User Behavior Study

In AI Shopping,
the Product Card
Is Your Storefront

Getting recommended in ChatGPT is where the competition starts. This study follows what shoppers did next, from the product card to the offer card.

Download the study

40 U.S. shoppers · 224 recorded shopping tasks across 6 product categories

Key findings

AEO and GEO are about getting your brand mentioned in ChatGPT’s written answer. Shoppers in this study read that answer, but they rarely chose there. Instead, they picked a product from the product cards, then a seller from the offer cards. To get chosen, brands and retailers have to win both.

Buying decisions happen in the product cards.

If your product isn’t in the product cards, it’s missing where most choices get made.

84%

of final product choices in ChatGPT came from a product card

  • Shoppers go to the product cards first.

    That makes your product data your first impression: the image, title and price on the product card.

    75%

    of first interactions were with product cards, not the written answer

  • Product card position pays.

    With several product cards to choose from, shoppers picked the first one most often, by a wide margin. Treat the top spot as a priority.

    43%

    of choices went to the first product card

  • And it pays twice, at the offer card.

    Choosing the product doesn’t settle who gets the sale. Your offer has to win a second comparison on price, availability and delivery.

    76%

    of choices went to the first offer card

We turned this proprietary study into 8 actionable recommendations for brands and retailers. Get them in the full study, with every chart and shopper quote.Each share uses its own base of shopping tasks, listed in the report.

Download the study

What’s inside

  • 8 actionable recommendations for brands and retailers

    What to do about each finding, from showing up in the product cards to winning the offer card.

  • Why shoppers trusted their pick

    Specs, price and brand were the main reason shoppers trusted a pick in 82% of shopping tasks. Reviews were the main reason in 2%.

  • I’m a little confused because this is $140, but everything I see here looks like $160.
    Study participant

    Where product data broke down

    Shoppers caught wrong prices, mismatched sizes, missing images and attributes that contradicted each other.

  • How shoppers asked

    Full sentences, few follow-ups, and pushback when the answer added a requirement nobody asked for.

  • What shoppers made of sponsored results

    When ads helped, when they backfired, and whether shoppers could tell what was sponsored.

About this study

Research partnerClickstream Solutions

ReFiBuy commissioned this study. Clickstream Solutions designed the research, coded the sessions, and ran the analysis. ReFiBuy contributed the Agentic Commerce Optimization (ACO) lens: what the findings mean for brands and retailers, and what to do about them. The study measured selections, not completed purchases.

  • Eric Van Buskirk

    Eric Van Buskirk

    Research and analysis

    Clickstream Solutions

    Founder of Clickstream Solutions. Eric has directed research studies for some of the best-known names in SEO and AI search, most recently behavioral studies with Kevin Indig and Profound on how people use AI platforms.

  • Scot Wingo

    Scot Wingo

    Commerce strategy

    ReFiBuy

    Co-founder and CEO of ReFiBuy. Scot is a serial entrepreneur with deep ecommerce experience, founding five startups and taking one public: ChannelAdvisor (now Rithum).

  • Brian Chapman

    Brian Chapman

    Editorial

    ReFiBuy

    VP of Growth at ReFiBuy. Brian has over a decade of experience building go-to-market engines at B2B SaaS and high-growth startups, including Spiffy and Dude Solutions (now Brightly).

6
product categories
158
ChatGPT tasks
66
Google AI Mode tasks
5,500+
coded observations

Forty U.S. participants took part in remote, unmoderated desktop shopping sessions using ChatGPT and Google AI Mode. After consent, UXtweak, a usability research platform, captured their screens and voices as they narrated their actions. Trained annotators coded every recording across 32 research fields.

Agentic commerce is reshaping where products compete

Shoppers researched and picked products on the product cards, then chose where to buy on the offer cards. Get 8 actionable recommendations for winning both.

Download the study

ReFiBuy finds and fixes product
 data gaps before shoppers do.

ReFiBuy continuously evaluates, enriches, and monitors the product data behind your product cards and offer cards, so AI shopping agents get them right.

Book a demo