We Tried OpenAI Operator—Here’s What Ecommerce Brands Need to Know to Succeed in the Age of AI Shopping Agents

AI

Table of Contents

Share

The way consumers shop online is rapidly changing. This time, the shift isn’t about new payment methods, flashy website redesigns, or social commerce trends. It’s about AI shopping agents, like OpenAI Operator, that are starting to do the shopping for customers. We predicted way back in 2023 that AIs would start shopping for people, and now that vision is truly becoming a reality.

We put OpenAI Operator to the test, and there’s a clear set of criteria emerging for what will set up brands for success with these types of AI agents. The thing that’s really interesting here is that these agents aren’t just recommending products (which is primarily what we tested) —they’re filling carts, making purchases, and even comparing deals across retailers. The future of ecommerce isn’t just about human buyers anymore—it’s about how well your website interacts with AI-powered agents.

What is OpenAI Operator and How does it Work?

OpenAI Operator is an AI-powered agent that can navigate websites, fill out forms, and interact with buttons—all with little to no human intervention. It’s designed to act like a digital assistant, automating online tasks like ordering groceries, booking reservations, and comparing prices of goods.

Unlike traditional shopping assistants that rely on structured data feeds or retailer APIs, OpenAI Operator interacts with ecommerce sites like a human user. It “sees” web pages through screenshots, clicks buttons, and enters text into forms, making it a highly flexible AI shopping agent. 

Diving into the technical details a bit further – OpenAI Operator uses a new model referred to as “CUA” – Computer Using Agent. This model navigates to websites, then takes screenshots of web pages as they render. It then interprets the contents of those screenshots, looking for buttons, text inputs, filters, and other significant elements to know where to click and what to do with the web page to continue to work towards a user’s goal. 

A diagram explaining how OpenAI Operator works

A diagram of the CUA model – source

What’s really interesting about CUA is the fact that it doesn’t rely on structured data or crawling the actual HTML content of pages, like other web scrapers and crawlers do. It runs an actual web browser in a virtual machine and uses screenshots to determine what to do next. This makes the experience slightly closer to how a human would perceive and understand things, since we also go purely off of the visual cues on a page to know what to do. 

For ecommerce brands, this means that customers might never set foot on your website in the way they traditionally have. Instead, AI agents will be the ones engaging with your product listings, evaluating pricing, and making purchasing decisions on behalf of users. 

Taking OpenAI Operator for a Test Drive

We tested out OpenAI Operator through several ecommerce scenarios and were impressed with the results. Based on what we saw, there’s a clear-cut set of items to focus on for ecommerce brands.

Scenario #1 – Shopping for Legos 

The first scenario we tried utilized this prompt:

“Find the highest rated Lego set under $100 that a 9-year-old girl would enjoy playing with on target.com”

In this case we were targeting (no pun intended!) a specific retailer. Operator methodically worked through the process, starting on Target’s website, then searching for Lego, and utilizing the filters to get down to a group of Lego sets that made sense given the prompt. 

OpenAI Operator navigating to Target.com

We started by navigating to Target’s website

OpenAI Operator filtering Lego by price

Then the agent searched for “Lego” and filtered the results down to $50-100 and >4 stars

OpenAI Operator looking at a Lego product listing page

This brought up a series of Lego sets that reasonably met our criteria

OpenAI Operator looking at a Lego product detail page for an Animal Crossing Lego set

It then reviewed several Lego sets and copied their links

OpenAI Operator looking at a Lego product detail page for an Earth and Moon in Orbit Space Lego Set

The end results were recommendations for two Lego sets, with links to purchase

The above scenario took about 2 total minutes to complete. In viewing the results that were returned, they met the prompt criteria well, and with a more specific prompt on the preferences of the child being shopped for, no doubt even better results would have come back. Additionally, we could have asked for more options or details on the sets, or for Operator to purchase them for us, but we stopped here. 

The final list of actions that Operator took were also listed above the browser window it was utilizing:

Navigating to Target’s website now
Using search bar for LEGO sets
Filtering LEGO sets by age group
Applying filters for LEGO sets
Selecting price range, applying filter
Applying guest rating filter for LEGO
Applying 4 stars filter for LEGO
Selecting highest-rated LEGO sets
Reviewing details for LEGO set
Right-clicking to open new tabs
Clicking product images to open pages
Copying link for LEGO set
Returning to tab for LEGO set
Copying product link for LEGO set

One thing we noticed was the high level of interaction with on-site search and filtering features, which weighs into our recommendations later in this post. 


Scenario #2 – Shoe Shopping

The second scenario we tried utilized this prompt, against the website of one of our clients, Martin Dingman:

“Can you browse casual shoes from Martin Dingman and find a pair that my 60-year-old father would enjoy? He likes brown or blue shoes and wears a size 10. He’s a stylish dresser. I want to spend less than $300”

Like before, Operator got straight to work and started browsing the Martin Dingman website for options. Interestingly this time, it did a Bing search to start with, then used the results in that search to navigate forward to complete its task.

