On this page
If your company sells products online, someone has probably asked you this year to rewrite your Google feed for Google AI Mode, Google's chat answer page: longer titles, every attribute (a product fact such as size, colour or material) filled, Google's new conversational fields on top. The feed is the file of products and prices a store sends to Merchant Center, Google's tool where a store lists what it sells.
On a catalog of a few thousand products that is a long stretch of a merchandiser's time, and until recently nobody could tell you whether any of it moved anything.
| The catalog | By hand |
|---|---|
| Products in the catalog | About 4,000 |
| Products needing facts filled each month | About 400 |
| Missing facts per product | About four, at about three minutes each |
| Hours a month | About 80, across two merchandisers |
| Cost a year in pay and benefits | $28,474 |
Data entry keyers, the closest occupation for copying values from spec sheets, earned a mean $43,310 in May 2025, the Bureau of Labor Statistics reports; with benefits at 30% of private employer costs, that is about $61,900 a person. Merchandisers may earn more; the calculator takes your own figure.
Product Attribute Filling ROI Calculator
What Filling Missing Product Facts From Spec Sheets Costs You, And What An Agent Saves
Saved Per Year
$17,979
765 hours handed back, after $4,800 a year to run it
- Hours Handed Back
- 765 hours a year
- Agent Running Cost (Estimate)
- $400 a month
- Payback Per $10,000 Of Build
- 6.7 months
On these numbers, the agent saves $17,979 a year after its own running cost, and every $10,000 of the build is earned back in 6.7 months.
On those numbers the filling costs $28,474 a year, and $10,495 with AI agents taking about eight in ten of the values, our assumption ($5,695 of staff time left and $4,800 a year to run them): $17,979 and 765 hours handed back, and every $10,000 of the build earned back in 6.7 months.
The first catch-up run, over the whole catalog, is larger than the monthly one.
Merchant Center now has a tab for that question: its AI performance report counts how often your products appear in AI Mode and in AI Overviews, the box above the results. A UK store has no report yet.
Where Your Feed Shows Up in the Answer
Google AI Mode's shopping answers stack a recommendation on top of a product card, the tile with picture, price and store, and the feed is what fills the card. Each card is a single offer from a single store, and Google picks which one the shopper sees. ChatGPT builds its shopping answers another way, from copies of your product records it reads elsewhere.
If you make the products, the card is often already yours, and the job is keeping your own listing complete and correct.
If you sell other companies' products, the brand's own store is often the offer you compete with for the card, and your feed is the one place where you decide what your offer says.
What the Feed Work Costs Your Merchandisers
Google's price-mismatch guidance says Google's automated page reader checks feed prices against product pages. Mismatches can get products disapproved and repeated mismatches can lead to account suspension. Disapproved products, the ones Google refuses to list until the data is fixed, leave its unpaid listings until then, so correct price, availability, product identification codes, images and titles before adding new fields.
Start with attributes, not titles: they are facts you already hold, and a title rewrite needs a person's judgment on every line.
Two sources show whether the work changed anything: Google's report counts impressions, each one a showing of your product to a shopper, measured against competitors Google picks, and the daily answers show which store and price the shopper saw.
Where the Missing Attributes Come From Now
The agents we build for your catalog read only your own records: the category report your catalog owner exports, your product records or spec sheets, and your current Merchant Center data. They run once for a category, then monthly as the report refreshes.
They copy a value only where a record states it, and write a supplemental file, an extra file of facts added without changing the products already listed: each product, attribute, value and source line, plus the gaps.
| The agents may | The agents never |
|---|---|
| Read your report export, product records, spec sheets and current Merchant Center data | Infer a value from a photo or a product name |
| Copy a value a record states, with the line it came from | Upload anything to Merchant Center |
| List every gap for a person | Run searches on Google |
The agents never infer values from photos or product names. Missing facts go to a person, not into the feed, and every value carries the record line it came from.
The catalog or ecommerce manager reads the file, rejects what looks wrong and uploads it. The agents never do.
The agents flag any listing whose feed price disagrees with your product records; checking every night that the feed's price matches what your product pages send stays a separate job with a separate owner.
