Back to the library

Key question New

AI-processed · Author-approved

How do conventional marketing cues influence product choices and search depth of autonomous AI shopping agents?

When the Buyer Is a Machine: The Diagnosticity–Exploration Tradeoff in Agentic Commerce
Suhyeon Lee, Sang Pil Han, Donghyuk Shin · International Conference on Information Systems (ICIS) · 2026
Illustration for: How do conventional marketing cues influence product choices and search depth of autonomous AI shopping agents?

Loads Spotify's player; Spotify may set cookies. This device remembers your choice.

Key findings

  • All nine tested marketing strategies significantly increased product selection relative to the control baseline (11.5% in the primary model), demonstrating that human-designed cues reliably sway AI purchasing.
  • Third-party and market-validated cues produced the strongest selection effects, led by the Best Seller Badge (70.0%), Quality Certification (61.1%), and Purchase Volume (59.6%), whereas seller-provided Stock Scarcity had the weakest impact (22.5%).
  • Stronger marketing cues systematically caused search compression through a diagnosticity–exploration tradeoff, leading agents to view fewer detail and review pages and stop searching earlier.
  • Marketing cue effects persisted across all display positions and generalized across different underlying language models and product categories, though individual cue magnitudes varied across settings.

What it means for you

CIO / IT Executive

Audit your organization's internal procurement AI tools by Monday morning to ensure they are configured to override market-validated cues (like Best Seller Badges and Quality Certifications) and enforce a minimum search depth, preventing automated agents from falling victim to search compression.

IT Manager

On Monday morning, update the system prompt and tool-use parameters for your automated shopping agents to mandate the retrieval of a fixed minimum number of product reviews and detail pages before any purchase decision can be executed.

Business Strategist

Allocate budget this Monday morning to prioritize third-party validations, quality certifications, and sales volume metrics on your e-commerce platform, as these cues have proven most effective at capturing autonomous AI agent selections.

Researcher

Design a controlled experiment this Monday morning to test whether varying the prompt architecture or temperature of large language models mitigates the diagnosticity-exploration tradeoff observed during agentic search.

Policymaker

Draft a transparency framework by Monday morning requiring e-commerce platforms to expose standardized, machine-readable metadata for product badges and certifications to prevent manipulative third-party cues from unfairly biasing automated agent decisions.

Summary

This study evaluates whether e-commerce marketing cues designed for human consumers influence autonomous AI shopping agents during product search and selection. The authors built a sandboxed e-commerce environment featuring nine marketing strategies and conducted 23,015 experimental trials across four product categories using three large language models. Through a multi-step tool-use architecture, the researchers measured how marketing messages alter both final product choices and search depth.

Problem

As consumers delegate purchasing authority to autonomous AI agents, e-commerce marketing messages originally designed for human cognitive biases must be processed by machine decision-makers. Prior research primarily evaluated AI reactions to persuasion in static, forced-choice settings without sequential browsing. It therefore remained unknown whether marketing cues affect an agent's multi-step information search or induce premature commitment.

Transcript
Host: Welcome to A.I.S. Insights, powered by Living Knowledge. I’m Anna Ivy Summers. Today, we are discussing a study presented at the International Conference on Information Systems 2026 titled "When the Buyer Is a Machine: The Diagnosticity–Exploration Tradeoff in Agentic Commerce." The authors are Suhyeon Lee, Sang Pil Han, and Donghyuk Shin. Alex, what core question does this study address?

Expert: Hello Anna. It examines what happens when purchasing decisions are delegated to autonomous AI shopping agents. Specifically, do marketing cues designed around human psychology still influence software agents, and do those cues alter how agents search before buying?

Host: That is increasingly relevant as industry forecasts project AI agents handling a notable share of e-commerce transactions by 2030. What was missing from earlier research in this area?

Expert: Earlier studies mostly tested language models in static, forced-choice scenarios where all product choices were visible at once. In real shopping, however, purchasing is sequential. A buyer searches, browses product details, reads reviews, and decides when to stop. We did not know whether marketing messages alter an agent's multi-step exploration or trigger premature commitment.

Host: How did the researchers test this dynamic search process?

Expert: They built a sandboxed e-commerce environment featuring 17 HTML pages across four product categories: facial serums, smartwatches, organic milk, and dresses. They surveyed 24 major retail sites to identify nine common marketing cues. Then, they deployed an autonomous shopping agent using the ReAct tool-use architecture, running over 23,000 trials across three language models: OpenAI's o4-mini, gpt-4o-mini, and Google's gemini-3.1-flash-lite. In each trial, the agent could search, view product details, read reviews, and choose an item.

Host: And what did the results show regarding marketing effectiveness?

Expert: All nine marketing cues significantly increased the selection rate above the control baseline, which sat at 11.5 percent in the primary model. However, their impact varied considerably. External, market-validated signals were the most effective. In the primary model, a Best Seller Badge raised the selection rate to 70 percent, Quality Certification reached 61.1 percent, and Purchase Volume reached 59.6 percent. In contrast, seller-provided cues like Stock Scarcity were much weaker, achieving only 22.5 percent.

Host: That differs from human shoppers, where scarcity warnings often drive immediate action. What happened to the agent's browsing behavior?

Expert: This is the central finding: the diagnosticity–exploration tradeoff. When an agent encountered a highly effective marketing cue, it compressed its search. Agents visited fewer product detail pages, read fewer review pages, and stopped searching earlier. For example, the Best Seller Badge reduced average search steps from 4.92 in the control to 4.41. Cues that strongly drove choice prompted the agent to commit without inspecting remaining alternatives.

Host: What are the practical takeaways here for business professionals and e-commerce managers?

Expert: First, business-to-agent marketing requires distinct playbooks. Tactics relying on emotional urgency, like low-stock alerts, appear far less persuasive to AI agents than validated popularity signals or certifications. Second, search compression creates a severe risk for competitors. If a rival product displays a strong badge early in the search results, the agent may terminate search immediately. That means competing products may never enter the agent's consideration set at all, regardless of their actual price or quality. Companies need to validate which cues capture agent attention in their specific software environment.

Host: That shows how critical early visibility and verified signals will be as agentic commerce expands. Alex, thank you for breaking down this study.

Expert: Thank you, Anna.

Host: And thank you for listening to A.I.S. Insights, powered by Living Knowledge. Join us next time for more data-driven discussions.
Artificial Intelligence & Automation Information Systems & Management Human-Computer Interaction Data Analytics & Big Data Digital Transformation & Strategy Cybersecurity & Privacy Cloud & Infrastructure E-Commerce & Digital Platforms Operations & Supply Chain Sustainability & Green IS

Agentic commerce, AI shopping agents, marketing cues, consumer search, e-commerce, search compression