Online shopping is moving beyond the traditional search box. Customers can now ask AI tools to find a gift within a budget, compare two products, check whether an item can arrive before Friday, or identify a retailer with simple returns. In some cases, an AI shopping experience can take the customer all the way to checkout.
For UK small businesses, the practical implication is straightforward: your shop needs to be understandable not only to people, but also to systems that collect, compare and recommend product information. AI shopping agents cannot reliably interpret vague claims, outdated stock messages or information hidden behind confusing page journeys. They need clear, consistent facts.
This is not a reason to chase every new platform or rebuild a working website overnight. It is a reason to improve the commercial data that already supports search, shopping feeds, marketplaces and your own storefront. The businesses best placed for AI-driven discovery will usually be those that make it easy to answer basic buyer questions accurately: what is this product, who is it for, how much does it cost, is it available, when will it arrive and what happens if it needs to be returned?
OpenAI’s current shopping guidance says product and merchant results can be informed by product metadata from third-party providers or merchants directly, while merchant ranking can take account of availability, price, quality and whether the seller is the maker or primary seller. It also notes that product data can be supplied through a direct feed. That makes trustworthy retail information an operational priority, not simply an SEO task. Read OpenAI’s shopping guidance.
What AI shopping agents need from an online retailer
An AI shopping agent is software that helps a customer research, compare, recommend or purchase products. It may work inside a chatbot, a search engine, a browser assistant, a marketplace or a retailer’s own service. The technology is developing quickly, but the core inputs are remarkably familiar.
Think of an agent as an exceptionally literal shopping assistant. If a shopper asks for a “machine-washable navy throw under £60, in stock, delivered to Manchester this week, with easy returns”, the agent needs structured, current evidence for every part of that request. A beautiful lifestyle image and a clever product name are useful, but they are not enough.
Start with a reliable product record
Every sellable variation should have its own dependable product record. That means a clear product title, descriptive copy, brand, SKU, barcode or GTIN where one exists, condition, dimensions, material, colour, size, compatibility details and high-quality images. If a customer can choose different sizes or colours with different prices or stock levels, those differences must be explicit rather than implied.
For example, “Luna travel mug” is too broad if the 350ml cream version is sold out but the 500ml black version is available. An agent should be able to distinguish the variants, their images, their price and their availability. Otherwise it may recommend an option that cannot be bought, or omit a relevant option altogether.
Use language customers use, but avoid keyword stuffing and unsupported superlatives. “Water-resistant recycled polyester daypack with 18-litre capacity and padded laptop sleeve for devices up to 14 inches” is more useful than “the ultimate everyday bag”. The first statement gives a buyer and a machine concrete attributes to compare.
Where manufacturer identifiers are available, include them accurately. Google’s free product listings, for example, require key fields such as title, landing-page link, image, price, description and availability, and require GTINs for products that have them. See Google’s free listings product-data requirements. Even if Google is not your main sales channel, this is a sensible benchmark for a complete product catalogue.
Make price a fact, not a surprise
Price is one of the strongest signals in a shopping decision, so it must agree everywhere: product page, variant selector, structured data, feed, basket and checkout. Show the currency clearly. Make sale prices understandable by displaying the usual price, current selling price and, where appropriate, the dates the promotion applies.
Do not rely on “see price in basket”, unexpected handling charges or a discount that only appears after an unclear condition is met. Aside from damaging trust, inconsistency makes it harder for comparison systems to establish which offer is real. Google’s Merchant Center guidance requires submitted prices to match the landing page and checkout, and says additional merchant charges should be included in the shipping information supplied. Review Google’s price requirements.
If you run frequent promotions, establish one source of truth for prices. Your ecommerce platform, product information management system, point-of-sale system and feed should not each contain an independently edited number. Schedule promotion start and end times carefully, and run a spot check as soon as a sale launches.
Publish truthful stock and delivery promises
Availability is not merely an “in stock” badge. It is a promise about whether the selected product can be ordered now. Use meaningful statuses such as in stock, out of stock, pre-order and back-order, and show an expected availability date where it matters. Never leave a discontinued or unavailable variant technically purchasable just to preserve a product page’s conversion rate.
Delivery information is equally important because many AI-led shopping questions contain a deadline. State the dispatch cut-off, delivery services, destinations, estimated delivery window, cost, free-delivery threshold, exclusions and any click-and-collect option in plain English. Put the essentials close to the purchase decision and maintain a complete delivery page for detail.
A homeware retailer, for instance, should not say only “fast UK delivery”. It should say something usable: “Orders placed before 2pm Monday to Friday are dispatched the same working day. Standard UK delivery is £3.95 and normally takes two to three working days. Free standard delivery applies to orders over £50.” The precise promise should reflect your real operations, not an aspiration.
Google supports structured information for shipping and returns alongside product data, and its documentation maps fields such as price, availability, shipping details and merchant return policies to schema.org properties. Explore Google’s supported structured-data attributes. This helps systems retrieve consistent offer information rather than guessing from page text.
Use structured data and feeds to make facts machine-readable
Visible page copy remains vital for customers. Structured data and product feeds make the same information easier for machines to identify and process at scale. They are not a magic ranking switch; they are a way to reduce ambiguity and avoid mismatches.
Ask your developer or ecommerce partner to implement Product and Offer structured data in the initial HTML of product pages, using JSON-LD. At a minimum, the data should match the page’s product name, description, image, SKU or GTIN, price, currency, condition and availability. Add shipping and returns information where your platform supports it. Do not mark up information that the customer cannot see or that is no longer true.
Google specifically advises that structured data must match the values shown to users and says its automatic item updates use fields including price, price currency, availability and condition. Follow Google’s structured-data implementation guidance. Test representative pages after deployment, especially variants, sale items and products that regularly move in and out of stock.
