A shopper asks ChatGPT which product to buy, and since October 10, 2026, as many as 20% of those answers on a given day have put a button under each recommended product: Shop at Best Buy, View at Walmart, Shop at Target. Where do those buttons send the shopper? Does ChatGPT still name your brand? Why did visits from ChatGPT fall on the stores we measure in the same week? Across the brands we track, 56% of the buttons pointed to a big-box retailer and about 5% to the brand’s own store, while ChatGPT named the brand as often as it did before October 10.
ChatGPT started adding “Shop at” buttons to shopping answers on October 10
Before October 10, fewer than 1 in 1,000 ChatGPT answers we tracked carried a retailer button. On the October 10 run, 11% of answers to shopping prompts did. From October 11 through 13, the share stayed between 16% and 20% every day.
| Day (UTC) | ChatGPT shopping answers with a “Shop at” button | Perplexity | Google AI Overviews |
|---|---|---|---|
| September 20 through October 9 | under 0.1% | under 0.1% | 0% |
| October 10 | 11% | 0% | 0% |
| October 11 | 20% | 0% | 0% |
| October 12 | 20% | 0% | 0% |
| October 13 | 16% | 0% | 0% |
The buttons appear only on shopping and buying prompts. Questions about how a product works get none. The buttons read “Shop at”, “View at”, “Buy at” or “Order from”, each followed by one retailer.
In the live answer, each recommended product renders as a block with a photo, a badge such as “Best overall” or “Best budget pick”, the price, a short description, pros and cons, and then the button. Some answers also end with a short interactive form asking a follow-up question.
OpenAI presents Intelligent UI as a better way to answer, and for a shopper the product blocks are an improvement: the product, the price and a place to buy it sit together, with no digging through citations to find a store. In our test, though, the links behind those buttons carried no referrer and no tracking tag, so the store that receives the shopper can’t tell that ChatGPT sent them. OpenAI’s engineers shortened the shopper’s path to checkout, and the clicks along that path no longer carry the signal brands used to measure what ChatGPT sends them. We can’t call that a fair exchange for the brands on the receiving end.
More than half of the buttons send the shopper to a big-box store
Across the four product brands we track, 56% of the buttons pointed to a big-box retailer such as Best Buy, Walmart, Target or Amazon. About 5% pointed to the brand’s own store, and competitor brands and specialist retailers took the remainder.
| Where ChatGPT’s buy buttons pointed, October 10-13, 2026 | Share of buttons |
|---|---|
| Big-box retailers | 56% |
| Competitor brands and specialist retailers | 39% |
| The brand’s own store | 5% |
The pattern held when we tested categories we don’t track. Asked for a durable carry-on suitcase, ChatGPT recommended a Travelpro and put Shop at Nordstrom under it, a Briggs & Riley with Shop at Macy’s, and a Samsonite with Shop at Walmart. None of the three buttons went to the suitcase maker’s own store. Asked where to buy a cast iron skillet, it linked Target, Wayfair and Walmart, and also linked three skillet makers directly.
A drop in ChatGPT referrals doesn’t mean ChatGPT stopped recommending you
We asked the same 250 shopping prompts before and after October 10. ChatGPT named the brand in 43% of answers afterward, against 40% before, and it cited big-box retailer sites as sources as often as it had.
| Same prompts, before and after October 10 | Before | After |
|---|---|---|
| Answer names the brand | 40% | 43% |
| Answer cites a big-box site as a source | 15.2% | 15.3% |
| Answer links to a big-box store | 3.7% | 19.2% |
| Answer links to the brand’s own store | 3.2% | 12.8% |
Answers carrying a link to a big-box store rose from 4% to 19%, and answers linking to the brand’s own store rose from 3% to 13%. The buttons send shoppers to brands more often than before, and to big-box stores far more often.
A prompt tracker will keep showing your brand holding steady while more of the purchases it leads to go through a retailer’s checkout.
Getting your product page into the answer’s sources is the strongest lever we found
When the brand’s own product page was among the answer’s sources, 34% of the buttons on that brand’s products went to the brand’s store, against 6% when it wasn’t.
| The brand’s own product page was among the answer’s sources | Button went to the brand’s store | Button went to a big-box store | Button went to another retailer |
|---|---|---|---|
| Yes | 34% | 42% | 24% |
| No | 6% | 33% | 61% |
A second analysis that matched products to brands more strictly found 39% against 7%.
Over those four days, one of the brands got none of the buttons on its own products: every one went to a big-box retailer.
These are four days of data, and the per-brand counts are small, so we’re treating the direction as solid and the exact percentages as provisional.
