A shopper asks Google for a washable rug for a busy dining room, under AED 800, delivered to Dubai Marina before the weekend.
That is not a simple product search. The shopper has given Google a budget, a use case, a location, a deadline, and a preference. They may also ask for something easy to clean, suitable for children, or close in colour to a photograph they uploaded.
An AI shopping agent has to sort through all of that. It needs to understand the products, compare them, check whether they are in stock, work out delivery, and eventually help the shopper buy. This is why product descriptions alone are not enough anymore. A polished paragraph cannot fix a price that changes at checkout or a stock status that has been wrong since Tuesday.
Google launched the Universal Commerce Protocol, or UCP, in January 2026 as an open standard for agentic commerce. Google says it is designed to cover the shopping journey from discovery and buying to post-purchase support. Its first checkout feature is for eligible US retailers on Google’s AI surfaces, with wider expansion planned. That does not mean every retailer in the UAE can switch on agentic checkout today.
So the sensible approach is preparation, not theatre. Fix the shop before asking an agent to represent it.
What An AI Shopping Agent Needs To Know
A normal product page can survive a little vagueness. A human visitor may send a WhatsApp message and ask whether the blue version is available in medium. An agent cannot rely on a reply that might arrive tomorrow.
It needs clear information about the product, its variants, price, availability, shipping, returns, payment, and any conditions that affect the sale. It also needs a working path to complete the order.
Google’s UCP documentation says retailers remain the Merchant of Record. In ordinary language, the retailer still owns the transaction and the customer relationship. Google is not taking responsibility for whether the wrong size arrives or whether the return policy is honoured.
That is a useful point to keep in mind. Letting an agent help with discovery does not hand the business responsibility for the messy bits.
The Agent Reads More Than The Product Description
Imagine a product page for a dining chair. The description says “premium velvet chair with modern design.” Fine. But the shopper wants to know:
Is the seat height suitable for a standard dining table?
Does the price include one chair or a set of four?
Is the beige colour closer to cream or sand?
Can six chairs arrive before Friday?
What happens if the fabric looks different in daylight?
Some of these answers belong in the product description. Others belong in structured product data, delivery settings, variant fields, policies, or the checkout itself.
The page should not force the customer to solve a small investigation before buying.
Start With Product Data, Not AI Copy
Google uses product data to match products with relevant searches and support its shopping formats. Missing or conflicting information can lead to disapprovals, limited eligibility, or incorrect displays.
The boring attributes matter. Product identifiers, categories, colours, sizes, item group IDs, price, availability, images, shipping, and returns all affect how a product can be understood.
This is where many stores get caught. The Merchant Center feed says “in stock.” The product page says “ships in 5 to 7 days.” Checkout says delivery is unavailable in the customer’s area. The agent has not failed. The shop has given it three different answers.
Keep The Feed And Website In Step
Google’s guidance says the submitted price and availability should match the landing page, checkout, and structured data. That sounds obvious until a store has several systems updating at different speeds.
Check a few real products manually. Look at the feed, the page, the cart, and checkout. Do not only test the products that nobody buys. Test the popular items, sale items, products with variants, and products that regularly go out of stock.
If you run a growing catalogue, an ecommerce SEO service can help organise the technical work around feeds, category pages, product templates, and internal links. But the store owner still needs to know which data is actually true.
A feed cannot decide whether “available” means sitting in your warehouse, available from a supplier, or ready to deliver tomorrow. Someone has to define that.
Product Variants Need Plain Answers
A red shirt in small is not the same item as a red shirt in large. A sofa in 2-seater and 4-seater sizes may have different prices, dimensions, delivery charges, and stock.
Yet some websites show one generic product page and expect shoppers to work out the details from a dropdown.
Use consistent variant information. Make the colour, size, material, dimensions, item group, and price clear. The selected variant should change the image and availability where necessary. If the green version is unavailable, do not leave the page showing a green “in stock” message after the shopper selects it.
A real person dislikes that experience. An agent will have trouble with it too.
Do not make claims that belong to another variant. “Free delivery” might apply to the large model but not the compact one. “Ships today” might apply to black but not ivory. Put the condition next to the relevant choice instead of hiding it in a general delivery page.
Make Prices Boringly Clear
AI shopping agents may compare products by price, but price comparison is only useful when the numbers mean the same thing.
A product appears at AED 249 in a listing. At checkout, the customer sees AED 249 plus a mandatory handling charge. Another store shows AED 279 with delivery included. Which one is cheaper? The answer depends on what the prices include.
Google’s Merchant Center guidance requires submitted prices to match the landing page and checkout. For countries outside the US and Canada, VAT or GST treatment also needs to follow the applicable requirements. UAE retailers should make the currency obvious and explain delivery or other unavoidable charges before the final payment step.
Sale prices need care as well. If a discount ended last week, remove it from the feed and the page. If the “from” price only applies to the smallest size, say so near the product choice.
I would rather see a store show a slightly higher but honest delivered price than advertise the cheapest number and surprise people later. The second approach may win a click. It tends to lose the customer.
Shipping And Returns Are Part Of The Product
For an agent, delivery is not a footer detail. It can decide whether the product fits the request.
A shopper may want a desk delivered to Abu Dhabi by Thursday. Another may need a replacement filter delivered to a remote area. A third may be willing to wait three weeks for a lower price.
The website should make the delivery area, cost, time estimate, and restrictions easy to find. If the time changes by emirate or product size, explain that without making the customer open six pages.
Returns need the same treatment. State the return window, exclusions, collection process, refund timing, and condition requirements. “Easy returns” is not a policy. It is a slogan.
