Nobody has ever been promoted for fixing a spreadsheet.
You spent this year on creative. Hooks, UGC, a founder video, three rounds of feedback on a thumbnail. Meanwhile the system deciding whether your product appears at all never watched a single second of it.
It read a file. A file most brands have never opened.
Here is the shift almost everyone has under-reacted to. Across Google Shopping, Performance Max, AI Max and now the AI assistants people are actually shopping in, the machine no longer takes instructions about who to show your product to. You cannot pick the audience. You cannot pick the keyword. What you can do is describe the product, and the machine decides everything downstream from that description.
Your feed stopped being housekeeping. It became the targeting layer.
A definition, because the term gets used loosely: a product feed is a structured file listing every product you sell and its attributes, including title, description, price, availability, images, identifiers like GTIN, and shipping data. It is the machine-readable version of your catalogue, and it is what advertising systems and AI shopping assistants read instead of your website.
Why Google Shopping feed optimisation became your targeting
Because Google removed the levers, one at a time, and replaced them with inference.
Because Performance Max is now the default path for most ecommerce accounts, and it builds its matching from your product data rather than your keyword list.
Because on 24 July 2026 Google launched AI Max for Shopping and Travel campaigns (Search Engine Journal, July 2026), extending AI-driven query matching into product ads. The model builds its understanding of what you sell from your titles, descriptions and landing pages.
Because in January 2026 Google launched the Universal Commerce Protocol, an open standard for agentic commerce, and has since expanded it across Search, Gemini, YouTube and Maps, with simplified Merchant Center onboarding added in March 2026 (Google Developers Blog; Google blog). UCP turns the Merchant Center feed into a checkout-ready data layer that AI surfaces can read and act on.
Because the AI shopping assistants building on top of that standard inherit its requirements. Google’s product data specification requires a GTIN wherever the manufacturer has assigned one, so a catalogue with missing identifiers is a catalogue that agentic surfaces built on that spec struggle to match against anything.
Every one of those systems reads the same thing. Not your brand story. Not your award-winning campaign. A row in a file.
And the file is usually terrible.
What a bad feed actually looks like
Open your Merchant Center and look at a product title. If it says something like “Coastal Tee – Blue – M”, your feed is a Shopify default that was never written for a machine, and it is costing you impressions you will never see reported anywhere.
Nobody types “Coastal Tee”. They type “men’s organic cotton t-shirt blue”. Your title contains almost none of that.
The systems reading your feed are pattern matchers. They match the words in your product data against the words in a query, or against the intent in an AI conversation. When your title is a product name your marketing team invented, you have handed the machine a word that exists nowhere in customer language.
Store Growers analysed 151 product titles across 120 advertisers and found the winning structures front-load brand and product type before variant detail, and that fewer than half of advertisers included a brand name in the title at all (Store Growers). That research is US-based, so treat the specific formulas as a starting template rather than gospel. Your own Search Terms report in Google Ads is the free local version, sitting in your account right now, telling you the exact words Australian customers used before they bought.
Most brands never copy those words back into the feed. The data goes in one direction and stops.
Vendors will quote you dramatic lift numbers for feed work. Treat those as directional, because nobody publishes the losers. What I can say from working on Australian ecommerce accounts is narrower and more useful: a week spent rewriting titles and filling missing attributes has repeatedly moved more volume than a month of bid adjustments, because it changes what you are eligible to appear for rather than what you pay when you do.
The 2026 specification changes nobody put in a calendar
Google updated the Merchant Center product data specification in April 2026 and set deadlines that mostly land after everyone’s Q4 planning was finished (Google Merchant Center Help).
- Minimum product image resolution rises to 500 x 500 pixels, with warnings from 14 April 2026 and enforcement from 31 January 2027
- A new optional video_link attribute accepts product video URLs, with serving plus policy and quality validation beginning 30 June 2026
- New product-level shipping attributes: handling cutoff time, minimum order value, and loyalty-related shipping labels
The image change is the one that bites. Google will auto-optimise some undersized images using near-duplicates or AI upscaling to prevent disapproval, which sounds generous until you realise the machine is now inventing your product photography. Anything it cannot rescue goes dark in January, and January is exactly when nobody is looking at Merchant Center.
