Consumer-First AI: Designing Consent & Personalization in AI Systems
Written by Yadi Wang
The Challenge
Personalized AI advisors and AI agents are rapidly becoming a reality. These AI systems have the potential to make the consumer marketplace more convenient and efficient – but they risk bringing new threats, especially if consumers lose a sense of choice and control. Therefore, it’s critical to ensure these systems are designed with the consumer’s best interests at their core.
In the summer of 2025, I joined the Innovation Lab at Consumer Reports (CR) as a Siegel Family Endowment PiTech PhD Impact Fellow. I was tasked with designing consent and personalization features for AI systems centered on consumer interests and agency.
The Research Process
To explore the current landscape of consumer-facing AI advisors and AI agents, I analysed 11 AI advisors and 11 AI agents, supplemented by a literature review of 20 academic papers on user preferences regarding consent and personalization in digital systems. The analysis revealed that general-purpose AI advisors draw on a wide range of data sources - including chat memory, manual user configurations, and integrations with external services - to personalize user experiences. In contrast, domain-specific and more agentic AI systems tend to offer much more limited personalization. I also documented the user interface designs these systems use to request consent and deliver personalized experiences, highlighting both effective and problematic examples.
To better understand how consumers perceive these practices in today’s rapidly evolving AI environment, I conducted semi-structured interviews with 10 participants who had used AI in their shopping journey. The findings showed that participants were highly interested in personalized AI shopping systems and were generally open to sharing substantial personal data in exchange for personalized experiences. Notably, most participants reported not actively using privacy settings when interacting with AI, often assuming that their data was already “out there” on the internet. However, they expressed greater caution toward AI agents that take actions on their behalf, and cited concerns about potential errors and the need to remain in control. Many also shared specific preferences for how they would - and would not - like to interact with personalized AI systems.
The Proposed Design
The landscape analysis and user research helped me develop high-level design principles for AI systems that give consumers real agency and the ability to personalize their experience. The 3 core design principles are:
Transparency. Users should always understand what the AI knows about them and how it uses that information.
Contextual Consent. Rather than asking for broad permissions upfront, consent should be collected when and where it becomes relevant and useful to the user.
User in Control. Users should remain the primary decision makers, with the AI system serving their choices rather than making choices for them.
Guided by these high-level design principles and the user interface designs identified in my earlier research, I developed interactive prototypes based on the current version of AskCR - CR’s expert-powered adviser that answers consumer questions based on CR’s trusted data and research. In particular, I prototyped the following interfaces:
Onboarding: New users can go through a 9-step setup process that establishes their preferences for memory storage, shopping integrations, privacy controls, and AI agent configurations.
Conversation: Users can chat with AskCR, while the interface saves and uses relevant information in conversations and integrates AI agent functionality into the chat experience.
Settings: Users can view and manage the AI system’s memory, external integrations, privacy controls, and agent configurations.
Agent Dashboard: Users can directly view and manage specialized AI agents’ actions through an interactive dashboard.
Yadi Wang
Ph.D. Student, Information Science, Cornell University
The Impact and Path Forward
My work has provided the initial insights on how a personalized shopping AI system might engage with consumers in a way that preserves their choice and control Moving forward, more work will be needed to translate these design insights into real products at CR. This will involve creating more comprehensive prototypes, as well as conducting user testing with actual consumers to refine interaction flows.
Additionally, my work also highlighted the need for more digital literacy education for consumers in the era of AI. Looking ahead, CR can play a vital role in developing and disseminating educational resources through its digital platforms and advocacy initiatives, to empower consumers to safely and confidently navigate and control their interactions with AI technologies.