For about twenty years, shopping online has followed the same routine: open a website, search, compare, read reviews, click "buy". Now software can do most of those steps for you. This week Amazon began letting a small group of invited US customers text an AI shopping agent. A few weeks earlier it had blocked a rival company's agent from its store, and it has spent months in court over a third. The question underneath all of this is simple and important: when a machine shops on your behalf, who is in charge of the customer, the shopper, the store, or the agent? This guide explains what happened, how shopping agents work, and what buyers and sellers should do about it.
Key Takeaways
- Amazon has started a text-message shopping agent for a small group of invited US customers, with wider access promised.
- The same company blocked Meta's Muse agent in September, and fought Perplexity's Comet agent in court with mixed results.
- A shopping agent searches, compares and sometimes buys for you. The more it acts alone, the more you need rules and limits.
- Buyers should keep approval switched on, check what data is stored, and compare suggestions with a normal search.
- Sellers should make product data clean and complete, and avoid depending on a single marketplace.
- The legal and business rules are not settled. Expect more changes in the next year.
What Is an AI Shopping Agent?
A shopping agent is software that takes a request in plain language and then carries out the steps needed to meet it. You might say, "I need a waterproof jacket for hiking, under a price I set, delivered before the weekend." A normal search engine would give you a list of links and leave the work to you. An agent goes further. It searches several sources, filters out items that do not match, compares prices and delivery times, and returns a short list. Some agents can also place the order.
Three levels are worth separating, because the risks are different at each:
- Assistant. It answers questions and suggests products. You do all the clicking and paying.
- Agent with approval. It prepares the purchase and asks you to confirm before it spends money.
- Agent on autopilot. It buys by itself when a condition is met, such as a price falling to a target you set.
Most people will be comfortable with the first two. The third is useful for repeat purchases and price targets, but it is the level where mistakes cost real money.
What Happened This Week
Amazon announced a text-message version of its Alexa shopping assistant. According to The Next Web, customers who opt in get a contact called "Alexa Shopping". They can text a request, such as needing more dog food, or send a photo of a recipe or an outfit, and the agent replies with options. It can buy with the customer's approval, or automatically when a price reaches a target the customer sets. It can also track price drops, reorder past purchases and check on packages. Conversations started by text carry over to Amazon's app, its website and Alexa devices.
The rollout starts with a small group of invited US customers, and Amazon says millions more will get access in the coming weeks. The same report says sponsored products can appear among the agent's options. That detail matters, and we come back to it below.
The Fight Over Who Is Allowed to Shop for You
Amazon's own agent is only half the story. The other half is how Amazon treats agents built by other companies.
The same article notes that Amazon blocked Meta's Muse shopping agent from its site in September. It gives no further detail, so we should not guess at the reasons.
Earlier, Amazon went to court against Perplexity over Comet, its browser assistant that can shop. A lower court granted Amazon an injunction in March. Later, an appeals court overturned it. The court found that Amazon was unlikely to prove that Perplexity had broken federal or California computer access laws, because the three-judge panel saw Amazon's customers, not Perplexity, as the ones accessing Amazon's systems. The case went back to the lower court for further proceedings.
This is an early test of a question that will matter for years. When an AI agent acts for a customer, is it the customer using the site, or a third-party bot? The first answer favours open access for agents. The second favours stores that want to keep control. Courts are still sorting it out, and different courts could reach different answers.
Why Platforms Care So Much
It helps to see why a large store might welcome its own agent but resist outside ones. Several business reasons are plausible, although the companies involved have not all stated them:
- Control of the customer relationship. If an outside agent does the shopping, the store becomes a warehouse behind someone else's screen. The store loses the chance to show its own recommendations and offers.
- Advertising income. Many stores earn money from sponsored listings. An outside agent may skip them. An agent run by the store can include them. The Next Web reports that sponsored products can appear among Alexa's options.
- Data. Search and purchase behaviour is valuable. A store usually prefers to keep it.
- Security and fraud. Automated buying can be abused. Stores have real reasons to want to know who, or what, is placing an order.
- Service quality. If an agent makes a mistake, the customer may blame the store.
None of these reasons is unreasonable. But together they explain why the same company can launch its own agent and block others in the same season.
How a Shopping Agent Works, in Plain Terms
You do not need to understand the technology to use an agent, but a basic picture helps you spot where things can go wrong. Most shopping agents follow roughly the same loop:
- Understand the request. A language model reads your message, or looks at your photo, and works out what you want and what limits you set.
- Find products. The agent searches a store's catalogue, or several sources, using product data such as titles, prices, specifications and stock.
- Filter and rank. It removes poor matches and orders the rest. This is where hidden choices enter. The ranking may reflect your needs, the store's profit, advertising, or a mix.
