On 10 October 2026, a black "Shop at" button started appearing under products in ChatGPT's shopping answers. Before that date it was essentially absent. Within three days it was showing up in roughly one in five shopping answers.
The button matters more than its size suggests, for two reasons. It is now the most prominent way out of a shopping answer — and it is a button, not a link, which quietly breaks several things marketers rely on.
Key takeaways
- "Shop at" buttons went from under 0.1% of shopping answers before 10 October to 11% on 10 October and 16–20% per day from 11 to 13 October (WISLR, 4 brands, 250 prompts).
- Of those buttons, 56% point at big-box retailers, 39% at rival brands or specialist stores, and about 5% at the brand's own store.
- The single strongest lever we found: when a brand's product page is among the answer's sources, the button points at the brand's own store 34% of the time. When it is not, 6%.
- A button is not a link. You cannot hover it, you cannot right-click to copy its URL, and most of the time it sends no referrer.
- Buttons frequently carry
utm_medium=feed, meaning the destination came from a merchant catalogue rather than a crawled page.
How fast it arrived
WISLR tracked four brands across 250 shopping prompts and dated the switch precisely. Before 10 October, fewer than one in a thousand shopping answers carried a buy button. On 10 October it was 11%. From 11 to 13 October it settled into a 16–20% daily band.
Two things are worth noting about that chart. First, the rise is a step function, not a ramp — this was a deployment, not a gradual rollout. Second, over the same window, Google AI Overviews showed the button in 0% of answers, as did the other AI assistant WISLR observed. This is, for now, a ChatGPT-specific surface.
Where the buttons send people
The left half is WISLR's distribution across those 250 prompts: 56% of buttons opened a big-box retailer — Best Buy, Walmart, Target, Amazon and the like — 39% opened a rival brand or a specialist store, and roughly 5% opened the brand's own store.
Our own hands-on numbers point the same way. Across 97 buy links logged on 21 October, 51 opened a retailer product page against 20 on a brand product page, and that is with several prompts explicitly requesting official-site links. Two independent samples, two methods, same direction.
The one number worth acting on
The right half of that chart is the finding we would put in front of a CMO.
When a brand's own product page appears among the sources cited in the answer, the buy button points at the brand's store 34% of the time. When the brand page is not among the sources, it is 6%. On a stricter definition of "brand store", the split is 39% versus 7% — different absolute values, same direction.
That is a roughly five-fold difference, and it reframes the problem. Teams tend to treat "getting cited" as a PR and content metric and "getting the sale" as a commerce metric. In a ChatGPT shopping answer they are two ends of the same mechanism. If your page is not in the evidence, the purchase link almost certainly goes to a retailer. That is not a conversion-rate problem you can fix with a better landing page. It is a visibility problem, and it is fixed upstream — see how AI search picks sources and how to get cited by ChatGPT.
There is a second-order effect worth planning for. Because the source panel now shows only one site name per paragraph, with the rest folded away, being a source is no longer enough — the first-ranked source takes most of the exposure. We cover that collapse in why ChatGPT now shows one source per paragraph.
A button is not a link, and that breaks things
We checked every clickable element in the answers one by one. Most of them do not leave ChatGPT at all.
Element | Type | What happens on click |
|---|---|---|
Product name | Button | Opens a side panel; the page does not navigate. Usually a product description; a quote list when you asked for cards or comparisons |
Product image | Button | Opens inside ChatGPT; never goes to a merchant |
Text buy link ("Shop at", "Buy from", "View at") | Link | Hover reveals the URL and it can be copied; click navigates with |
Black buy button ("Buy at …") | Button | No "Copy Link Address" in the context menu; click navigates, usually carrying |
"Visit" inside the side panel | Link | Navigates to the merchant's product page |
Source marker (site name + N) | Button | Opens a source card; only clicking through the card opens the original page |
The screenshots above show the difference on one screen. Hovering "View at Apple" surfaces the apple.com URL — complete with its truncated ?utm_s… parameter — and the context menu offers Copy Link Address. The "Buy at Lowe's" black button offers page actions only.
Why this is not a pedantic distinction:
- You cannot audit destinations at scale by hand. Hovering tells you nothing; you have to click and land.
- Referrer is unreliable. Button-type entry points are generally opened with
noreferrer, so you cannot identify ChatGPT from the referring domain. The only dependable signal is the UTM parameter on your own landing page — see utm_source=chatgpt.com explained. - Buttons do not behave like links for users either. No middle-click to a new tab, no preview, no copy-and-send-to-a-friend.
One thing we did not test: whether the URL behind a black button is already present in the page data before the click.
The multi-store trap
In all 28 prompts, two Soundcore earbud models were sent to four different domains inside the same corporate group: soundcore.com, us.soundcore.com, us.anker.com and us.ankersolix.com. In the run where we asked for product cards and buy buttons, the "Buy at Soundcore" button for the Sleep A30 landed on an Anker SOLIX store page, with utm_campaign=openai_catalog and utm_medium=feed in the URL — proof that the destination came from the merchant catalogue, not from a crawl.
If you run multiple storefronts for the same group, ChatGPT will pick which one receives the traffic, based on catalogue data, and it may not pick the one you would. Before anything else: align price and stock across stores, decide which store should receive ChatGPT traffic, and submit your catalogue accordingly. The mechanics are in the ChatGPT product feed.
FAQ
What is the "Shop at" button in ChatGPT? A black, full-width button that appears under a recommended product in a shopping answer and navigates to a merchant page. It started appearing on 10 October 2026 and is implemented as a button, not a hyperlink.
Why does the "Shop at" button point at Amazon instead of my store? Most likely because your product page is not among the sources the answer cited, and because your product data reaches ChatGPT through a retailer's catalogue rather than yours. Brand pages that are cited get the button 34% of the time; pages that are not get it 6%.
Can I control which store the button opens? Not directly. You influence it by being cited as a source, by submitting an accurate catalogue feed, and by keeping price and stock consistent across your storefronts.
Can I track clicks on the "Shop at" button? Only from your own landing page. The button usually sends no referrer, so identify the traffic with utm_source=chatgpt.com and separate catalogue-sourced clicks with utm_medium=feed.
Do other AI assistants have this button? Not in the same window. Google AI Overviews showed it in 0% of answers over the same period, as did the other assistant WISLR tracked. If you want a single view of how you appear across ChatGPT, Google AI Mode, Gemini, Google AI Overviews and Microsoft Copilot, that is what GEOly's AI shopping monitoring tracks.
Sources and method
Prevalence and destination percentages are from WISLR, published 14 October 2026 and updated 16 October: 4 brands, 250 shopping prompts, 10–13 October. Four days of data; treat the figures as directional, not as a benchmark.
The element-type table, the 97-link destination count and the multi-store observation are from GEOly Research: 28 shopping prompts run in a signed-out Chrome window on 21 October 2026, with every buy link clicked and logged. The full method and the rest of this series start at how ChatGPT shopping results work after GPT-6.
