Google Shopping feed: the complete guide to structuring, auditing and optimizing it

What is a Google Shopping feed?

A Google Shopping feed is a structured list of your products that you send to Google Merchant Center. Each row is a product and each column an attribute: ID, title, description, price, availability, product page link, image, brand, GTIN, category. Google reads this data to decide which products it can show, for which searches, and in what form.

The feed powers Shopping ads, the product inventory in Performance Max, free listings in the Shopping tab and the way Google presents your products in its AI-generated answers. Shopping campaigns have no keywords: Google matches the shopper’s query against the information in your feed. A thin title or a missing attribute cuts the number of searches a product can appear for.

That puts the feed on the traffic manager’s desk as much as on the IT team’s. It shapes ad relevance, click-through rate and, further down the line, the cost of each sale. For how the platform itself works (accounts, diagnostics, programs), our complete guide to Google Merchant Center complements this one.

Formats and data sources

Accepted file formats

Google Merchant Center accepts several formats for file-based feeds:

• Delimited text (TSV or .txt): a table with a header row listing attribute names. Easy to produce from an ERP export or a spreadsheet.
• XML (RSS 2.0 or Atom 1.0): the most common format for exports generated by e-commerce platforms and feed tools, because it handles repeated attributes such as additional images well.
• Google Sheets: handy for a small catalog or a test, less suited to anything beyond a few hundred products.

Update frequency matters more than format. Google treats a product that has not been updated for 30 days as expired, and a price or stock level out of sync with your site leads to disapprovals.

Data sources in Merchant Center

File uploads can be manual, scheduled fetches from a URL or SFTP transfers. Platform integrations (Shopify, WooCommerce, PrestaShop and others) sync the catalog through an app. Google can also discover products on your website from the structured data on your pages, which helps you get started but gives you little control over what gets sent.

Content API and Merchant API

For large catalogs, or prices and stock that change several times a day, an API connection remains the most reliable option. Google announced that the Merchant API replaces the Content API for Shopping, with the Content API sunset on August 18, 2026. New integrations should be built on the Merchant API, and existing ones should already have migrated. Ask your developer or feed provider which API your connection relies on.

Supplemental feeds

A supplemental feed (or supplemental data source) adds or overrides attributes on products that already exist in your primary feed. It joins on the id attribute and cannot create products.

It is the most useful tool for optimizing without touching the source feed. You can push rewritten titles, custom_label values, a more detailed product_type or missing attributes, while your site export keeps handling price and stock. Feed Enrich, Dataïads’ product feed optimization tool, works this way: it publishes enriched data as a supplemental feed to Google Merchant Center, with no change to your site or your existing feed tool.

Required attributes and the ones that drive performance

Without the required attributes, Google disapproves the product. For performance, some optional attributes weigh more than required fields that are already filled by default.

Our article on how well-filled attributes boost Shopping campaigns goes through them one by one. Three groups deserve extra attention.

Identifiers: GTIN, MPN and brand

The GTIN (UPC in North America, EAN in Europe) lets Google identify the product without ambiguity and link it to the matching searches and listings. When several merchants sell the same product, a missing GTIN holds back serving. If the manufacturer never assigned a GTIN, provide brand and MPN. If the product has neither (handmade, custom-built), the identifier_exists attribute says so. Making up a GTIN or reusing another product’s leads to disapprovals.

Taxonomy: google_product_category and product_type

Google assigns a google_product_category automatically if you leave it empty, and it can get ambiguous products wrong. Set it yourself for sensitive categories such as apparel or restricted products.

You control product_type. A three- or four-level tree (“Home > Kitchen > Cookware > Frying pans”) gives Google context and lets you structure product groups in Google Ads.

Custom labels

The five custom_label attributes let you segment the catalog on your own criteria: margin, seasonality, stock level, best sellers, new arrivals, price band. Google does not use them for serving, but they decide how finely you can steer: without them, Performance Max treats a high-margin product and an end-of-line item the same way. Each label takes up to 100 characters, and Google caps each label at 1,000 unique values, so use categorical values rather than IDs.