Next, maintain a product feed for the channels that matter to your business. That may include Google Merchant Center, your marketplace listings and platform-native catalogues. Shopify merchants, for example, are already integrated into ChatGPT shopping through Shopify Catalog according to OpenAI’s guidance, while other merchants can seek access to direct product feeds. The lesson is not to duplicate work blindly; it is to ensure your underlying catalogue is complete enough to travel accurately between channels.
Third-party signals help agents assess trust
AI shopping agents may draw on more than your own website. Reviews, ratings, reputable editorial coverage, marketplace performance, business identity and customer-service signals can all shape how a product or merchant is understood. OpenAI says its shopping experience may show review summaries based on public websites and that product labels can draw on information available to the model, including third-party data. See how ChatGPT describes reviews, labels and merchant selection.
That does not mean you can buy your way into a recommendation, and it certainly does not justify manufactured reviews. It means your public reputation should match the experience you deliver. Claim and maintain your business profiles. Keep your company name, contact details and trading address consistent. Respond professionally to genuine reviews. Correct inaccurate listings. Make it obvious who is behind the shop, how customers can get help and how to contact you before and after purchase.
For own-brand products, explain why you are the maker or authorised seller. Include origin, materials, care instructions, warranty information and certifications where relevant. For resale products, identify the exact model and condition. This reduces the risk of an agent confusing your listing with a similar item from another seller.
Returns and policies are recommendation data
A return policy is not fine print. For a shopper comparing near-identical offers, a clear return window, return cost and refund process can be decisive. Publish a dedicated policy page, link it in the footer and from product pages, and use straightforward terms. Specify which products are excluded, how customers start a return, whether they pay return postage, the condition required, the refund timetable and how exchanges work.
UK distance-selling rules give consumers a cancellation period for many goods bought online, subject to exemptions and conditions. Government guidance explains that consumers generally have 14 calendar days to change their mind and, in many cases, then have 14 days to return goods. Read the UK Government’s Consumer Contracts Regulations guidance. Get legal advice if your product category, subscription model or policy wording is complex.
Clarity matters commercially as well as legally. OpenAI’s commerce policies prohibit misleading practices, including misrepresenting price, availability or key product characteristics, and using unclear or unfair terms or misleading refund and return policies. Read OpenAI’s commerce policies.
Legitimate shopping agents versus harmful bot traffic
Not every automated visit is a customer-serving shopping agent, and not every bot should be welcomed. Treating all automation as either good or bad is a mistake.
Legitimate agents and crawlers may access public product pages to discover information, keep search or shopping listings current, compare public offers, or act on a customer’s request. They should have a declared user agent, behave within sensible request volumes, access pages that a normal customer can view and follow your published access rules where applicable. Their activity may support visibility, but it can still create load, so monitor it.
Harmful bot traffic is designed to exploit rather than assist. Warning signs include rapid repeated requests that strain the site, scraping of protected content, attempts to test stolen cards, checkout abuse, credential stuffing, fake account creation, inventory hoarding, unauthorised price harvesting, malformed requests and attempts to evade rate limits. These visitors often rotate IP addresses or imitate legitimate user-agent strings.
Do not trust a user-agent label by itself. Google warns that Googlebot’s user agent is often spoofed and recommends verifying suspicious requests through reverse DNS lookup or its published IP ranges. Read Google’s crawler-verification guidance. Apply the same principle to any service claiming to be an AI agent: check its published documentation, verify technical identity where possible, and ask your hosting, CDN or security provider for help if the traffic is costly or suspicious.
The practical response is layered. Allow public product and policy pages to be discovered if visibility is a goal. Protect login, account, basket, checkout, search and payment endpoints with rate limits, bot management, web application firewall rules and fraud controls. Separate research traffic from transactional actions: an agent reading a product page is not the same as an automated process repeatedly attempting payments. Log blocked activity, review false positives and give your team a clear escalation path.
A simple AI shopping readiness audit
Use this checklist monthly, and every time you change platforms, suppliers, fulfilment partners or promotion rules.
- Product identity: Does every variant have an accurate title, description, image, SKU and GTIN or MPN where applicable?
- Attributes: Are size, colour, material, dimensions, compatibility, condition and care information present where relevant?
- Price: Do product page, feed, basket and checkout show the same current price, currency and promotion terms?
- Availability: Is stock status accurate for every variant, with honest pre-order or back-order dates?
- Delivery: Are costs, cut-offs, destinations, delivery windows and free-shipping thresholds clear before checkout?
- Returns: Is the returns policy easy to find, plain-English and aligned with your legal obligations?
- Structured data: Does Product and Offer markup match what customers can see on the page?
- Feeds: Are your key sales-channel feeds active, validated and refreshed after stock or price changes?
- Trust: Are your business details consistent, reviews authentic and customer-service contact information prominent?
- Bot controls: Can you distinguish verified crawlers from suspicious automation, while protecting account and checkout routes?
Make accuracy your competitive advantage
AI shopping will not replace the fundamentals of retail. It raises their value. Clear product information, dependable fulfilment, fair returns and a credible reputation have always helped customers buy with confidence. The difference is that these details increasingly determine whether an automated assistant can find, compare and confidently recommend your offer in the first place.
Start with your 20 best-selling products. Audit the facts a customer would need to make a decision, correct inconsistencies, add structured data, check your feeds and test the journey from product page to checkout. Then make ownership clear: someone in the business should be responsible for product-data accuracy whenever a price, supplier, stock level or delivery promise changes.
The goal is not to optimise for a mysterious algorithm. It is to become the retailer whose information is clear enough for customers, search services and legitimate shopping agents to trust. Make that standard part of everyday ecommerce operations, and your small business will be much better prepared for the next way people choose to shop.
