Google Shopping Feed Optimisation in Three Passes

Optimise a Google Shopping feed in three passes. First, get every product approved, with price, availability and shipping matching your site. Second, tell Google exactly what each product is: GTIN, brand, category. Third, improve what matches searches and earns the click: title, main image, description. Within each pass, fix the field that is empty on the most products, starting with products that get ad spend. Error-hunting is a small part of the work: in 109 of 142 stores in our data, under 1% of the catalogue carries any flag.

What does feed optimisation change?

Your product feed is the file of products, prices, images and attributes that your store sends to Merchant Center, Google’s service for product data. Google builds Shopping ads and free listings from it. According to Google’s Product data specification, Google uses this data to match your products to the right queries. The same page warns that incorrect or missing information can lead to disapprovals, limited eligibility or products shown incorrectly.

So the feed decides two things: whether a product may show at all, and which searches Google thinks it fits. You set bids and budgets in Google Ads. A product appears in ads only when both sides work: Merchant Center has approved it, and a campaign includes it. How Google Shopping ads work follows one product from the feed to a Shopping result.

Performance Max starts from the same data. Google’s guide for retail campaigns, Performance Max for online sales with a product feed (retail), states that you must link your Google Ads account to a Merchant Center account. So feed fixes apply to Standard Shopping and Performance Max alike. Our Performance Max guide for online stores covers the campaign side.

In what order should you fix the feed?

Fix the feed in the order a product meets Google: Google lets it show, then understands it, then a shopper chooses it. A later pass pays off only for products that clear the earlier ones.

The last column shows how the GetProfit portal calculates its feed score: nine checks with fixed weights that add up to 100. It is our rule, not a Google formula, and it lets you compare fixes in points. A separate article, Google Shopping feed attributes, covers which attributes Google requires, which are optional and which change results.

Pass 1: make every product eligible to show

Google’s specification marks a short list of attributes as required: id, title, description, link, image_link, availability and price. brand is required for new products other than movies, books and music. “Required” has a precise meaning there: without the attribute, the product can’t serve in ads and free listings. Google also requires shipping costs for Shopping ads and free listings in 21 countries, including Czechia, Poland, Germany and the UK.

Next, fix mismatches between the feed and your site. The price and availability must match your landing page, checkout and structured data (the product markup in your page code). Outside the US and Canada, the price must include VAT. If prices and stock change often, two things help:

• Automatic updates. Merchant Center can update price, sale price, availability and condition from the structured data on your product pages (Allow Merchant Center to update product information automatically). Google describes this as a fix for temporary problems on a small share of products; you still need regular feed updates.
• Structured data and a feed together. Google Search Central (Introduction to Product structured data) says that providing both maximises your eligibility and helps Google verify your data.

Measure this pass by approval: the share of products Merchant Center has approved. The portal treats 95% or more as healthy, 85–95% as a warning and below 85% as an error. It also raises a warning when 3% or more of products are disapproved and marks 10% or more as critical. Our feed audit in nine checks shows how to work through the issue list in Merchant Center, most damaging issues first.

Pass 2: tell Google exactly what each product is

GTIN. A GTIN is the manufacturer’s barcode number. Google’s help page GTIN [gtin] strongly recommends it for every product that has one and warns that products with missing or incorrect GTINs may have limited visibility. In its words, “Accurate matching to products can’t be assured when GTIN is missing.” Submit only real GTINs: according to the specification, a wrong GTIN gets the product disapproved.

Store-brand products and products only you sell usually have no GTIN. For them, Google asks for brand and mpn (the manufacturer part number). When a product has neither a GTIN nor an MPN with a brand, set identifier_exists to no. Our article on GTIN in Google Shopping shows what the “missing identifiers” warning costs in practice.

Brand. Fill the brand field with the name shoppers recognise. According to the specification, use your own store’s name only if you make the product or sell it under your own label. If a product truly has no brand, leave the field empty rather than writing “N/A” or “Generic”.

Category. Google automatically assigns every product a product category from its own category list. Its help page Google product category [google_product_category] says that good titles and descriptions, plus accurate price, brand and GTIN, help it get the category right. Set google_product_category yourself only when you need to override Google. Google accepts an override in these cases:

• a wrong category adds extra required fields, as apparel does;
• you build your campaigns on Google’s categories;
• the product is alcohol.

To decide whether an override is worth the work, see Google product category.

Separately, product_type holds your own category tree, such as “Home > Women > Dresses > Maxi Dresses”. Google’s page Product type [product_type] says it helps Google understand what you sell and lets you organise Shopping campaigns by it. Only the first value counts for bidding and reporting. If you organise campaigns by it, Google recommends that you submit one value with all its levels.