- Present options. You see a short list instead of hundreds of results.
- Act. With your approval, or on a rule you set, the agent places the order, tracks it and handles follow-up.
Every step can fail in a different way. The request can be misunderstood. The product data can be wrong or out of date. The ranking can be skewed. The final action can happen too easily. When you judge an agent, ask which step you are trusting it with.
What This Means for Buyers
The good news
- Less effort. You describe what you want once, and the agent does the searching and comparing.
- Better follow-up. Tracking packages, watching prices and reordering are tasks software does well.
- Access. A text message is easier than an app for many people, including older users and people with limited time.
The things to watch
- Narrower choices. An agent shows a short list. If you never see the other options, you cannot tell what you missed.
- Hidden influence. If sponsored items can appear among the options, you should know which results are paid. Check whether the agent labels them.
- Automatic purchases. A rule that buys when a price drops is helpful until you no longer want the product, or the price drop is a mistake.
- Data. Past orders, photos and preferences become inputs. Photos of your home, your clothes or your family can be sensitive.
- Disputes. If an agent orders the wrong item, who is responsible: you, the store or the agent's maker? Read the return and refund terms before you rely on it.
A Simple Safety Checklist for Shoppers
- Keep approval before purchase switched on, especially at the start.
- Set a spending limit if the agent allows one.
- Check which results are sponsored, and how the agent says it ranks products.
- For any large purchase, repeat the search yourself in a normal browser.
- Review the agent's order history every week for the first month.
- Find out what data the agent stores, and how to delete it.
- Avoid sending photos that show private information.
- Keep proof of what the agent said it would buy, in case you need a refund.
What This Means for Sellers and Small Shops
For a small online business, shopping agents are both a chance and a risk. The chance is that an agent can send buyers to you without a single advertisement. The risk is that the shopper may never see your website, your brand or your story, and that the store running the agent can favour itself.
1. Your product data becomes your storefront. If an agent chooses between products, it relies on what it can read: title, price, size, material, stock, delivery time and returns. A clear, complete listing is easier to match to a request than a vague one. Photos and brand still matter once a buyer sees you, but data decides whether you are shown at all.
2. The platform can favour itself. A marketplace that runs its own agent can promote its own results, include paid listings and limit outside agents. If most of your sales come through one marketplace, you have little say over how an agent treats you.
3. Reputation travels in numbers. Agents are likely to weigh review scores, delivery reliability and return rates. A shop with honest reviews and a good record is better placed than one that relies on a polished banner.
4. Rules are unsettled. The Amazon cases show that platforms and courts are still deciding what outside agents may do. Marketplace terms may change. Read them for any rule on automated or third-party buying tools.
A Practical Checklist for Sellers
Make your product data complete
- Write clear product titles that say what the item is, not only the brand name.
- List size, colour, material, compatibility, weight and other facts a buyer might filter by.
- Keep prices and stock accurate in every place you list them. Out-of-date stock leads to cancelled orders and bad reviews.
- State delivery times and costs plainly.
- Publish a clear returns and refund policy on its own page.
Make your data easy for software to read
- Use structured product data on your own site, so search engines and other software can read price, availability and reviews without guessing.
- If you sell on marketplaces or shopping platforms, keep your product feeds complete and consistent.
- Use the same product names and identifiers across channels, so one product is not treated as three.
Do not depend on one channel
- Keep your own website live, fast and trustworthy.
- Build an email or messaging list of your own customers.
- Sell through at least two channels where you can, so a policy change on one does not end your income.
Watch the rules
- Read each marketplace's terms on third-party tools and automated buying.
- Decide your own position on agents visiting your site, and check what your website's access rules say.
- Keep an eye on the Amazon and Perplexity case. A final ruling could change what stores can and cannot block.
Three Scenarios to Think Through
These are made-up examples to show how the issues play out. They are not real cases.
Scenario 1: The reorder. A family sets an agent to reorder pet food every month. After a year they switch to a different brand but forget the rule. For two months an unwanted bag arrives. Lesson: automatic rules need a regular review, and the agent should remind you.
Scenario 2: The short list. A shopper asks an agent for a laptop bag. The agent shows three options. Two are sponsored. A better, cheaper bag from a smaller seller is not on the list. Lesson: ask how the list is chosen, and run your own search for larger purchases.
Scenario 3: The small seller. A small shop sells handmade leather wallets. Its product pages say only "Classic wallet, brown". A rival lists the same kind of wallet with the leather type, dimensions, card slots and delivery time. When a buyer asks an agent for "a slim leather wallet that fits six cards", the rival's listing matches the request and the first shop's does not. Lesson: complete data is a form of marketing.
