Walmart wired around 200,000 SKUs into OpenAI’s Instant Checkout. Then their EVP of product shared the data: ChatGPT-driven purchases converting at roughly a third of Walmart’s own site rate. He described the result as “unsatisfying” — which is corporate for “we’re not doing that again.”
What Walmart did next is more interesting than the failure itself. They built Sparky — their own assistant — and embedded it inside ChatGPT. The buyer logs into Walmart from inside the chat, carts sync across platforms, and the purchase completes inside Walmart’s system. Gemini is next. Owned environment, owned moment of decision. The chat window becomes the shop window — not the till.
The layers are coming apart
Two conclusions follow, and both matter for anyone selling online.
First, discovery and checkout are separating — faster than expected. The working assumption across the industry was that assistants would absorb both: the natural endpoint being a checkout inside the answer, with the retailer reduced to a fulfilment partner. Walmart’s experiment was the largest single test of that thesis so far, and it didn’t hold. Conversion several times worse isn’t a tuning problem — it’s structural. Buyers need their account, saved payment methods, order history, and returns flow: the entire scaffolding of trust that an owned checkout carries. The chat window doesn’t have it, and apparently can’t fake it well enough to matter.
Second, one number keeps recurring: AI-referred shoppers convert substantially better than traditional-channel visitors — but only once they land on the merchant’s own surface. Discovery on the assistant. Decision on your site. That gap is where the value is.
What this means for merchants
The question we hear most, usually from founders running stores between £1M and £20M: “Should I be optimising for the checkout in the chat?”
The answer, more confidently than we’d have given it a month before Walmart’s data: no. Not yet. Maybe not ever, for most categories. Optimise for being the brand the assistant names. The conversion happens on your turf, where it always has. The naming is the new work — and the naming is where most merchants are presently invisible.
Two practical shifts follow:
- Drop AI-referred conversion rate as your obsession metric. The cohorts are still thin and noisy. The metric that matters is whether the assistant names you at all — visibility, citation presence, mention consistency across the engines that matter. Conversion looks after itself once the visibility is real.
- Stop treating “agentic commerce” as one thing. There are at least two layers moving independently. The discovery layer is going through real, measurable change right now. The transaction layer is going through the change everyone talks about — and the evidence so far is that it’s slower and less certain than the noise implies.
The smartest operators have stopped trying to be everywhere in the stack and started picking a layer. The discovery layer — being the answer when a shopper asks — is the one where the change is real today, and it’s the one Geoffy is built for.
Are the assistants naming you? Get your free GEO Score and find out.