Here is a number you will not see in any AI keynote: 3x worse.
Walmart embedded 200,000 products into ChatGPT with an instant checkout feature. Type what you want, the AI finds it, you buy it without leaving the chat. The future of shopping, right?
The conversion rate was one-third of their regular website.
Not 3x better. 3x worse.
Walmart’s EVP called the experience “unsatisfying”. They are now pivoting to embedding their own chatbot inside ChatGPT instead of using the native checkout.
Why this is the most important AI data point of the month
We are drowning in AI demos. Every company has a “powered by AI” announcement. Every keynote promises that AI agents will handle everything from shopping to scheduling to filing your taxes.
Walmart actually did it. At scale. With real customers and real money. And the result was: people do not want to buy things inside a chatbot.
This is not a failure of technology. ChatGPT works. The products were real. The checkout was functional. The failure is a failure of understanding how humans actually shop.
Why chat is wrong for commerce
Shopping is not a conversation. It is a visual, comparative, exploratory process. You browse. You compare. You read reviews. You look at photos from multiple angles. You check the price against three other stores. You put things in a cart and come back tomorrow.
A chat interface collapses all of that into a text stream. “Show me running shoes”. Here are three options. Pick one. Buy now.
That is not how anyone shops. It is how someone who has never shopped imagines shopping works.
The agentic commerce reality check
Google launched Universal Commerce Protocol at NRF in January. Every major AI lab is pitching “agentic commerce” where AI agents shop on your behalf. The entire thesis is that conversational AI will replace traditional e-commerce interfaces.
Walmart just provided the first real data point against that thesis. Not a theoretical argument. Not a prediction. Actual conversion data from the largest retailer in the world, using the most popular AI chatbot in the world.
3x worse.
What builders should take from this
If you are building an AI product that replaces an existing workflow, your competition is not other AI products. Your competition is the existing workflow. And existing workflows are optimized by decades of iteration and billions of data points.
Walmart.com converts well because every pixel on that page has been A/B tested to death over 15 years. The search algorithm, the product ranking, the image sizes, the review placement, the checkout flow. All of it optimized.
Putting that inside a chat window and expecting it to work better is not innovation. It is ignoring the optimization that already happened.
The lesson: AI is not automatically better than what it replaces. Sometimes the old way is the old way because it works.
Source: Search Engine Land