Amazon Adds 24/7 Monitoring to Seller Assistant, With Merchants Approving Actions

A merchant may no longer have to ask Amazon’s Seller Assistant what needs attention. Amazon is adding continuous monitoring, memory and seller-defined automations so the AI can spot changes and prepare a response. Mary Beth Westmoreland, Amazon’s vice president of Worldwide Selling Partner Experience, described the expansion in a September 23 interview with PYMNTS. Sellers must still approve actions.

Work between questions

Westmoreland said the assistant can assess inventory, pricing, advertising, demand and compliance together. Its long-term memory is intended to build an understanding of a seller’s goals, habits and preferences, rather than start each interaction without that context. A seller can also describe a task in plain language and leave the assistant watching for the relevant change.

One task could be watching a merchant’s top products for falling ratings and preparing a response. Another could be looking for an opening in a product category and proposing a change to advertising spending. Those examples show the intended shift: the seller sets the watch, and the assistant does the monitoring instead of waiting for another question. The proposed work is not the same as an approved action.

Westmoreland illustrated the decision step with a fictional homewares seller preparing for the fourth quarter. The assistant identifies demand for a planter and suggests changes to its description, search terms, price and promotions. The seller can inspect scenarios in a generated visual interface before deciding whether to authorize the changes. It is an example of the workflow, not a reported result from a merchant using it.

A route outside Seller Central

Amazon is also introducing a selling-partner plugin for Amazon Quick and, initially in beta, Anthropic’s Claude. It is meant to make Seller Assistant’s seller information and capabilities available in those tools, so a merchant need not start every AI task in Seller Central. Westmoreland said roughly 90% of Amazon’s selling partners already use some form of third-party AI. That figure is a company executive’s estimate, not a measure of how many will use this plugin.

Roughly 90%Third-party AI use

Westmoreland’s estimate of Amazon selling partners who already use some form of third-party AI.

More than 90%Recommendations accepted

Westmoreland’s figure for Seller Assistant recommendations that sellers accept; it is not the plugin’s adoption rate.

PYMNTS described a Proof Culture use case that reaches further than a single answer. An AI agent spotted an upcoming sneaker release, checked Amazon inventory, recommended a restock, drafted a purchase order and prepared an email to employees about expected demand. The account shows the range of work an agent can prepare. It does not establish that the purchase order was sent or the restock completed without approval.

The boundary around action

The plugin creates a distinct access decision. Westmoreland said sellers choose what information it can reach and can revoke that access. She said seller data remains inside Amazon’s infrastructure, while sellers must explicitly approve actions. The distinction is between allowing an assistant to keep watch and giving it permission to carry out what it proposes.

That boundary may matter most as the assistant becomes more persistent. A seller can authorize access and define what to monitor without approving every recommendation it produces. For now, Amazon’s stated design keeps the final action decision with the merchant. Whether that arrangement saves sellers attention will depend on the usefulness of the work the assistant puts before them—not simply on how much it can watch.