The Bigger Picture: Who Owns the Customer?
For years the web has worked on a simple bargain: stores compete for attention, and customers choose. Agents change the bargain. If the customer talks to an agent, the agent becomes the front door. Whoever controls the agent controls what is shown first, what is hidden, and what counts as the best choice.
There are roughly three futures, and the real one will probably mix them:
- Walled gardens. Each large store runs its own agent and blocks others. Customers get convenience inside one store, but cross-store comparison gets harder.
- Open agents. Independent agents shop across stores for the customer, and stores must be ready for software visitors. This favours open access and competition on price and quality.
- Negotiated access. Stores and agent makers sign agreements, with rules on data, advertising and fees. This is likely to be messy at first.
The court decision in the Perplexity case points toward the second future, at least on one legal question. Amazon's launch and its blocking of Meta's agent point toward the first. Both are happening at once, which is why this is worth watching.
How to Judge News About AI Shopping Agents
Expect plenty of headlines in the coming months. A few habits help you separate signal from noise:
- Check whether the story is about a product that people can actually use today, or an announcement of what is planned.
- Ask who gains. A company's own blog describes its strengths. A court filing describes a dispute.
- Look for the number of users, but be careful with figures that come from one source and are not confirmed.
- Notice what is not said. In this story, reasons for blocking Meta's agent are not given in the report, so we should not assume them.
- Check the date. Court cases and platform rules change quickly.
Plain-English Glossary
- Agent: software that takes steps toward a goal for you, instead of only answering questions.
- Agentic commerce: buying and selling where software, not a person, does the searching, comparing and sometimes the paying.
- Sponsored listing: a product that a seller has paid to promote. It may appear among the agent's options.
- Product feed: a file or service that gives a store or search platform your product details, such as title, price, stock and delivery time.
- Structured data: product details written in a standard format on a web page, so software can read them without guessing.
- Injunction: a court order that stops a party from doing something while a case continues.
What to Do This Week
Shoppers: open the settings of any AI shopping tool you use, switch on approval before purchase, and check what data it stores. Sellers: pick your five best-selling products, and make sure each listing has a clear title, full specifications, accurate stock, a delivery time and a returns policy. These small steps cost little and help whichever way the market moves.
Frequently Asked Questions
What is an AI shopping agent?
It is software that takes a plain-language request, searches and compares products, and in some cases places the order for you. It works as a layer between you and the store.
Is Amazon's text shopping agent available to everyone?
Not yet. It started with a small group of invited US customers. Amazon says millions more will get access in the coming weeks.
What can Amazon's agent do?
According to The Next Web, customers can text a request or send a photo, and the agent replies with options. It can buy with approval or when a price target is met, track price drops, reorder past purchases and check on packages.
Why did Amazon block other shopping agents?
Amazon has not given a full public explanation for each case, and the report on Meta's Muse gives no detail. In the Perplexity case, Amazon argued the agent broke computer access laws, but an appeals court was not convinced. Stores also have commercial reasons to control how customers reach them, such as advertising income and customer data.
What did the court decide in the Perplexity case?
A lower court granted Amazon an injunction in March. An appeals court later overturned it, finding Amazon was unlikely to prove Perplexity broke federal or California computer access laws. The case went back to the lower court, so it is not over.
Will shopping agents replace websites?
Probably not entirely. Many people enjoy browsing and comparing for themselves. But for repeat and simple purchases, agents may handle more of the work, so sellers should be ready for both kinds of visitor.
Are automatic purchases safe?
They carry risk if you set a rule and forget it. Use approval before purchase, set spending limits and review your order history regularly.
How can a small seller prepare?
Make your product data complete and accurate, use structured product data on your own site, keep prices and stock consistent across channels, build your own customer list, and avoid depending on one marketplace.
Can an agent show me sponsored results?
It can. The Next Web reports that sponsored products can appear among the options in Amazon's agent. Check whether the agent you use labels paid results, and compare with a normal search for important purchases.
Is my data safe with a shopping agent?
It depends on the company and its settings. Past orders, photos and preferences become inputs, so read what is stored, how long it is kept and how to delete it. Avoid sending photos that show private information.
What should I watch next?
Watch whether Amazon widens access and how it labels sponsored options, what happens in the Perplexity case after the appeal, and whether other large stores launch agents or block outside ones.
Final Thoughts
Shopping agents make buying easier, but they also move power. Whoever runs the agent decides what you see first, what is promoted and who may take part. For buyers, that means keeping control of approvals, spending limits and data. For sellers, it means making products easy for software to understand and not depending on one platform. The courts and the platforms are still deciding the rules, and this week's launch is only one step. Keep your approvals on, keep your data clean, and expect the story to keep changing.