How to optimize titles, with before and after examples

Titles pay back the most when you rework them. Google uses them to match products to searches, and shoppers read them first. Google often truncates the displayed title, so the first words have to carry the meaning.

A structure for each category

No single formula fits every catalog, but the logic holds: what shoppers type should appear early in the title. A few structures that work:

• Apparel: brand + product type + gender + key attribute (fabric, fit) + color + size
• Home: brand + product type + material + dimensions + color
• Auto parts: brand + part type + vehicle compatibility + part number

Put the brand first when people search for it, and further back when it is little known and the product type matters more to shoppers.

Before and after examples

The examples below are built to illustrate the method.

Each rewritten title names the product type, adds the attributes shoppers filter on (gender, size, volume, compatibility) and drops internal abbreviations. Avoid keyword-stuffed titles, all caps and promotional text (“Free shipping”, “20% off”), which Google rejects or ignores.

Images and visuals in the feed

In a Shopping carousel or a Meta Catalog Ad, shoppers see the image before they read the title or the price. When ten similar products sit side by side, the image wins or loses much of the click. It also decides eligibility: a rejected image means a rejected product.

Google currently sets minimums of 100 × 100 pixels (250 × 250 for apparel), raised to 500 × 500 pixels for all products from January 31, 2027, and recommends 1,500 × 1,500 pixels so shoppers can zoom. When you replace an image, change its URL (a timestamp in the file name is enough) so Google fetches the new version sooner.

What Google rejects

Promotional text (“50% off”, “Free delivery”), watermarks, retailer logos, borders, collages and generic placeholder images all lead to disapprovals. Promotions belong in sale_price, not in the picture. For a customizable product, the image has to show the exact variant being sold.

Packshot or lifestyle

The packshot remains the standard for the main image: it complies, reads well at thumbnail size and compares cleanly across merchants. A lifestyle shot shows scale, use and mood, which helps shoppers picture the product at home, on themselves or in a beauty routine. Keep it for lifestyle_image_link and additional_image_link, then compare on a test segment, since the winning format varies by category and placement. Our article on e-commerce visuals that convert covers these levers on the product page side.

The feed image should also match what shoppers find after the click: same color, same variant, same pack size. Google flags mismatches between the image and the product page, and a disappointed shopper leaves. Watch product families grouped under item_group_id, where a single photo sometimes stands in for every color.

Visual search and AI-generated visuals

With Google Lens, shoppers photograph a product to get Shopping results, and that match relies on feed data. Since April 2025, AI Mode also takes questions that combine an image and text, through Lens. Sharp photos from several angles give these matches more to work with.

An AI-generated visual belongs in the feed as long as it stays faithful to the product. A generated background is fine; a changed color, shape or accessory misleads the shopper. Google expects IPTC metadata on AI-generated images, and Article 50 of the EU AI Act requires machine-readable marking of generated content. Our Merchant Center guide to images, videos and AI compliance and our breakdown of the AI Act for e-commerce visuals cover the rules in detail.

Producing visuals at catalog scale

Feed Enrich handles visuals the way it handles the rest of the feed. Its vision models pick the main image on compliance and appeal, then generate lifestyle visuals and additional images from your product photos, keeping the product faithful to the original. Your team approves a preview before anything goes live. The new image URLs reach Merchant Center through a supplemental feed, segment by segment, with no photo shoot and no change to the source feed.

Descriptions and secondary attributes

The description adds use cases, materials, care, dimensions and what sets the product apart. Google takes its content into account to understand the product. A manufacturer description copied word for word, identical across every reseller, adds no distinct signal.

Secondary attributes (color, material, pattern, product_detail) let a product appear on long queries such as “long linen dress ecru size 12”, where purchase intent runs highest. The information often already sits in the description or on the product page, without having been extracted into the right fields.

Conversational attributes: preparing the feed for AI Mode

At Google Marketing Live in May 2026, Google documented six optional Merchant Center attributes, available in every country. They help AI Mode answer shopping questions asked in natural language, where a title and a description run out of material. Our complete guide to Google Merchant Center places them within the platform. This section covers how to produce the data and send it.

