What is agent-based commerce? How will artificial intelligence affect e-commerce?
This unfortunate-sounding term refers to making purchases that are “assisted” by AI purchasing assistants (AI agents). What is it supposed to look like from the shopper’s point of view? Ultimately, he will no longer have to browse dozens of store pages in search of the perfect products, but an agent will do it for him, based on an appropriately formulated query. For sellers, however, they may end up having to redesign the shopping path – to make it suitable not (just) for users, but for AI agents.
The Internet is full of visions describing how AI agents can change the lives of consumers, online store owners and more. The key word here, however, is may. This is because at this stage it is difficult to predict which way agent commerce will go.
Two things are certain. First: the potential of this extremely modern form of shopping is gigantic. Second: the changes that will take place will be acutely felt by both sellers (the cost of adapting stores to the new reality will be enormous) and buyers (ads in the ChatGPT application are already there, and it will get even more interesting). Join us for a wide-ranging discussion of the topic!
OpenAI and Perpexity offensive
Let’s start with a still relatively recent development. In November 2025, both OpenAI and Perplexity launched AI shopping assistants and made them available to users. The announcements, as usual, were very enthusiastic. And what is the reality? We conducted brief tests of both functions.
Shopping research at ChatGPT
OpenAI has unveiled Shopping Research, a tool that allows the user to enter a query about a product (name, technical specifications, price range). In response, the AI assistant retrieves and analyzes information available on the Internet. It compares reviews, offers and technical data, and then prepares a summary, which makes it easy to choose the best offer and make a purchase.
So much for theory. In practice, Shopping Research is an extremely imperfect solution that needs not so much polishing as improving its basic functions. Although ChatGPT claims to search websites for deals, it repeatedly returns incorrect results – product information without links, incorrect links (for example, leading to 404 pages or the manufacturer’s homepage), or directs to products that have been discontinued.
As it stands now (as of December 2025), using Shopping Research is pointless. The function simply does not work as it should, and trying to achieve the expected result resembles a conversation not with an assistant, but with an average-smart two-year-old.
Important!
ChatGPT is gradually introducing the Shopping Research function in a form with a graphical interface, making it easier to use. Unfortunately, the changes are being introduced in stages, so we were not able to test this variant as it was not available in a test account.
It is possible that in the US market, the Shopping Research function works better (that is, at all). OpenAI boasts an Instant Checkout feature, which allows you to finalize a purchase directly in a conversation window with a bot in the ChatGPT app. However, after what I saw while testing the current version of Shopping Research, my expectations have been drastically revised – so I’m not setting myself up for a revolution. Well, it’s no secret that these kinds of features are optimized first with the most promising and profitable markets in mind, of which the Central Europe area is unlikely to be one. Maybe someday.
Does Shopping Research make sense for shoppers?
Contrary to the tone with which I described my experience with this feature… yes. Shopping Research Today doesn’t offer what it promises, but it can collect and condense data on the product category you’re looking for, thus shortening tedious research.
Unfortunately, based on multiple sources, ChatGPT often prepares summaries in which the details of the technical data differ significantly. What I’m getting at is that it ‘s still best practice to independently verify the results presented by the AI assistant. In my tests, there have been times when the assistant “forgot” one of the three criteria, resulting in proposing me a product – for example – that is unsuitable for the size of the room I clearly indicated in the prompt.
Of course, when searching on our own, we also have to verify, but I expected the artificial-intelligent assistant to show higher efficiency under these conditions.
Buy with Pro – online shopping at Perplexity
Perplexity launched the Buy with Pro feature at the end of 2024, but at the time it could only be used by Pro plan users living in the United States. Today, however, not much has changed in this regard – although the developers assure that the service will also be made available in other countries in the future, you can still use its full potential only in the US.
The crux of the Buy with Pro feature is the ability to make a purchase directly in the Perplexity app or chat window. How it works.
Step 1: the user enters a query, for example, “humidifier for a room of 25m2, budget up to £1500, quiet for the bedroom, easy cleaning.”
Step 2: Perplexity generates a list of models that meet the criteria, along with price, a brief description and links to stores. That’s where the Buy with Pro functionality ends outside the US. Indeed, users who are in the United States may notice a “Buy with Pro” button next to many of the products presented.
Step 3: The user selects the “Buy with Pro” option next to the product he is interested in, and then – still at the Perplexity app level – he receives an order summary panel. In it, he will find the product name, price, shipping costs, store details, shipping address and selected payment method. Perplexity retrieves the user data from his account, so the user must first save it in his profile.
Step 4: After checking the summary, the user confirms the order and waits for the ordered product.
As with Shopping Research, the Buy with Pro feature will be made available in European countries “in the future,” but Perplexity does not provide any dates or roadmap. So we are not in a position to assess the priority of launching a fully functional AI shopping assistant outside the US. It remains to wait, and until then, shop on your own.
What’s next for agent-based commerce? How are AI shopping assistants changing commerce?
Purchases made with the support of AI agents have a future, but not a present. This is important news for consumers, but also for sellers.
The former will learn that using chat rooms for shopping purposes is still wandering around a big city full of signposts, each claiming to lead to a destination.
The latter, in turn, should use this knowledge to prepare for doing business in the new reality. For there is no doubt that it will change – one way or another.
AI shopping assistants from a consumer’s perspective – advantages and disadvantages
How might consumers look at all this? From their point of view, much will change, but not necessarily for the better. There are plenty of reasons to be interested, less to be delighted.