The same applies to warranties. Put the coverage and duration where the customer can see them. A support agent should not need to search through an old PDF to find out whether a damaged part is covered.
Your Checkout Has To Work Without A Human Rescue
Google’s UCP is designed to enable agentic actions on Google AI surfaces, starting with direct buying for eligible retailers. Google says retailers can choose integration paths and remain responsible for the transaction.
That does not mean every shop should rush into a new protocol. First, test the checkout you already have.
Try it on a phone using a normal mobile connection. Add a product with a variant. Change the quantity. Apply a discount. Enter a UAE address. Check shipping. Remove the item. Start again. Then ask someone who did not build the store to complete the purchase.
Watch where they hesitate.
A Shopify web development company in the UAE can help with the technical connection, theme, checkout logic, and product structure. But no integration removes the need for accurate fulfilment behind it.
If the warehouse takes two days to confirm stock, do not present the product as ready for same-day dispatch. If the payment gateway sometimes fails, give the customer a useful message instead of a blank error screen.
Prepare The Support Side As Well
Google announced Business Agent, a branded AI agent that can answer product questions in a retailer’s voice on Search. The initial availability is for eligible US retailers, and Google says more capabilities are planned, including direct purchases and related offers.
The quality of those answers will depend on the business information behind them.
If the store has ten versions of the warranty in different folders, the agent may not know which one to use. If the customer support team gives a different answer from the website, a branded voice will not make the situation better.
Write down the questions customers ask repeatedly. For a furniture store, they might ask whether the item fits in a lift. For a beauty retailer, they may ask whether a product suits sensitive skin. For an electronics shop, compatibility and warranty may matter more than another paragraph about innovation.
Answer those questions on the product page or in a clear support section. A conversion rate optimisation process in Dubai should include these moments because unanswered questions often cause abandoned carts.
Be Careful With AI-Generated Product Descriptions
There is nothing wrong with using AI to help organise a catalogue or produce a first draft. The trouble starts when every product receives the same soft paragraph.
“Elevate your lifestyle with premium quality and timeless elegance” tells a shopper almost nothing. It does not say whether the item scratches easily, fits a narrow hallway, needs assembly, or survives a humid balcony.
Add details that come from the product and the people who sell it. Mention the awkward delivery issue. Explain the difference between two materials. Say that the white finish shows fingerprints. Tell customers the shoe runs narrow if that is what the returns team keeps seeing.
Those details are useful because somebody learned them through actual sales, support calls, returns, or product testing. They also make the catalogue less interchangeable.
A store can use content optimisation to improve product pages, but the source material should come from the business. AI can rearrange knowledge. It cannot honestly invent a warehouse experience the team has never had.
What To Check Before You Explore UCP
Start with a small group of important products. Do not begin with ten thousand SKUs and a large presentation about the future of shopping.
Check one best seller, one sale item, one product with variants, one low-stock item, and one product with unusual delivery rules. Compare the information in Merchant Center, the product page, structured data, cart, checkout, shipping policy, and returns policy.
Look for these uncomfortable gaps:
Different prices in the feed and checkout
Stock that changes only once a day
Variant information hidden in images
Delivery promises that do not apply to every emirate
Returns language that sounds reassuring but says nothing
Product titles that omit the model, size, or main use
Old PDFs or help pages with expired terms
Fix those before adding another AI tool.
Google’s UCP documentation also mentions future capabilities such as multi-item carts, loyalty account linking, tracking, and returns support. Those features may be useful, but they depend on clean commerce operations. A store that cannot tell a customer where an existing order is will not become organised simply because an agent can ask the question faster.
The 2026 Shopping Agent Reality
AI shopping is moving from product discovery towards actions such as comparing, answering questions, and completing purchases. Google’s current UCP checkout announcement begins with eligible US retailers, while the company says it plans to expand globally and add more capabilities.
That is enough to take the change seriously. It is not enough to claim that every website can offer agentic checkout today.
For most retailers, the best preparation is not a shiny AI badge. It is a product catalogue that tells the truth, a checkout that works, a delivery promise the warehouse can keep, and customer support that knows the answer.
If an agent sends more people to the store, those details will be tested more often. Better to fix them before the traffic arrives.
FAQs
1. Is Google’s AI checkout available to UAE retailers?
Not as a universal feature. Google’s current Merchant help describes eligibility for products and participating merchants or partners in the US, Canada, and Australia. UAE retailers should not advertise Google agentic checkout as available unless their account has received clear eligibility and the complete purchase flow has been tested.
2. Do I need UCP to appear in Google AI shopping results?
No. UCP is the protocol for supported agentic commerce actions, including direct buying. A retailer can still appear through product listings and Merchant Center without implementing UCP. Accurate product data remains the first job. UCP becomes relevant when the business wants to support the eligible checkout or other agentic actions.
3. What product information should I fix first?
Start with price, stock, variants, product identifiers, shipping, returns, and the landing page. Make sure the feed, product page, structured data, cart, and checkout agree. A perfect product title will not help if the green size medium is shown as available when the warehouse has none.
4. Will Google handle the customer relationship after an agentic purchase?
No. Google says the retailer remains the Merchant of Record. The retailer still owns the transaction, customer relationship, fulfilment, returns, support, and post-purchase experience. If a delivery is late or the wrong item arrives, the business still has to resolve it.
5. Do I need to rebuild my website for AI shopping agents?
Usually not. First improve the catalogue and checkout you already have. Use readable product pages, accurate structured data, clear delivery and return policies, stable feeds, and a checkout that works on mobile. A new integration cannot compensate for a broken payment step or unreliable stock information.