Six attributes, six sources in your catalog

Reviews and customer service tickets tell you what shoppers ask. The answers come from verified specs. Google bans prices, dates, promotions, lead times, company names and keyword lists from Q&A, and caps each question and each answer at 1,000 characters.

In TSV, a pair is written "question":"answer", and you can repeat the column to send several pairs:

id	question_and_answer	question_and_answer
KET-1700	"What is the capacity?":"1.7 liters, with a visible water gauge."	"How long does keep-warm last?":"Up to 15 minutes at the selected temperature."

Use TSV rather than CSV: questions and relationships are full of commas and colons.

The no-duplication rule

Google asks you not to repeat in these attributes anything already in description, product_highlight or product_detail, and not to send Q&A whose content already sits in a document_link PDF. Split the roles before you generate anything: the description presents the product, highlights sum it up, Q&A handles the questions that remain.

What these attributes change

They have no effect on approval: a malformed value makes the field unusable without disapproving the product. Google does not present them as a ranking lever for Shopping or Performance Max either. Their stated use is AI Mode and conversational experiences. Google recommends sending them in a supplemental feed joined on id, next to the primary feed that keeps price and stock. The Merchant API accepts these fields too.

Where to start

1. Isolate the 5 to 10% of products that matter most (traffic, margin, sales).
2. Fill question_and_answer for that segment.
3. Add related_product where merchandising data already exists.
4. Connect document_link for appliances, flat-pack furniture and DIY.
5. Handle item_group_title and variant_option for apparel and electronics.
6. Calculate popularity_rank last, from your sales data.

Feed Enrich generates Q&A from product data, specs and product pages, then publishes it as a supplemental feed, without touching price or availability in the primary feed.

Feed quality, Shopping, Performance Max and AI answers

A better feed moves three measurable levers. Coverage: more approved, eligible products. Matching: complete titles and attributes make a product eligible for more searches, long-tail ones in particular. Relevance: a better-described product shows up on searches closer to its real use, which lifts CTR and cuts wasted clicks.

Performance Max relies on the Merchant Center feed for its product formats, and your custom_label values set how fine your asset groups can get. A thin feed limits the algorithm whatever the budget.

AI-generated answers give you one more reason to look after the feed. Google describes its Shopping Graph as fed by Merchant Center data among other sources, and AI Mode draws on that information to answer shopping questions. Precise, structured facts (material, compatibility, use cases, common questions) make a product easier to recommend. In 2026 Google introduced attributes designed for these uses, covered in our article on Merchant Center conversational attributes. Other assistants, ChatGPT among them, also accept merchant product feeds, so the same groundwork serves several channels.

A better feed raises your chances of being shown and clicked on the right searches. ROAS also depends on prices, bids, the product page and the competition. Measure the effect on a test segment before drawing conclusions.

How to audit a Google Shopping feed in five steps

A useful audit combines three sources: Merchant Center diagnostics, the feed content itself and product-level performance in Google Ads.

1. Measure coverage. In Merchant Center, record the share of approved, limited and disapproved products. Rank issues by the number of products affected, then by commercial weight: a warning on your top 200 sellers comes before an error on 2,000 products that never sell.
2. Calculate fill rates. For each required and high-impact attribute, calculate the share of products where it is filled, category by category. A GTIN filled for 60% of products, or color missing on half the apparel range, stands out at once.
3. Review titles. Length, brand and product type in the first words, differentiating attributes, duplicates across variants. A sample of 50 titles per category shows the patterns.
4. Join with Google Ads. Add product-level performance to the feed. Products with no impressions often point to a data problem. Products that spend without converting sometimes raise a title relevance question.
5. Prioritize. Fix what blocks serving first, then enrich the highest-value categories. Our article on Shopping feed errors that inflate your ad costs lists the most frequent cases and how to fix them.

On a large catalog, Feed Enrich automates this audit: it scores the feed on about forty criteria (compliance, completeness, content quality, visuals, pricing, variants, conversational readiness) and ranks fixes by impact.

Manual, rules or AI: three ways to optimize a feed

Manual optimization suits a catalog of a few hundred products, or the best sellers of a large one. It gives the most editorial control and cannot keep up with new arrivals and seasonal changes.