The video attribute is the interesting one. A field that accepts product video, on a surface where AI assistants are increasingly summarising products for shoppers, is Google building a lane. Brands with product video already shot are one CSV column away from occupying it.
What you can do right now
1. Read your own search terms and put the words back. Export the last 90 days of search terms from Google Ads. Find the phrases that converted. Those exact phrases belong in your product titles, in the first 60 to 80 characters. This is a two-hour job with an immediate effect on what you are eligible to show for.
2. Rewrite titles to the pattern, not to the brand. Brand, product type, key attribute, variant. “Thrive Organic Cotton T-Shirt, Men’s, Blue, Medium” beats “Coastal Tee – Blue – M” every time, because the second one only makes sense to people who already know you. Feeds are not the place to be clever.
3. Fill in GTINs. Not because Google nags you, but because the identifier is how every system downstream, including AI shopping surfaces built on Google’s spec, knows your product is the same product it saw somewhere else. If you manufacture your own products and have no barcodes, get them. It is a small annual cost that determines whether you are matchable at all.
4. Audit for missing attributes, then fix the top 20 SKUs first. Colour, size, material, age group, gender, condition. Automated systems learn slower on sparse data and allocate budget away from products they cannot understand. You do not need a perfect catalogue. You need your revenue-driving SKUs complete.
5. Check your images against 500 x 500 today. Warnings are already live. Enforcement lands 31 January 2027, in the dead zone right after Christmas when nobody is checking dashboards. Do it now while you are thinking about it.
6. Treat product descriptions as machine input, not shelf copy. AI assistants summarising your product read the description. If yours is three lines of atmosphere about coastal mornings and no mention of fabric, fit, or care, the assistant has nothing to work with and will describe a competitor instead.
There is a strategic point underneath the admin. Every time a platform removes a targeting control, it moves the leverage upstream into the inputs. Meta did it with Advantage+ and Andromeda, where creative variety became the targeting. Google is doing it with AI Max and PMax, where the feed became the targeting. The same thing keeps happening: the machine takes the steering wheel and hands you responsibility for the map.
The brands that keep winning are not the ones with the best prompt or the cleverest bid strategy. They are the ones who noticed that the inputs got more important, not less, precisely as the controls disappeared.
Your competitor’s feed is probably as bad as yours. That is the opportunity. It expires the moment one of you opens the file.
If you want a straight assessment of what your feed is costing you before Q4, book a growth call and we’ll go through it with you.
FAQ and TL;DR Version:
What is a product feed? A product feed is a structured file listing every product you sell along with its attributes: title, description, price, availability, images, identifiers such as GTIN, and shipping data. Advertising platforms and AI shopping assistants read this file rather than browsing your website, which makes it the primary input for Google Shopping, Performance Max, AI Max and agentic shopping surfaces.
Why does feed quality matter more in 2026? Because automated campaign types removed manual targeting controls. Performance Max and AI Max build query matching from product data rather than keyword lists, and AI shopping agents query structured catalogue data directly. When you cannot choose the audience, the product description becomes the targeting instruction.
What changed in Google’s Merchant Center product data specification in 2026? Google’s April 2026 update raised minimum product image resolution to 500 x 500 pixels (warnings from 14 April 2026, enforcement from 31 January 2027), introduced an optional video_link attribute with serving beginning 30 June 2026, and added product-level shipping attributes including handling cutoff time, minimum order value and loyalty shipping labels.
How do I get my products into AI shopping assistants? Through structured product data. Google’s Universal Commerce Protocol, launched January 2026, uses the Merchant Center feed as a checkout-ready data layer across Search, Gemini, YouTube and Maps. Assistants reading that layer rely on complete attributes and valid product identifiers, so clean, complete, frequently refreshed feed data is the entry requirement.
