Google Shopping feed optimisation used to be an ads chore. Somebody set the feed up, it passed diagnostics, and nobody opened it again. That same file now feeds three separate places customers find you and Google has since added six new attributes to Merchant Center. Most of their Help pages carry the same note: primarily intended for conversational experiences such as AI Mode. ChatGPT’s shopping answers cite Merchant Center feed URLs as well, so the extra information and context you’re now able to give will increase the chances of being recommended.
Content Overview
Six new fields, and how they can help you get recommended
The six are question_and_answer, document_link, related_product, item_group_title, variant_option and popularity_rank. All were launched at Google Marketing Live 2026, all are optional, and all are available in every country.
Open the Help page for question and answer, document link, related product or popularity rank, and each one opens with the same line. The attribute is primarily intended for use in conversational experiences such as AI Mode in Google Search. Shopping ads don’t get a mention on any of them, and neither do free listings.
You don’t need to know the possible use cases to show that a product is available (primarily what Shopping feeds do) but if your users are having a conversation about their preferences and use cases having the additional context confirmed and citable by AI tools is a huge advantage.
Feed data has always been read by more than the ads system. Free listings, the Shopping tab and Google’s product graph all pull from it.
Which puts your feed in an awkward spot on the org chart. Your ads agency probably owns it and a developer built the export years ago. The person responsible for how you show up in AI search has probably never logged into Merchant Center. We’ve written about how to adapt your Google Ads for AI Mode.
What each of the six fields does
question_and_answer
Both fields are capped at 1,000 characters, you can submit up to 30 pairs per product, and everything together caps at 10,000 characters. You’re now allowed to add 10,000 characters of confirmed use cases or context which is more text than most of your product pages.
Google’s framing is user, merchant and manufacturer authored questions, for people researching a product in detail. Its best practices ask you to vary them across specifications, ingredients, package contents, and what occasion or activity the product suits.
Say you sell 12 volt camping fridges. The questions that matter are practical:
- Does it run off a standard cigarette lighter socket?
- How many amps overnight?
- Does the lid open with an awning bracket over it?
Go and ask your sales team or consult your IG direct messages for every question (silly or logical) that customers have had doubts about, these are the thoughts that arrive at 9pm on a casual scroll before someone buys and having a follow up confirmed instead of guessed reduces potential customer friction. They’re not in your product description as they’re niche and shouldn’t take up space, but they’re close to word for word what a shopper types into AI Mode so if they do come up you want an answer ready to go.
Pro tip: Use TSV rather than CSV for the upload. Answers are full sentences and full sentences contain commas.
document_link
Up to five PDFs per product, 50 MB maximum each, with URLs up to 2,000 characters starting with http:// or https://. Google’s document link Help page lists training manuals, user guides, assembly instructions and package inserts. Packaging designs count too, including ones carrying QR or barcodes.
The PDF has to be publicly accessible and crawlable by Googlebot, so no logins, nothing blocked in robots.txt, and stable URLs that won’t expire. And Google’s best practice is that the document is about the product, not about your company. So this is more likely the technical questions rather than the question_and_answer is more situational and use case.
popularity_rank
A number between 0.0 and 100.0 that you supply. It expresses a product’s popularity as a percentage of your total inventory, so it’s scored against your own catalogue not the market. Google asks you to keep it accurate and update it when popularity shifts substantially.
This is the Google data version of your best sellers list, but you could also use it to nudge people towards certain things. Ask a model what a shop’s most popular option is and it has nothing to go on unless the shop has said.
The other three are connective tissue: which products go together, and how variants sit inside a group. related_product records relationships to other items in your catalogue. item_group_title gives a variant group one shared title of up to 150 characters, while each variant keeps its own. variant_option names the properties that separate one variant from another, as name and value pairs, up to thirty of them.
The same data is how ChatGPT finds you
Ethercycle, Kurt Elster’s US agency, published a study in July 2026 called State of Ecommerce 2026. One of their side findings was that ChatGPT’s shopping answers cite Google Merchant Center feed URLs. They could see it in the tracking, where the srsltid parameter Google appends to feed links turns up stacked on ChatGPT’s own tag.
Inside that sample, ChatGPT sessions arriving on feed-tagged links converted at 6.2%, against 2.9% for the rest of its ChatGPT traffic. Their entire AI channel was still about a tenth of a percent of revenue, so it’s a strong result inside a small slice.
If a model reaches into Merchant Center to answer a shopping question, the feed earns its keep whether or not you run Shopping ads. It’s a structured, machine-readable description of your catalogue, sitting in a database that AI shopping answers already read. We covered the wider version of this in Google’s AI is shopping for your customers.
Shopify already put you halfway there
Most of the platform work is already done.
On the Shopify builds we work on, product structured data ships with the current themes. The schema on your product pages is handled, unless somebody has taken a hammer to the theme code. Category metafields have been in the admin for a while too, and the merchants we onboard are usually defaulted into them. A good chunk of the taxonomy Google wants is already sitting in their product records.
WooCommerce takes more assembly, though one decent plugin closes most of the gap. If you’re still choosing, we’ve compared Shopify and WooCommerce in detail.
In the ecommerce audits we run, the platform is almost never the problem, the fields are just empty. Nobody wrote the answers, nobody made the PDFs crawlable, and nobody has ever ranked the catalogue against itself.
That’s a content job in a technical costume, which is why it falls between teams. It’s the same pattern behind most of the reasons an ecommerce website isn’t working: not a broken build, just work nobody was assigned. If you’re not sure what’s in your feed right now, it takes about twenty minutes to find out.
Where to start if you’ve got 2,000 products
Nobody is writing 30 questions for 2,000 SKUs.
Sort the catalogue by revenue over the last twelve months and take the top 20 products. The questions are already written, they’re in your support inbox, your live chat transcripts and the FAQ block at the bottom of your product pages. Those are the questions real customers asked, in the words they used. That’s much closer to how someone phrases a question to AI Mode than anything a copywriter writes from scratch.
Two of Google’s rules are worth reiterating before you go nuts adding extra content. Don’t duplicate what’s already in your title, description, product_detail or product_highlight. Repeating your own spec sheet in a Q&A field adds nothing to the answer. And if you’re submitting document_link PDFs that already contain the FAQ information, don’t also submit question_and_answer. Google says it will extract the FAQ information from those documents.
Then set popularity_rank once and diarise a refresh quarterly or yearly. If you’re an Australian ecommerce business already running Google Shopping, do this on your best sellers now. You’re already paying for the impressions those products get, and this is the cheapest improvement available to them. If you’re not running Shopping at all, keep the feed current anyway as it’s the only place your catalogue exists in a form both Google and ChatGPT can read without crawling your site.
It’s one job, make sure one person owns it
Three teams look at this data and none of them think it’s theirs. The ads agency treats the feed as a Shopping input and tunes titles for click-through. The web team treats product data as a CMS field. The SEO person works on pages and has no reason to open Merchant Center, because for a decade there wasn’t one.
Google has now stamped its AI Mode label on the fields sitting in the middle of all three. Fill them on your top sellers and you’ve improved Shopping ads, AI Mode and your odds of turning up in a ChatGPT shopping answer. That’s one afternoon in a spreadsheet, leave them empty and you’ve opted out of all three at once.
We run Google campaigns and build ecommerce websites, so we spend a lot of time inside feeds nobody has loved for years. If you want to know whether yours is one of them, come and have a chat with us. We’ll tell you which fields are empty and which products are worth filling first.
Rob Sharif
Technical SEO Lead