Bring Us Your Missing-Attribute Export
Bring the missing-attribute export and your spec sheets. Your catalog owner has the first file two weeks after access, and uploads it.
What the agents may read and do is set in a written agreement, your data is processed where that agreement says, and every action the agents take is logged.
Frequently asked questions
Can an AI agent change our feed on its own?
Not the agents we build. They write a supplemental data file from your own records, with the source line behind every value, and list what they cannot fill. The catalog or ecommerce manager approves and uploads it.
What should we bring for the attribute agents?
The report's attribute export, access to your product records or spec sheets, and your current Merchant Center data. Name the catalog owner who will check and upload the supplemental file.
Does a Google Shopping feed affect Google AI Mode answers?
Yes, through the product card. Google AI Mode stacks a recommendation on top of a product card, the tile with picture, price and store, and the feed is what fills the card. Each card is one offer from one store, and Google picks which.
Which store does Google AI Mode show on its product cards?
In our research, 48.09% of product cards named the brand's own store, 27.29% another store, usually a specialist, and 24.62% a large retailer or marketplace.
Should a retailer rewrite product titles for AI Mode?
Start with attributes, not titles. Attributes are facts you already hold, like size, colour or material, while a title rewrite needs a person's judgment on every line.
Can you measure how often products appear in AI Mode?
Merchant Center's AI performance report counts how often your products appear in AI Mode and in AI Overviews, measured against competitors Google picks. A UK store has no report yet.
What happens if feed prices don't match product pages?
Google's automated page reader checks feed prices against product pages. Mismatches can get products disapproved, and repeated mismatches can lead to account suspension; disapproved products leave the unpaid listings until the data is fixed.
What does filling missing product attributes cost?
In this post's example, about 400 products a month need about four facts each at about three minutes a fact: about 80 hours a month across two merchandisers, $28,474 a year.
How much can AI agents save on product attributes?
With agents filling about eight in ten of the values, the example falls to $10,495 a year, $4,800 of running cost included: $17,979 and 765 hours back. The first catch-up run over the whole catalog is larger.
Do catalog agents guess product attributes?
No. They never infer a value from a photo or a product name. They copy a value only where a record states it, and missing facts go to a person, not into the feed.
What rules govern access for product-data agents?
A written agreement sets what the agents may read and do, and where your data is processed. Every action the agents take is logged. Their access covers only your own records.
How long does the first product-attribute file take?
The first file reaches your catalog owner two weeks after access. The agents run once for a category, then monthly as the report refreshes. Your catalog owner uploads that first file.
Does ChatGPT use Google Shopping feeds?
ChatGPT builds shopping answers from copies of product records it reads elsewhere. Google Shopping feeds fill Google AI Mode's product cards. Each card shows one store's offer, chosen by Google.
When do product-attribute agents repay their build cost?
The example repays every $10,000 of build cost in 6.7 months. It assumes agents fill about eight in ten values, leaving $5,695 in staff time and $4,800 in annual running costs.
Does a retailer of other brands compete with the brand's own store?
Often. If you sell other companies' products, the brand's own store is often the offer you compete with for the card, and your feed is the one place where you decide what your offer says.
Is filling missing product facts a one-time job?
No. The agents run once for a category, then monthly as the report refreshes. In the post's example, about 400 products a month need facts filled. The first catch-up run is larger.
What pay should you assume for product data entry?
Data entry keyers, the closest occupation for copying values from spec sheets, earned a mean $43,310 in May 2025, the Bureau of Labor Statistics reports. With benefits at 30% of private employer costs, that is about $61,900 a person.
Which product data should be correct before anything else?
Correct price, availability, product identification codes, images and titles before adding new fields. Google refuses to list disapproved products until the data is fixed. Those products leave its unpaid listings until then.
Do catalog agents check feed prices?
They flag any listing whose feed price disagrees with your product records. Checking every night that the feed's price matches what your product pages send stays a separate job with a separate owner.
How do you know the feed work changed anything?
Two sources: Google's AI performance report counts impressions of your products against competitors Google picks, and the daily answers show which store and price the card showed for each shopper question.
