Google Shopping feeds: the optimization layer most retailers ignore.

In a Search campaign you choose keywords, so it is obvious that targeting is something you control. In Shopping and Performance Max there are no keywords, and most retailers conclude there is nothing to control. The opposite is true. The product feed is the targeting. Google’s own guidance says the title is used to match your product to a customer’s search, and the rest of the attributes do similar work. The feed is not a data export. It is one of the most powerful optimization surfaces in the account.

Yet a common setup is a raw platform export, generated once when the store connected and untouched since. Bids get adjusted weekly while the data deciding which searches those bids apply to has never been reviewed. That inversion of attention is the opportunity.

Titles are your keywords now

The product title is the closest thing Shopping has to keyword targeting. Merchant Center allows up to 150 characters and recommends using them, putting the most important details first, because people usually notice only the first 70 or fewer characters depending on screen size. A title that leads with your internal product name spends that space on nothing. A title built the way people search (brand, product type, then the attributes that define the purchase) enters more of the right auctions.

• →Lead with what people search for, not your internal naming conventions
• →Front-load the most important terms, since the end of a long title is the part fewest people see
• →Include the attributes that disambiguate the purchase: size, color, material, compatibility, quantity
• →Write for the query, not the brochure: "womens waterproof hiking boots" beats "The Alpine Collection"
• →Skip promotional text, all caps, and symbols, which Google’s editorial rules ask you to avoid

What changed in 2026: Google may rewrite the title

Two changes make the feed matter more, not less. First, Merchant Center now has a structured title attribute, and titles created with generative AI must be submitted through it and marked as AI-generated. If you use AI to write titles at scale, that is a data field to set, not an optional disclosure.

Second, Google’s AI Max for Shopping, currently a beta, includes text customization that rewrites product titles to match individual searches using content from your feed, plus final URL expansion that can send shoppers to dynamic landing pages. Google notes text customization is limited to English feeds for now.

The operator reading: when Google writes the title shown for a query, it writes it from what your feed says. A feed missing material, fit, or compatibility gives the rewrite nothing to work with. Rich, accurate attributes become the raw material for every title Google generates, so completeness is now also creative input. Keep your own titles strong regardless, because not every campaign, market, or language will use the rewrite.

Completeness is not compliance, it is reach

GTINs, brand, correctly mapped product categories, color, size, gender, material: every one of these fields helps Google understand what you sell and which searches it belongs in. Google states that products submitted without unique product identifiers are hard to classify and may not be eligible for every Shopping program or feature, and that including a GTIN where one exists can improve performance. The shift is to stop treating attributes as fields you fill to avoid disapproval and start treating them as reach you are either claiming or leaving unclaimed.

Where a product genuinely has no GTIN, such as custom or handmade goods, Google does not require one. Submit brand and manufacturer part number where they apply instead. The failure is leaving the field empty for products that do have one, usually because the store never captured it.

Custom labels: the steering system

Custom labels do nothing for matching. They exist so you can structure campaigns around what matters to your business. Merchant Center gives you five, custom label 0 through 4, each with up to 1,000 unique values across the account. Label products by margin tier and you can bid high-margin products differently from clearance stock. Label by price band, seasonality, stock depth, or lifecycle stage (new launch, core line, end of life). Every label becomes a lever, because campaign structure can follow it.

This is where feed work connects to profit. An account that cannot separate high-margin from low-margin products in its structure is forced to bid to a blended average, and a blended average is wrong for almost every product in it.

At a single 3.0x target, the high-margin products are held back well above what they could afford, and the low-margin products lose $40 for every $100 spent while still reporting a respectable ROAS. The breakeven ROAS calculator gives you the floor for each tier; the fuller method for bidding to margin rather than revenue is in profit on ad spend.

To go further, add the cost_of_goods_sold attribute to your products. Google uses it with conversions with cart data to report gross profit in Google Ads, which turns profit by product from a spreadsheet exercise into a report you can read weekly. The companion piece on profit by product shows what that report tends to reveal.

Rules and supplemental data: change without touching the store

The practical objection to feed work is that product data lives in the store and changing it is an engineering ticket. Merchant Center removes that excuse. Supplemental data sources let you add or override attributes, such as titles and labels, without touching the source platform, and attribute rules let you set, extract, or take the latest value of an attribute, with conditions. A spreadsheet can drive a title test, and a rule can assign margin labels from a product type or price. There is no dependency standing between you and the work.

One caution. Rules are easy to add and easy to forget. Keep a short written list of every active rule and what it changes, because a rule written for last season’s promotion will keep rewriting titles long after the promotion ends.

Where to start if the feed has never been touched

Order the work by how much revenue it can move, not by how easy it is. Our usual sequence for a feed nobody has reviewed:

1. Clear disapprovals on products that sell, because a disapproved product enters no auctions at all
2. Add missing GTINs for products that have them, starting with your best sellers
3. Rewrite titles for the products behind most of your revenue before touching the long tail
4. Add a margin label and split campaigns only where the breakeven figures are far apart
5. Then move to descriptions, secondary attributes, and seasonal labels

The weekly discipline

• →Review disapprovals and warnings in Merchant Center every week, not once a quarter
• →Check which products get impressions and which are invisible, then fix the invisible ones, usually a title or attribute problem
• →Maintain the custom label taxonomy as margins, stock, and seasons change
• →Audit active attribute rules and supplemental sources for anything out of date
• →Run one deliberate feed test at a time, with a control group, like any other experiment

None of this is glamorous, which is exactly why it is still an edge. Many competitors are adjusting bids on top of a feed nobody has read. Do the feed work and you compete in better auctions before the bidding even starts. For how feed quality fits alongside offer, creative, and measurement as a growth lever, see how to scale an ecommerce business, and for the campaign side, the Google Shopping operator’s guide.