Pass 3: improve what matches searches and earns the click

Title. Of the two text fields, the title and description, the title does more of the work. Google’s page Title [title] and structured title [structured_title] calls it one of the most prominent parts of an ad and says Google uses it to match your product to a search. You have 150 characters. Put the most important details first: people usually see only the first 70 characters or fewer.

Add the brand where it sets the product apart, and leave out prices, shipping and promotional text. Google may also show its own version of a title, built from your landing page and other signals to fit a search. Google Shopping title optimisation covers what to put first in your category.

Description. You have up to 5,000 characters. Write about the product only, and make it match the landing page. The specification rules out links to your store, sales information and promotional text. Google sets no minimum length; the portal counts a description under 50 characters as short, and that threshold is ours.

AI-written text. If generative AI wrote a title or description, the specification asks you to submit it in structured_title or structured_description, marked as AI-generated.

Images. The main image, image_link, must show the product accurately. The specification rules out promotional text, watermarks, borders, placeholders, thumbnails and upscaled images. You can add up to 10 additional images to show the product in use.

Google has announced a minimum size of 500 × 500 pixels for all product images and will enforce it from 31 January 2027. To choose which image should lead, see Google Shopping images.

Optional attributes. The specification lists product_highlight (2 to 100 short points) and product_detail as optional. It describes optional attributes as ones you can submit to help boost a product’s performance. Fill them in once the three passes are done.

How much is each fix worth?

The portal’s feed score puts nine checks on one scale. For each check, it multiplies the share of products that pass by the check’s weight. It adds up the results and divides them by the total weight of the checks that had data. A check with no data drops out of the score.

The 0–100 score then gets a grade: 85 and above is an A, from 70 a B, from 50 a C, from 30 a D, and below 30 an F. The title and description check averages two shares: products with a title, and products with a description of 50 characters or more.

That makes the cost of each gap easy to read: points lost = weight × share of products that fail the check.

Example store, not client data.

A tableware shop has 3,000 products. Of them, 180 are not approved, 1,800 have no GTIN, 30 have no image, 1,500 have a description under 50 characters and 450 have no brand. All other fields are filled on every product.

The score is 100 − 15.76 = 84.24: a B, just short of an A. Three things follow from the table.

1. GTIN is the biggest gap, and you can close only part of it. Say 1,200 of the 1,800 products without a GTIN are the shop’s own brand: they have no GTIN to add. Adding GTINs to the other 600 earns 16 × 600 ÷ 3,000 = 3.2 points.
2. Descriptions are the cheapest way to an A. Lengthening the 1,500 short descriptions recovers 3.5 points and lifts the score to 87.74.
3. Approval still comes first. The 180 unapproved products cost only 1.32 points, yet they get no impressions at all until you fix them. Points measure gaps in the data; the order of passes decides what you do first.

In most stores, Merchant Center flags almost nothing

We looked at what Merchant Center flags in our stores’ catalogues. These are observations, not an experiment.

Flags sit in a few large catalogues. Merchant Center puts at least one flag on 3,211,061 of the 5,744,507 products in our stores’ catalogues, or 55.9%. Yet the median store has 0.0% of its catalogue flagged, and 109 of 142 stores have less than 1% flagged (GetProfit data, catalogues of 142 stores with 100+ products, fact bank compiled in August 2026).

Data errors are rare among the flags. In a sample of 400,000 flagged products (not a random one, so the large catalogues dominate it), 263,483 carried “not eligible in any campaign” and 209,328 “out of stock”. Third came a missing shipping setting for the country, on 38,032 products.

Classic data errors, such as a missing required attribute or an image that is too small, numbered in the hundreds. The two leading flags reflect decisions about campaigns and stock rather than broken feed data. The full breakdown is in what Merchant Center flags in 142 stores.

The link with results is weak. Stores with a larger flagged share had slightly lower ROAS, but the correlation is weak (Spearman’s rho −0.161, 116 stores) and shows no cause.

Budget reaches few products. In our study of 1.4 million products across 130+ stores over 13 months, 55.0% of the 1,404,808 products got no ad budget at all. In the median store, the top 10% of products took 68.1% of spend.

So for most stores, feed optimisation means filling gaps more than hunting errors. Several gaps that matter sit outside Merchant Center’s errors: a GTIN is strongly recommended rather than required, and Google sets no minimum length for a description. The work pays off first on the minority of products that your budget reaches.

Which products should you fix first?

Fix by hand where the money goes, and fix by rule everywhere else.

