Your best salesperson is a spreadsheet. Most businesses have not looked at theirs in a year.
At Google I/O in May 2026, Google introduced Universal Cart, a single shopping cart that works across merchants. It is rolling out in the United States first, on Search and the Gemini app, with YouTube and Gmail to follow. Checkout runs through Google's Universal Commerce Protocol, and a separate payments protocol lets AI agents buy for a shopper inside spending limits the shopper sets.
Australia is already on the list. Google's Merchant Center help names products eligible in the United States, Canada and Australia as the three markets for the protocol's checkout feature, limited to participating merchants for now.
Look at what Google built for those agents to read. In January 2026 it announced dozens of new Merchant Center data attributes made for discovery in AI Mode and Gemini, covering things like answers to common product questions and compatible accessories. All of it is product data. Your campaign creative is not on the list.
So your beautifully shot hero video does not get a vote, and neither does your clever headline. The agent reads the title and the attributes. It checks the price, the stock status and your policies. Then it decides.
That is a salesperson's job. For a lot of stores, the salesperson is turning up with the wrong price list and no idea what is in the warehouse.
If you sell products online, your feed is the first thing an AI shopping agent can judge you on. Complete and accurate product data gets you shortlisted. Thin or stale data gets you skipped, however good the product is.
Why the feed matters more now
Feeds have always powered shopping ads. For years that made them an ad ops chore: upload once, fix the disapprovals, forget about it.
What changed is the reader.
A human scanning a shopping carousel is forgiving. They see a nice photo next to a vague title and click to find out more. An agent working for a buyer has no reason to do that. Either it has the information to match your product to the request, or it moves on to a product that does.
Say someone asks for waterproof hiking boots in a wide fit, under $250, delivered before the weekend. Your feed does not say waterproof. It does not list width, and it has no delivery information. You are out of the running, and it does not matter how good the boots are.
Ads persuade people. Feeds qualify products. When a machine does the shortlisting, qualification comes first and persuasion may never get a turn.
We have written before about the protocols that let agents buy. The feed is the layer underneath them. The protocols decide whether an agent can transact with you, and the feed decides whether it wants to.
What a product feed is
If you have never opened yours, a product feed is a structured file with one row per product or variant. It is usually managed through your store platform or a merchant account such as Google Merchant Center. Each row carries fields: an ID, a title, a description, a link, an image, a price, availability, brand, and a long list of optional attributes.
Platforms like Shopify and WooCommerce can generate one automatically through an app or plugin, and that is where the trouble starts. Automatic feeds copy whatever is on your product pages, and product pages are written for humans who can see the photo and fill in the gaps. A machine does not fill in gaps. If a fact is not in a field, it does not exist.
Think of it as the spec sheet your salesperson carries into every meeting. A human shopper might forgive a blank. An agent comparing forty options against a precise request will not.
The feed hygiene checklist
Work through it in order. The early items do the most damage when they are wrong.
Titles written for matching
A product title should read like a precise answer to a request. Lead with brand and product type, then the attribute that sets it apart, then size, colour or material where relevant. Google allows 150 characters and says shoppers usually notice only the first 70 or fewer, so the important words go at the front.
"The Explorer" is a name. "Explorer Men's Waterproof Leather Hiking Boot, Wide Fit, Brown" is a match. Keep the brand voice for your product page.
Attributes filled in completely
Every optional field you leave blank is a question the agent cannot answer. Go through colour, size, material, pattern, age group, product category, condition, and the GTIN or manufacturer part number.
Then add the details customers ask about, like dimensions and compatibility. If buyers keep asking a question in your inbox, the answer belongs in structured data.
Stock status that is true
Nothing burns trust faster than recommending something that cannot be bought. Sync availability as often as your platform allows. When the queen size in oatmeal sells out, the feed should say so before a customer finds out at checkout.
Prices that match the page, every time
The price in your feed has to equal the price on the landing page and at checkout. Include sale prices with their dates. Google's crawler compares the two, and a product with a mismatched price may be disapproved in Merchant Center. An agent has even less reason to recommend a product whose price it cannot trust.
Reviews connected to your merchant account
Star ratings on your product page help the humans who land there. They do nothing for a machine unless they are passed through to Google. The Product Ratings program takes them from your own data source or from a supported reviews aggregator, and you need at least 50 reviews across all your products before ratings show on your listings. Check that yours are flowing and not sitting in a widget on your site. Why that matters beyond the feed is covered in how AI reads your reviews before it recommends you.
Shipping and returns spelled out
Buyers decide on delivery cost and speed, and on how easy a return is. A buyer who wants something by Friday with free returns needs an agent that can check both. Set them in your merchant account. A footer page on your website is not enough.
Images that match the listing
The feed image should show the exact variant listed, with no promotional text over it. Creative may not persuade an agent, but an image that breaks Google's rules will not appear on Google at all, and a wrong image still confuses the human who makes the final call. Accurate product photography earns its keep twice here.