OpenAI Operator searching Bing for Martin Dingman Shoes

Starting off with a Bing search, since that’s OpenAI Operator’s browser’s default search engine

OpenAI Operator navigating to Martin Dingman's Website

Operator got straight down to business, navigating to shoes from the home page

OpenAI Operator filtering to size 10 shoes on Martin Dingman

Once Operator found the shoes, it immediately went to filter by size 10

OpenAI Operator filtering to blue and brown shoes on Martin Dingman

It then filtered further by color

OpenAI Operator looking at a blue shoe 
 product detail page on Martin Dingman

It then double checked on the product detail page that the size and color needed were available

OpenAI Operator recommending a specific Martin Dingman shoe product

And there’s the final result – a pair of casual blue joggers

Operator performed similarly well on this task. In retrospect, we probably should have asked it to give us a few different options back to choose from, but it likely would have been up to the task given the Lego example. You can see the final list of steps it completed below – and note that it had to do some additional scrolling and moving around the site to meet its goal:

Searching Martin Dingman for shoes
Clicking Martin Dingman for shoes.
Hovering over shoes for categories
Clicking on shoes category for options
Selecting casual shoes category now
Selecting casual shoes category filter
Closing irrelevant sign-up pop-up
Exploring filters for casual shoes
Selecting size 10, filtering colors
Scrolling back to select size 10
Scrolling to find size 10
Scrolling “Size” filter for 10
Scrolling up to select size 10
Scrolling down to find size 10
Scrolling size filter for option
Scrolling up, searching for size 10
Selecting size 10, filtering colors
Selecting colors: Blue and Brown
Applying color filters for selection
Adding blue color filter option
Scrolling through products, checking prices
Sorting products by lowest price
Sorting by price, seeking budget shoes
Sorting products by ascending price
Checking prices of casual shoes
Scrolling down for shoe prices
Clicking on loafers for pricing
Retrying click, accessing loafers details
Checking availability of jogger sneakers
Sharing link for stylish sneakers

One last thing that’s interesting to note is that Operator will run in the background – so you can set it and forget it on tasks like this while you focus on other things. 

So, what can ecommerce brands do to prepare and optimize for these scenarios? A few clear patterns emerged from our testing.


How Ecommerce Brands Should Prepare for AI Shopping Agents

To stay competitive, ecommerce brands need to ensure their websites are AI-friendly—optimized for both human and AI shopping experiences. There are several steps you can take to make sure this happens. 

1. Make Sure You’re Found on Bing (Yes – Bing)

In the test we ran, Operator did a search for the Brand on Bing (which appears to be the default search engine in the virtual browser it uses). It would then click on the most prominent result to get to the website of the brand. 

While we often put our focus on Google, it is important to remember that Bing is a default search engine for many users, and potentially in the future for certain AI agents, and should be a part of your SEO strategy. You’ll want to ensure that you’re ranking first in branded search both on Bing and Google to be sure you get your site clicked on by the AI agent.

2. Optimize for AI-Driven Product Discovery

From what we observed, the AI shopping agent prioritized efficiency. There was no browsing around like humans do—it looked for clear signals that guided it to the best product match.

In order to optimize for this scenario, brands should focus on the following:

  • Easy to find on-site search – that’s probably where the agent will go first if there’s no clear category labels. 
  • Clear labeling on navigation menus – if an agent sees what it wants right off the bat in the navigation (e.g. “shoes”) – it will go there. 
  • Robust filtering and sorting – we can’t stress this enough – the agents relied very heavily on sorting and filtering to find what they were looking for, vs scrolling and browsing

The good thing about everything above is these already align to the best practices your store should already be following. It’s just that they take on an extra layer of importance for the shopping style of an AI agent. You’re essentially creating a website for two top level personas now – the human that will browse and potentially meander through the site more, and the AI agent that will very directly interact with search and filtering to get to a destination that matches specific criteria. 

3. Focus on Accessibility for AI Agents

AI shopping agents interact with websites through buttons, menus, and text fields—just like humans. If your website has poor accessibility, AI may struggle to navigate it. 

For example – if low text contrast on buttons makes them hard to read, the AI agent might not be able to click the right buttons, since it is going off of an actual picture of the website, not the code of the website. Following the Web Content Accessibility Guidelines (WCAG) gives your website that much better of a chance of being easy to understand for AI agents.

You should already strive to make your website ADA compliant, but this adds yet another incentive to do so. An AI agent will not have nearly the level of comprehension as a human will either, and so the simpler and more explicit you can make things to navigate, which also aligns to the goal of creating an accessible website, the better it will be. 

4. Build Trust with AI Shopping Agents and Consumers

AI shopping assistants prioritize security and reliability. If your brand lacks strong customer reviews, verified seller badges, or transparent return policies, AI agents may recommend competitors instead if they’ve been instructed to do comparisons across multiple sites.

There’s a couple of simple steps you can take to ensure you don’t get let behind here, and these also typically have the benefit of boosting your conversion rate with humans too:

  • Display trust signals like SSL certificates, secure checkout badges, and transparent return policies.
  • Prioritize positive customer reviews and social proof to increase your ranking in AI-generated shopping recommendations.
  • Maintain a clear and transparent pricing strategy to prevent AI agents from skipping your listings due to hidden fees.

AI Shopping Agents Are Here—Are You Ready?

The rise of AI agents like OpenAI Operator is changing the ecommerce landscape in ways we’ve never seen before. Consumers are starting to outsource their shopping decisions to AI assistants, making it crucial for brands to adapt. 

By optimizing your ecommerce site to show in branded search, focusing on clear product discoverability, creating an experience that’s accessible, and showing trust signals, you can ensure the AI agents that start shopping for consumers don’t pass you by. Want to see how well your ecommerce site is prepared for AI shopping agents? Let’s talk. We can help you future proof your digital presence and ensure AI shopping assistants choose your brand first. 🚀

Keep Reading

Why It’s Good for Your Company to Stand for Something

Do I Need a CRM System?

How Free, Fast, and Sustainable Delivery Can Boost Your Ecommerce Business

Ready to talk ecommerce goals?

Schedule a consultation with an ecommerce expert.