• By hand: products that get spend. Improve the titles, main images and descriptions of your top sellers and of the products with the most spend. A few products take most of the budget, so an improvement there reaches the most searches.
• By rule: patterns across the catalogue. Typical cases are a brand missing from every title, a wrong condition, or a category tree in the wrong field. Merchant Center’s attribute rules transform your data to fit the specification (Set up your attribute rules). One of Google’s examples builds a title from the brand plus the existing title. Rules need the Advanced data source management add-on, and they can take values from a supplemental data source as well as from the main one.

To split a catalogue of thousands of products between the two methods, see which products to fix first. Our playbook for quick wins with product data shows both methods for titles step by step.

How do you know a feed change worked?

Record the date and the exact change. Then compare the changed products with similar products you left alone, over periods of equal length before and after the change. A season, a sale or a bid change can move results by itself, and the untouched group shows how much.

Look at impressions and clicks as well as orders: a feed change affects which searches a product shows for and whether people click. Google may show its own version of a title, so the ad can carry Google’s text instead of yours. The method and its pitfalls are in testing feed changes.

On the data side, the portal’s Feed and Merchant Center section re-runs its nine checks with every scan. For each check, you see how many products are affected and what the previous scan showed. You can download the list of affected products as a table.

A feed optimisation checklist

1. Count each check across the whole catalogue. For every field, note how many products fail it. A gap on 1,800 products matters more than one on 30.
2. Clear disapprovals and mismatches. Fix price, availability and shipping until approval reaches the portal’s healthy level of 95%. An unapproved product gets no impressions, whatever else you fix.
3. Add real GTINs and brands. Use the manufacturer’s GTIN where one exists. Set identifier_exists to no only where the product has neither a GTIN nor an MPN with a brand. Missing GTINs can limit visibility, and wrong ones get products disapproved.
4. Fix wrong categories and build product_type. Override Google’s category only in the three cases Google accepts, and give every product one full product_type path. That keeps category-specific requirements right and lets you split campaigns by your own tree.
5. Rewrite the titles of your top-spend products. Lead with what the shopper searches for, and keep the key details within the first 70 characters. Google matches the title to searches, and people usually see 70 characters or fewer.
6. Replace weak main images. Use one clean image of the product, with no text or watermark, at least 500 × 500 pixels. Google will enforce that size from 31 January 2027.
7. Extend short descriptions. Describe the product the way its landing page does. Google sets no minimum, but the portal counts anything under 50 characters as short.
8. Measure again and compare. Check the new score, then the impressions and clicks of the changed products against an untouched group. The score shows that the data improved; the comparison shows whether it paid off.

The product is in the catalogue, but Google doesn’t have it. The portal goes through your feed and shows which fields are empty, on how many products, and what that does to impressions. The portal changes nothing without your consent.

Check your feed →

Sources

• Product data specification — Merchant Center Help: how Google uses product data; what missing or conflicting data causes; required attributes and what “Required” and “Optional” mean; rules for price, availability, brand, GTIN, MPN, identifier_exists, description, images and AI-generated text; countries where shipping costs are required; 500 × 500 pixel images from 31 January 2027. Checked 2 October 2026.
• Title [title] and structured title [structured_title] — Merchant Center Help: the title matches products to searches; 150-character limit; the first 70 characters or fewer are visible; dynamic titles. Checked 2 October 2026.
• GTIN [gtin] — Merchant Center Help: GTIN strongly recommended; missing or incorrect GTINs may limit visibility; store brands without a GTIN. Checked 2 October 2026.
• Google product category [google_product_category] — Merchant Center Help: automatic categorisation and the three cases for an override. Checked 2 October 2026.
• Product type [product_type] — Merchant Center Help: your own category tree; only the first value is used for bidding and reporting. Checked 2 October 2026.
• Allow Merchant Center to update product information automatically — Merchant Center Help: automatic updates of price, sale price, availability and condition; not a replacement for regular feed updates. Checked 2 October 2026.
• Set up your attribute rules — Merchant Center Help: attribute rules, the add-on they need, primary and supplemental data sources. Checked 2 October 2026.
• Introduction to Product structured data — Google Search Central: structured data and a Merchant Center feed together maximise eligibility. Checked 2 October 2026.
• Performance Max for online sales with a product feed (retail) — Google Ads API: Performance Max for retail needs a linked Merchant Center account. Checked 2 October 2026.
• GetProfit data: catalogues of 142 stores with 100+ products, fact bank compiled in August 2026 — share of products and of each store’s catalogue flagged by Merchant Center; the most common flags in a sample of 400,000 flagged products; correlation with ROAS in 116 stores.
• GetProfit study of 1,404,808 products, 130+ stores, 13 months — products with no ad budget; share of spend on the top 10% of products.
• GetProfit portal methodology — feed score: nine checks and their weights, grades A to F, approval and disapproval thresholds, the 50-character description threshold.