How can AI assistants help customers?
Consumers are already using assistants in a shopping context. With them, they can:
- Save time – in theory, all you have to do is enter the type of product you’re interested in and a few parameters, and you’ll get a short list in response, from which it will be easier to pick a winner. This may or may not work. In many cases, nuances are decisive (for example, when buying a TV or a car), which assistants may have trouble analyzing.
- Learn more about products – AI assistants will work well during hectic shopping periods, when we are looking for many different products that we… often don’t know about. Delving into all the technical data can be a road to nowhere. The assistant can instead prepare a brief product description, compare the selected ones and highlight the differences between them.
- Take advantage of recommendations – an AI assistant (especially one accessible from an online store) can take into account not only the current inquiry, but also the customer’s past conversations and purchases. In this way, it can build a profile of the consumer and recommend products tailored to their needs and buying style.
- Ask 24/7 – a big, perhaps the biggest plus of AI assistants is their availability. Leaving aside the issues of potential server failures, which of course do happen, AI assistants can be used at night as well as on weekends or holidays.
What should consumers pay attention to?
It is said that there is no rose without thorns, but in this case we are almost talking about a minefield. Using AI assistants can help, but it can also let the user into the raspberry – a lot depends on the consumer’ s ability to use the assistants. What are the disadvantages and risks?
- Poor data quality – AI assistants today can support research, but they can’t verify the data provided to customers – is a task for the consumer. AI’s notorious hallucinations are an ever-present theme, although technology developers maintain that this is happening less and less frequently. In our tests, the assistant repeatedly directed us to 404 pages, store homepages or to product listings that have long since been withdrawn from sale.
- Lack of ability to finalize a purchase from an assistant – what corporations offering AI tools so eagerly boast about, i.e. the ability to shop directly “with an assistant,” is really a song of the future. These features are often run-of-the-mill or available only in selected regions (usually the US). As a result, you have to go to the store anyway, and then wade through the entire shopping path.
- Basic “knowledge ” – while there are assistants who can help choose the color and size of clothes before buying, they are more likely to linger over trying to order a table of a certain size and material. In contrast, when buying a computer, AI may not pay attention to seemingly insignificant elements that may matter to an enthusiast (such as the width of the rail in a graphics card or the location of the cooling in a computer case).
- What about privacy? AI assistants process massive amounts of data, including online store user data. The risk of payment data leaks is keeping some consumers awake at night.
In November 2025, Amazon sued Perplexity AI. The e-commerce giant’s list of allegations against Perplexity included hiding bot activity and gaining unauthorized access to customer accounts.
Perplexity AI authorities responded to Amazon’s offensive with an article on their blog. The company rejected the accusations and argues that Amazon issued the demands not for the sake of users, but for the sake of interference in serving ads to customers.
The dispute is ongoing and is sure to develop in 2026.
- Ads in AI? No one needs them, but everyone gets them . This was to be expected, but for a long time it was strangely quiet about it. ChatGPT already runs ads in conversations, and in the future it could become a real advertising pole. This is important because AI assistants, if used by customers regularly, accumulate a great deal of knowledge about them. This will make recommending products that those there want to buy easier than ever before. In this context, AI assistants have almost unlimited room for maneuver, and consumers… can only sigh. Or buy the assistant in a premium package – without ads, once such an option is made available.
Agent trading from the sellers’ perspective
The fact that customers are increasingly using AI assistants is extremely important news for the entire e-commerce industry. The conversational style of online shopping is changing the way consumers think about products, but more importantly, it is limiting their direct interaction with stores.
Of course, all this does not mean that users should be forgotten in the context of content design in online stores and product cards – on the contrary, these sites should still be tailored to the needs and expectations of consumers. For a long time to come, a large portion of customers will still – even if supported by AI – finalize purchases in the environment of the store’s website.
Nevertheless, when running an online store, it is worth considering that the content present on the site (especially product descriptions and technical data) should be readable and clear to AI assistants. At the same time, it is important to remember that AI tools do not browse pages in the sense that users do – instead, they break down the content into structured data, interpret it, and then process it to generate a response for the user based on it.
Three pillars of preparation
Wanting to prepare your store for “visits” from AI agents, you should focus on specific actions. One suggestion is to divide into three segments:
- Creating a product database with an open API,
- treating the data as if it were a language tailored for conversations with chatbots,
- place special emphasis on organizing and completing the data.
This last point is particularly important because AI tools usually recommend those products whose descriptions are as complete as possible. Missing data and a cluttered structure will cause AI assistants to omit an offer from their recommendations.
Summary, or how not to fall behind?
Companies that choose to wait and passively watch the situation unfold may fall behind the competition. While it’s easy to conclude that the biggest piece of the pie will be grabbed for themselves anyway by giants such as Amazon and Allegro, this doesn’t mean that smaller stores and companies have no say.
By systematically optimizing data, ensuring that the available information is up-to-date and that customers have good feedback, small companies can systematically appear in the recommendations of AI agents. However, it is not a one-time spurt that is needed, but a methodical, long-term action that is worth adding to the list of permanent duties in the company.
With all of this, it ‘s impossible to forget that at the end of every inquiry is the customer anyway – so it’s important to build accessibility for AI agents, but at the same time continue to emphasize delivering quality products and services, fast, hassle-free delivery, safe packaging and building customer relationships. Deficiencies in these areas will not be fixed by even the best AI assistant.
