authenticity in a way that consumers could interpret. We also needed to make relevant user rights understandable. Simply displaying a new badge or label is not enough if users do not understand what it represents or why it should influence their purchasing decisions.
For unfamiliar concepts, I believe the interface should provide easy access to explanations at the point where users encounter them. Contextual information or expandable descriptions could help. Clear entry points could also lead to more detailed explanations of seller classifications and associated protections. The information should be accessible without requiring users to leave the shopping journey or search through a help center. At the same time, designers need to balance explanation with efficiency, presenting the most relevant information first while allowing users to explore further when necessary.
As consumers encounter a new concept repeatedly and become more familiar with its meaning, the platform can gradually expand its visibility across different shopping touchpoints. However, research and evidence of user understanding should inform this progression rather than assumptions about familiarity. I see this as a process of developing consumer mental models over time. We first help users understand a concept and reinforce that understanding through consistent exposure. As that understanding develops, we can make the concept a more integrated part of the shopping experience.
Your AI shopping work was associated with improved commerce metrics, but higher conversion does not necessarily mean better decisions. How do you evaluate the impact, particularly when business incentives and consumer interests diverge?
In my work on AI shopping and commerce experiences at TikTok, quantitative performance has been an important part of evaluating design outcomes. However, an increase in conversion does not necessarily mean that users have made better decisions, and higher engagement does not always indicate a more satisfying experience.
Short-term indicators can measure adoption and task completion. Conversion is another short-term indicator, while longer-term indicators such as retention, repeat purchasing, customer satisfaction, returns, and support-related issues can help reveal whether the experience is sustainable. Qualitative research is equally important because behavioral data alone may not explain whether users understand AI recommendations and recognize their limitations. It may also leave us uncertain about whether users feel confident in the decisions they make.
When business metrics and consumer interests appear to conflict, I think designers have a responsibility to make those trade-offs visible rather than treating conversion as the only measure of success. An AI assistant might encourage faster purchasing decisions, but if those decisions subsequently lead to dissatisfaction or frequent returns, the initial performance gains may not translate into sustainable growth. Declining trust can have the same effect. The most successful AI commerce experiences will increase transactions while helping establish lasting relationships in which consumers continue to find the system useful and reliable enough to return to.
Your CSCW 2025 research examined Black users’ experiences with AI writing tools. What most changed your perspective, and what does that research suggest about personalization and standardized experiences in AI commerce?
Our paper, “Should AI Mimic People? Understanding AI-Supported Writing Technology Among Black Users,” received the CSCW 2025 Special Recognition for Contribution to Diversity, Equity & Inclusion. Through interviews and observational studies with 13 Black American users, we found that AI-supported writing tools sometimes failed to recognize common names and expressions in African American Vernacular English (AAVE), treating intentional language choices as mistakes. For some participants, these corrections were more than minor inconveniences. They created feelings of exclusion and raised concerns about whether the technology adequately represented their culture. At the same time, participants recognized the practical benefits of AI for professional communication.
While participants wanted AI systems to recognize their language and communication styles, some felt uncomfortable when the technology attempted to reproduce AAVE inaccurately or in ways that seemed stereotypical. Others saw potential value in systems capable of communicating more naturally across different linguistic contexts. This revealed that personalization is not simply about making AI behave more like the user. It requires understanding the user’s expectations and respecting their choices. It also requires recognizing that different people may want different levels of adaptation.
Shopping preferences and cultural expectations can vary significantly across different users and contexts. So can perceptions of credibility. An AI shopping assistant should therefore avoid assuming that a standardized recommendation or interaction pattern will work equally well for everyone. Instead, it should give consumers opportunities to express their preferences and correct the system’s assumptions. We can maintain consistent design principles while recognizing that a consistent experience does not always require an identical experience. My research reinforced my belief that teams should integrate inclusive design from the beginning of AI product research and development. It should also guide evaluation, rather than serve as a final localization or accessibility exercise.
You moved from architecture to BioMatrix before joining TikTok. Which lessons carried over, and how do you make sophisticated technology approachable without hiding risks or decisions users need to understand?
My background in architecture has had a lasting influence on how I approach product design because architecture teaches you to understand experiences as interconnected systems rather than isolated visual elements. When designing a building, you consider how people enter, navigate, interpret spaces, and move between different activities. Those decisions can make an environment feel intuitive and efficient or leave people confused. I apply a similar perspective to digital products by looking at how users navigate information and move between interactions. I also consider how their decisions unfold over time.
At A&C Tech, I worked on BioMatrix, a consumer platform that combined AI and Web3 technologies with biometric identity. The challenge was that many users were unfamiliar with the technical concepts underlying the product, including biometric verification and blockchain-based digital assets. My work involved designing consumer experiences that translated technical capabilities into understandable interactions, including onboarding and digital content. I also worked on participation mechanisms. That experience reinforced the importance of progressive disclosure. This means helping users understand what they need at each stage while making additional information available as their familiarity and engagement develop.
I believe there is an important distinction between simplifying complexity and concealing information that affects users’ decisions. People do not necessarily need to understand the technical architecture behind a product, but they should understand the implications of using it and the choices they are making. These experiences have shaped my belief that designers can contribute far beyond the interfaces they create. My ambition is to help shape the next generation of intelligent consumer products, bringing together technological innovation and human-centered thinking. I want those products to deliver business impact while creating experiences that genuinely improve people’s everyday lives.
