Two of the largest marketplaces now answer an agent's request with a closed door, and a third answers it with a catalogue of a billion listings. As of November 12, 2026, access has become the first variable in agentic commerce, ahead of price and ahead of assortment, because an item that cannot be read cannot be compared. The Verge
The block is about the customer relationship
eBay's spokesperson described a "selective approach" to third-party access and said the company has restricted unauthorised scraping and training on its data, while evaluating how agents could help new customers discover inventory. The wording is careful and the intent is not: the platform will accept traffic an agent sends, and will not accept being replaced by one. The Meridiem
That distinction is easy to miss from the seller side. A merchant reads "restricted access" and assumes a technical outage. What actually happened is a decision about who owns the moment of discovery, and the seller is a third party to that decision.
Four platforms, four access regimes
The industry trend is toward selective access rather than open catalogues, and the four largest players now sit at four different points on that line. For a seller, the practical question is not which policy is right but which regime governs each channel it sells through. The Meridiem
| Marketplace | Agent access | What it means for a seller |
|---|---|---|
| Amazon | External agents blocked entirely | Only the platform's own assistant can surface the listing |
| eBay | Selective, restricted in November 2026 | Listing data must be assumed unreadable by outside agents |
| Target | Inconsistent: some requests pass | Visibility varies by agent and by request type |
| Shopify | Catalogue open to agent purchasing | Product data quality becomes the binding constraint |
The third row is the one that breaks reporting. An inconsistent regime produces inconsistent measurements, so a brand comparing its own visibility across channels is comparing four different definitions of visible. A category team that averages those four numbers is measuring nothing in particular, which is why access policy belongs in the analytics brief rather than in the legal folder.
A mirror site takes the description away
In testing, a blocked agent did not always give up. It sometimes redirected the shopper to an outside site mirroring the marketplace's listings, which means the product still appears in an answer while the description is no longer controlled by the brand or the platform. The Verge
What the shopper sees instead
A mirrored record carries a price from the moment of capture, a stock flag from the same moment, and whatever fields the scraper happened to keep. An item sold out three days ago can still be presented as available, and the complaint lands with the seller, not with the mirror.
This is the part of the agent story that rarely reaches a merchandising team. Blocking agents does not remove the brand from AI answers, it removes the brand's ability to correct them. And a mirrored listing is not a minor inaccuracy: the price is anchored to a capture date, the variant may be missing, and the shipping terms are whatever the scraper could see at the time.
What to test before the traffic arrives
Agent volume is still small enough that the temptation is to wait. Uber fulfilled four rides through third-party agents in the third quarter and DoorDash handled several hundred agent orders against 993 million total, which is a rounding error rather than a channel. The Meridiem
The counter-argument is that the platforms are behaving as if volume were already large. Amazon's blanket block, eBay's selective restriction and Shopify's open catalogue are all infrastructure decisions, and infrastructure is expensive to reverse once it is built into a seller agreement. Access rules also tend to arrive without notice, as eBay's did over a single weekend, which is a poor moment to discover that a channel depends on being readable.
Measure the door, not the ranking
Three checks are worth running now. First, whether an agent can read the listing at all, tested by sending a real agent to the category pages a brand cares about and recording where it stops. Second, whether the record it reads is complete: identifier, variant, price, stock with a timestamp, and shipping terms.
Third, whether the brand's own description survives into the answer. Where a marketplace blocks agents, the work shifts to the merchant's own surfaces: a catalogue the brand controls, structured attributes that a model can quote, and short factual statements with sources attached. Shopify's own data points the same way, with AI-referred sessions converting 80 percent higher than organic search between April and June 2026, and orders carrying 14 percent higher average values. AI Business Dispatch
None of that guarantees access, and no vendor can promise it while platform policy is being written week by week. It does mean that when a door opens, the brand is holding a record worth reading rather than a page built for a human to scroll. The brands that treat the catalogue as an interface, rather than as a by-product of the storefront, will be the ones that survive the next revision.
FAQ
Why did eBay restrict third-party agents?
It describes a selective approach to third-party access and says it restricted unauthorised scraping and training. The commercial motive is control of the discovery moment.
Does blocking agents remove a product from AI answers?
No. Agents sometimes redirect to mirror sites, so the item can still appear with a description nobody verified.
Which marketplace is most open to agents?
Shopify, which has opened a billion-listing catalogue to agent purchasing. Amazon is at the opposite end and blocks external agents entirely.
Is agent traffic large enough to matter yet?
Not by volume. Uber reported four agent-booked rides in a quarter and DoorDash several hundred orders, but platform policy is being set as if scale were imminent.
