Skip to main content
Agents Control Room is Topsort’s workspace for retailers to understand and manage how external AI shopping agents interact with their stores. An AI shopping agent is software that searches for and evaluates products on a shopper’s behalf. A shopper might ask ChatGPT to find an air fryer for two people with a dishwasher-safe basket, or a television under $400 that arrives before Saturday and has a simple remote. To evaluate those requests, the agent needs accurate product details, pricing, and availability. Agents Control Room helps your team understand what information agents requested, what your store returned, and what happened afterward.

What can I do?

Review agent visits and associated orders

In Agents > Sessions tab, you can review recorded sessions where an agent purchased a product or left without purchasing. Depending on what information the agent supports sharing, Topsort can show some or all of the below for a session:
  • The agent’s requests and product queries.
  • Product information returned to the agent.
  • Shopper requirements or additional context.
  • Whether the agent found answers to its questions in your catalog.
  • Associated orders, when an order can be linked to the session.
Select a session to examine the captured interaction and its outcome. For example, a television request may include requirements for price, delivery, and remote controls. Reviewing the session helps you understand which products were returned and whether their information addressed those requirements. For sessions that ended without a purchase, an unanswered question or another observable issue can reveal a potential barrier to purchase that your team can address. Activity must be captured through a supported agent connection or connected retailer systems.

Identify and fill gaps in product information

In the Catalog Health tab, the Answers for agents pane flags warnings about missing product information. Each warning highlights the number of times agents asked the question, the number of products affected and a severity rating, helping your team prioritize which gaps to address. Questions might include:
  • Is the air fryer basket dishwasher safe?
  • Does the television have a simple remote?
  • How long is the product warranty?
Suggested answers appear for unanswered questions. Your team can review each suggestion, confirm its accuracy, and approve or reject it. Approved answers become available for future agent requests. For example, an air fryer’s catalog entry may include capacity, wattage, and dimensions but omit whether its basket is dishwasher safe. Approving an accurate answer gives agents the information they need to evaluate that requirement, although it does not guarantee a purchase. Catalog Health is an existing Topsort capability for identifying catalog quality issues. Agents Control Room connects that work with questions arising from agent-led shopping.

Create and measure sponsored offers

Use the Offers tab to make brand-funded product offers available to eligible shopping interactions through supported agent connections. You can:
  • Select the products included in an offer.
  • Configure the offer, such as a product discount.
  • Set an offer budget.
  • Review associated orders, sales, and brand spend.
For example, a television brand can fund a $20 discount on a selected product. A supported agent connection can present that discount as a sponsored offer when the product is eligible for the interaction. The offer adds a commercial incentive to a relevant product. The product still needs to meet the shopper’s requirements, including delivery timing or the type of remote.

Manage agent connections and access

Use the Connections tab to review supported connections and manage their access to your store. You can allow or block individual connections (e.g. allow Gemini and ChatGPT, block Perplexity) and apply rules for unverified agents or identified price scrapers. For example, you may allow a shopping agent that requests product information for customers while blocking an identified bot that collects prices. Requests through a blocked managed connection are refused. These controls apply to access handled through managed connections. Their scope depends on how the connection handles access to your store.