In our test, no click from ChatGPT told the store where it came from
When we opened ChatGPT and inspected the answers, every outbound link opened its destination without passing a referrer or a tracking tag. That held for all 47 links across nine answers, in temporary chats, regular chats and web search mode, from the buy buttons to the ordinary citation in an answer about seasoning a cast iron skillet. Copying an answer with ChatGPT’s own copy button gave the same untagged links. We clicked through on that citation, and the skillet maker’s page recorded an empty referrer.
Analytics tools identify a ChatGPT visit in one of two ways: a referrer of chatgpt.com, or a utm_source=chatgpt.com tag on the link. Server-side tools read them from edge and server logs, and client-side tools such as GA4 read them in the browser. A visit with neither gets filed as Direct by all of them.
On one store we measure, the tag was already doing most of the work before October 8: 94% of identified ChatGPT visits arrived with the tag and no referrer. The links in our test carried no tag at all.
From October 9 through 13, arrivals from ChatGPT on that store averaged about a third of their earlier level. The fall came within a single hour, around 21:00 UTC on October 8, and hit both kinds of identified visit: tagged visits fell to 35 on the index below and referrer-only visits to 21. A second store we measure separately fell to 43 the same day. Arrivals from Gemini on the first store indexed to 116 over the same days.
| Day | Arrivals from ChatGPT | Arrivals from Gemini |
|---|---|---|
| October 1-7, daily average | 100 | 100 |
| October 8 | 86 | 111 |
| October 9 | 36 | 64 |
| October 10 | 33 | 108 |
| October 11 | 34 | 108 |
| October 12 | 36 | 128 |
| October 13 | 33 | 169 |
Indexed to the October 1-7 daily average = 100. The store’s ChatGPT arrivals had stayed between 83 and 128 on this index every day from September 14.
The ChatGPT visits that still show up nearly all carry the tag, and they come from every kind of device, with desktop computers keeping more of their visits than iPhones. They arrive through the same citation links as before, landing on the same kinds of pages in the same proportions, at about a third of the volume. None of the links we tested on ChatGPT’s web app carried the tag, so the remaining tagged visits most likely come from people whose ChatGPT hasn’t switched to the new link behavior yet. A rollout still in progress would show that remaining third shrinking after October 13, and we’re tracking it. An earlier and smaller loss on September 25 removed the referrer from untagged ChatGPT clicks, so the October 1-7 baseline already undercounts how many people ChatGPT sends.
The two findings fit together: answers that link the brand more often, alongside fewer visits recorded as coming from ChatGPT. We haven’t confirmed that one explains the other. Our live test was the first time we observed the missing referrers, with no earlier test to compare against, so we can’t date the change. The check that settles it is whether Direct visits to the same product pages rose from October 8 by about what ChatGPT referrals lost.
The change lines up with OpenAI’s Intelligent UI rollout
OpenAI rolled out GPT-6 with Intelligent UI in ChatGPT on October 7, 2026, for Plus, Pro, Business and Enterprise users, and on October 8 for Free and Go. Intelligent UI lets an answer mix text with interactive elements such as charts, buttons and forms. OpenAI hasn’t said how it chooses the retailer behind a button, and we found no announcement of a retailer partnership or affiliate program dated to that week.
Retailer links aren’t new to ChatGPT. Profound measured 22.5 million retailer offers on ChatGPT Shopping’s product cards in March 2026 and found Walmart holding the most first-position offers and Target the broadest presence. What appeared on October 10 is a single retailer button inside the answer itself, under each product the answer recommends.
The referral fall on the store we measure came on the evening of October 8 (UTC), the day the rollout reached Free users. The buttons first appeared in our tracking two days later, on October 10. The two-day gap doesn’t show the changes are unrelated, because a rollout reaches different accounts on different days.
How to read your numbers from October 8 onward
Annotate October 8, 2026, on every chart with a ChatGPT, AI referral or referral-channel series, and stop comparing windows across it. A month-over-month comparison of AI traffic that straddles October 8 isn’t an apples-to-apples comparison.
Track mentions separately from referrals. A prompt tracker shows whether ChatGPT names you, which referral counts can’t. Mentions holding while referrals fall points to a change in how ChatGPT sends clicks, and lost visibility shows up as a fall in both.
Watch Direct traffic on your product and category pages. ChatGPT clicks that arrive without a referrer land there as Direct, so a rise in Direct sessions on deep product pages from October 8 is the trace they’d leave.
Make your product pages readable without JavaScript. The button went to the brand’s own store far more often when ChatGPT had read the brand’s product page. Put the price, availability and Offer structured data in the HTML the server sends, so an assistant reading the page gets them.
For many shopping answers, the button goes to a retailer whatever you do. Your listing at Best Buy, Walmart or Target is where that shopper lands, so its price, stock and content belong in your AI visibility work.