Rules sit at the core of feed management tools: concatenate brand, type and color into the title, fill a custom_label based on margin, fix a price format. They run fast, give the same result every time and cost nothing to execute. They hit a limit when the information exists in no field: a rule cannot add a material the source does not contain.

AI takes over at that point. A model can read the description, the product page or the images to extract missing attributes, rewrite a title to the category structure or draft key points. It needs a framework: instructions per category, consistency checks and sample-based review by your team before rollout.

The three approaches work together. In Feed Enrich, deterministic transformations run as formulas with no AI cost, and generative AI steps in only where it adds new information. Google Ads performance joined to the feed lets you handle the products that matter first. If you are weighing approaches, our article on why most product feeds stay invisible explains the gap between syndication and optimization.

Multi-country feeds: what changes

Selling in several countries multiplies the details to get right. Each market expects its own language, currency and shipping costs. Word-for-word title translations miss the terms shoppers type: “hoodie” in the US, “sudadera con capucha” in Spain, and vocabulary that differs even between countries sharing a language, such as “sneakers” and “trainers”.

Size systems (size_system), units of measure and regulatory mentions also vary by market. Good practice is to keep one source feed and produce the market-specific attributes separately, often through supplemental feeds. Our article on optimizing multi-country feeds covers how to organize it.

Google Shopping feed checklist

Compliance

• Price, currency and availability identical in the feed and on the product page
• Updates at least daily, more often for volatile stock
• Valid, unique GTINs, brand filled, identifier_exists only where it applies
• Compliant images with no promotional text or watermark

Matching

• Titles structured by category, with product type and key attributes in the first words
• Descriptions written for your site, with use cases and specs
• color, size, material, gender and age_group filled where relevant
• product_type on three or four levels, google_product_category checked for sensitive categories

Steering

• custom_label values driven by margin, seasonality, stock and performance
• Variants grouped with item_group_id
• Optimizations published as a supplemental feed to keep the source feed intact
• Test and control segments defined before each optimization wave

AI answers

• Key points and specs filled (product_highlight, product_detail)
• Conversational attributes reviewed for priority categories

To see where your feed stands on each of these points, request a free feed audit from Dataïads: you get a scored report and a prioritized fix plan.

FAQ

What is a Google Shopping feed?

It is the file or API connection that sends Google Merchant Center the list of your products and their attributes (title, price, availability, image, GTIN, category). Google uses it to serve your Shopping ads, the product inventory in Performance Max and your free listings.

How do I create a Google Shopping feed?

Create a Merchant Center account, then add a data source: a file (TSV, XML or Google Sheets), your e-commerce platform integration or the API. Fill the required attributes, check diagnostics, then enrich titles and attributes.

What is the difference between a primary and a supplemental feed?

The primary feed creates products in Merchant Center. A supplemental feed adds or overrides attributes on existing products, joining on the product ID. Use it to optimize titles, labels or attributes without changing your site export.

How often should I update my feed?

Google treats a product not updated for 30 days as expired. Daily updates are the practical minimum, and prices or stock that change during the day call for an API connection.

Which images should I send in a Google Shopping feed?

A compliant main image in image_link (sharp packshot, plain background, no promotional text, watermark or border), up to 10 extra views in additional_image_link and up to 5 in-context shots in lifestyle_image_link. AI-generated visuals are accepted as long as they stay faithful to the product sold.

Do I need a feed management tool or a feed optimization tool?

A feed management tool distributes the catalog to several channels with rules. A feed optimization tool enriches the feed content to improve performance. The two work together: Feed Enrich plugs in as a supplemental feed alongside your existing feed tool.

Does my Google Shopping feed matter for AI Mode?

Google says its Shopping Graph draws on Merchant Center data among other sources, and AI Mode uses that information to answer shopping questions. Precise, complete attributes make your products easier to understand and recommend.

Should I fill Merchant Center conversational attributes?

The six attributes are optional and have no effect on product approval. They help AI Mode answer shopping questions. Start with question_and_answer on your priority products, sent as a supplemental feed, without repeating what the description and highlights already say.