A worked example: fixing one product row
Say you run a small Melbourne homewares store selling linen bedding online. Your best seller is a quilt cover set, and its feed row looks like this in plain English.
Title: "The Sunday Set." Description: two lines about slow mornings. Colour: blank. Material: blank. Size: one row for all sizes. Shipping: not set in the merchant account. Reviews: sitting in a widget on your product page, not connected.
A buyer asks an AI assistant for a queen size, 100% linen quilt cover in a neutral colour, delivered within the week. Your product might be perfect. The agent cannot tell.
Fix it in this order.
- Split the variants. One row per size and colour, each with its own ID, price and stock status.
- Rewrite the title. "Sunday Set 100% French Linen Quilt Cover Set, Queen, Oatmeal." That gives the agent brand, material, product type, size and colour in one line.
- Fill the attributes. Material, colour, size, pattern, product category and GTIN if you have one. Put care instructions in the description, because buyers ask about them.
- Set shipping and returns in the merchant account, with realistic delivery times for each region you ship to.
- Connect reviews so the product rating flows through to the platforms reading your feed.
Nothing about the quilt changed. It now answers the question the buyer asked.
How to run it without a data team
- Start with your top sellers. Fix the products that earn most of your revenue first. The long tail can wait.
- Read your diagnostics. Merchant Center already lists your errors and warnings. Clear the errors, then work through the warnings.
- Ask a machine to shop for you. Describe what a real customer wants to Gemini or ChatGPT and see what gets recommended. Note what information the winning products had that yours lacked.
- Give it an owner. One person checks feed health on a fixed day each week, because feeds decay as prices and stock change.
- Fix it at the source. Correct the data in your store or inventory system as well as the feed tool, or the old errors will flow straight back in.
Common feed mistakes that cost sales
Marketing copy in the title. "Best seller!" and "Free shipping" belong nowhere near a product title. They breach Google's title rules, which ban promotional text and shipping details, and they push the useful words out of view.
One row for every variant. If six sizes share one row, the agent cannot confirm the size the buyer needs is in stock. Give each size a row and each one can be matched.
Copying the manufacturer description. Every stockist of that product now has the identical text. Add what only you know, such as fit notes and the questions customers ask most.
Set and forget. Prices change, stock runs out, products get discontinued. A feed that was perfect at launch is wrong within months unless someone owns it.
Fixing symptoms in the feed tool. Rules and overrides feel fast. They also hide the real errors in your store data, and they break the moment someone edits the product.
How to measure whether feed work is paying off
Nobody notices feed improvements unless you track them.
- Error and warning counts in your merchant account, week by week. They should trend to zero.
- Impressions and clicks on shopping listings for the products you fixed, compared with the period before.
- Visits from AI assistants in your analytics, where they can be identified, and sales from those visits.
- Your own agent test. Repeat the same shopping prompts each month and note whether your products appear.
The advantage is boring, which is why it lasts
My bet is that most of your competitors will keep pouring effort into creative and leave their feed as it was set up years ago.
That is your opening. Feed work is dull and invisible to anyone looking at your ads. It is also the thing the machines are reading.
It does not wear out the way creative does, either. A well structured title written today keeps qualifying that product for every request it matches until the product itself changes.
If you would rather hand the job to people who do this every week, feed structure sits inside our digital marketing work, right next to the search and AI visibility it feeds.
Export your feed today and open your top sellers. Read each title as if you were an agent holding a precise request, and ask whether you would pick it.
Frequently asked questions
What is a product feed and why does it matter for AI shopping?
A product feed is a structured file listing every product and variant with fields like title, price, availability, brand and size. AI shopping agents read these fields to decide which products match a buyer's request. If the information is missing or wrong, the agent cannot confirm your product fits and is likely to recommend a competitor whose data is complete.
How should I write product titles for Google Shopping and AI agents?
Write titles as precise matches. Start with the brand and product type, then the attribute that sets it apart, then details like size and colour. "Explorer Men's Waterproof Leather Hiking Boot, Wide Fit, Brown" can match far more requests than "The Explorer." Keep promotional phrases and brand voice out of the title and on your product page.
How often should I update my product feed?
Sync stock and price as often as your platform allows, ideally automatically every day or more. Then have one person review feed health weekly and clear the errors Merchant Center lists. Do a deeper review of titles and attributes whenever you add products, change prices or run a sale, because feeds decay as the store changes.
Do I need a developer to fix my product feed?
Usually not. Most fixes happen in your store platform, whether that is Shopify or WooCommerce: rewriting titles, filling attributes and splitting variants. Shipping and returns are set in your merchant account. Start with your top selling products and the errors your merchant account already lists. A developer helps with custom data or complex catalogues, but most small stores can make real progress in an afternoon.