An AI agent for choosing an expensive purchase searches for product information, compares specifications, and narrows down the list of options. It helps you prepare, but product inspection and purchase approval are best left to a human.
The final decision remains with the buyer.
This approach is especially useful when specifications matter but personal impressions cannot be replaced by a catalog description. Ron Johnson, one of the creators of the Apple Store chain, expresses a similar idea: a buyer is unlikely to want to entrust an agent with independently purchasing a laptop for 1 000–2 000 dollars without seeing the device. In Johnson's view, AI should help study a product, not replace the final choice.
Start by Setting Your Selection Criteria
Before searching, define your budget, must-have specifications, and the compromises you are unwilling to make. For a laptop, mandatory requirements might include screen size, memory capacity, and port types. The color of the casing or its thickness may sometimes be acceptable compromises. Separately list the features you want to check in person: weight, keyboard comfort, brightness, and screen quality.
The phrase “find the best laptop” is not enough. Ask the agent to compare models within a specified budget, list suitable options, and explain the criteria by which each one is better or worse.
If the agent does not know what matters to you, it will fill in the gaps with its own assumptions. The result will not necessarily be bad, but the list may reflect someone else's priorities rather than yours.
Ask for a Comparison Based on Verifiable Data
Have the agent consolidate the models into a table listing the price, exact configuration, weight, dimensions, display specifications, warranty, and availability. Each parameter should include a source and a link to the seller's or manufacturer's page where the information can be verified. If a source does not confirm a specification, the agent should mark it as unknown rather than fill in the picture with a guess.
Configurations require separate verification. A single model may come with different processors, memory capacities, displays, and part numbers, so comparing products by name alone can be misleading. Ask the agent to specify the exact device variant and clearly note the differences between configurations.
What should you do if the price or availability in the response differs from the store listing? Treat the seller's terms at the time of purchase as current. OpenAI warns that price and availability information in ChatGPT product results may contain errors or be updated with a delay, so it must be checked with the seller before ordering.
Leave Preliminary Selection to the Agent
Automate the search, initial comparison, and availability checks. The agent performs these tasks using catalog and website data, but the result remains preliminary: the selection may not include every product, and the description may not always reveal differences that matter to you.
Do not give the agent unconditional authority to pay for an expensive purchase. Even if the service supports order placement, the decision about the model, seller, and final price should remain with the buyer. In Johnson's concept, AI makes a person better prepared to make a purchase, including a store visit, rather than eliminating the need to choose.
The division of tasks is simple: the agent searches and compares, while the human inspects the product and chooses. Before paying, check the final amount, delivery, seller, warranty, and return policy.
Check What the Specifications Do Not Show
A laptop's weight is listed in the specifications, but the figure does not tell you how comfortable the device will be to carry every day. An image of the screen does not reliably show its brightness in your lighting, reflections, viewing angles, or eye comfort. The dimensions in a product listing also do not show whether the device is comfortable to hold or will fit in your bag.
If possible, hold the selected model in your hands, open the screen, and try the keyboard and touchpad. When dimensions or ergonomics are especially important to the purchase, visit a store or find a seller with clear return conditions. A virtual image and a list of specifications cannot replace this kind of check.
Johnson connects the value of a physical store with the opportunity to try a device, get help, and learn how to use it. This is his assessment of the role of retail, not a universal rule for every buyer. For equipment whose comfort depends on personal impressions, an in-person inspection helps verify what is difficult to convey through a catalog.
Put Human Approval Before Payment
If the agent can place an order, configure the process so that before payment it shows the product, configuration, seller, total cost, and delivery and return terms. The buyer should personally confirm the order after checking this information. When a service does not clearly separate purchase preparation from purchase approval, do not give it your payment details or permission to place the order independently.
In OpenAI's published description of Instant Checkout from 29 September 2025, the user confirmed the purchase steps, while the seller processed the order, payment, and delivery. In a later description of OpenAI's product features from 24 March 2026, the company said it was focusing its work on product discovery and allowing sellers to use their own checkout process. This information shows why specific commerce features in chatbots need to be checked directly before use: they change.
Check Whether the Agent Distorted the Comparison
A recommendation for the “best option” does not mean that the product is objectively better in every respect. Ask the agent to show the basis for its conclusion and list the disadvantages of the selected model. Then compare the key data with several primary product listings, especially when the differences affect price or compatibility.
- Check that the agent compared identical or comparable configurations.
- Make sure the price applies to the intended seller and whether it includes delivery.
- Separate manufacturer specifications from summaries of reviews and advertising labels.
- Ask it to flag products for which confirmation of an important specification could not be found.
- Cross-check the warranty, availability, and return policy on the seller's side.
In its guide to ChatGPT product search, OpenAI states that names, descriptions, brief review summaries, and some labels may be generated by the model based on third-party data. The company also recommends confirming the final price, availability, warranty, and return policy with the seller. This list can also be used to check other AI agents.
When to Stop Automated Selection
Hand the decision over to a human if the agent cannot confirm a key specification, shows conflicting price or availability information, or fails to distinguish between configurations. Stop the process as well if the proposed product exceeds the specified budget, the seller is unclear, or the return terms do not meet your requirements.
Set a rule for the agent in advance: when the data conflicts, it should not choose a model but should list the discrepancies and ask the human. This process reduces the risk that a confident recommendation will conceal a lack of information.
For developers of shopping AI agents, this means separating comparison and purchasing in the interface. The agent can be entrusted with gathering information and checking availability, but before payment the buyer should see the specific product and transaction terms and then explicitly confirm the action.
A Short Action Plan
- Write down your budget, mandatory specifications, and unacceptable compromises.
- Ask the agent to compare several exact models and provide sources.
- Check the specifications, configurations, price, and availability on the seller's or manufacturer's websites.
- Inspect the device in person if the choice depends on weight, screen, keyboard, or dimensions.
- Before paying, check the total amount, warranty, delivery, and return policy.
- Confirm the purchase yourself; stop the process if there are discrepancies.
An AI agent is suitable for preparing for an expensive purchase: it gathers information and helps compare options. The final choice remains with the human, who checks the device's physical properties and confirms the order.
Compare models before you start
The service sets its plans, limits and model catalog. If they differ from this article, contact us so we can update it and record a new review date.
Browse modelsAffiliate link: your price stays the same and the project earns a commission.