Merchant Center and feed: so that goods pass without deviations
Setting up Google Merchant Center is not setting up an account, but bringing product data to the site's requirements: identifiers, mandatory fields by category, price and availability that match the page. We start the cabinet, collect or sort the feed, divide the deviation report into mass causes and individual causes. We start with the data, not the file. In the garden equipment store we measured, product submaps give 55,321 addresses — on such a volume, one system defect costs tens of thousands of lines. The term of work is 5–18 working days.
feed, feed rules, mandatory fields, analysis of the deviation report
What are we doing?
55,321
Dimension of the catalog of garden equipment
product addresses, 6 types of JSON-LD and no Product (31.07.2026)
in the cosmetics store, the outdated card gives 2,706 products against 3,947 in the canonical
Data source defect
analysis of the feed and deviation report, 2-4 working days, free of charge
Before the estimate
30
Guarantee
calendar days for completed work
Process steps
Transparent stages with approval at every step
Total duration:5–18 days
1
Analysis of data and deviation report
2-4 working days
We look at the download: identifiers, brand, categories, price match with the page. If the cabinet is connected, read the report and count the share of the catalog for each reason, not the number of messages.
2
Assembling or rearranging the feed
1-4 working days
We map the fields of the directory to the fields of the feed and fix the source of addresses: we take it from the database, not from the site map, until the map is checked. The identifier is taken from the article, not from the page address.
3
Categorization and mandatory fields
0-3 working days
Map of your categories to the product taxonomy of the site, summarization of sizes and colors into one dictionary, product status. The stage is dropped when the catalog is uniform and the fields are already set up separately.
4
Product and Offer marking on the card
0-3 working days
Price, currency, availability, terms of return and delivery in machine form - what the page confirms the feed. We mark only what is really on the page. The stage is dropped if the marking is standing.
5
Feed rules and additional feeds
1-2 working days
We remove mass causes at the entrance to the office: brand substitution, category template, cutting off items with zero price. What needs to be edited in the database for a long time, we put in an additional feed with the same identifier.
6
First read, report, connect to campaigns
1-2 working days
We wait for the platform to take the file, read the new report, remove what has been leaked. Next, we feed the sales campaigns and Performance Max and show your manager what to do with the next deviation.
Technologies & integrations
What we build on and what it connects to
Stack
Google Merchant Center is a feed, report and diagnostics cabinet. Restrictions: the account will be suspended for policy violations.
The XML commodity feed is the file that takes the site. Restriction: gives the state at the time of generation.
Rules of the feed — correction of data at the entrance to the office. Restrictions: change the file, not the page.
Additional feeds are a file on top of the main one. Constraints: Matches by id, different ids - silence instead of error.
JSON-LD Product / Offer - what the page confirms the feed. Restrictions: does not raise items.
MerchantReturnPolicy and OfferShippingDetails - machine return and shipping terms. Restrictions: without texts on the site, there is nothing to mark.
Update background tasks — scheduled regeneration. Limitations: on tens of thousands of items, the run lasts minutes.
Integrations
Google Merchant Center
Google Ads — trade campaigns and Performance Max
GA4
Google Search Console
accounting system for prices and balances
Rozetka and Prom — adjacent export
Meta directory
What's included
Complete list of work and what you get as a result
Assembling or rebuilding the feed in a format accepted by Merchant Center - items become available to trade campaigns and Performance Max
Identifiers: the seller's article from the database and the manufacturer's code where it is assigned - the product is not read as new after each regeneration of the catalog
Mandatory fields by category - size, color, gender, age group, state - are reduced to one dictionary, not to the text in the title
Categorization according to the product taxonomy of the site - the item falls into the group where it is searched, and not into the neighboring one
Rules of the feed for mass rejection reasons - the correction takes effect from the next read, without editing the catalog
Additional feeds for what is long or dangerous to manage in the database - matching with the main feed by ID
Reconciliation of regular and special price and availability between feed, markup and page - one source instead of three
Product and Offer markup on the card, if there is none - the page has something to confirm what is written in the feed
Reading schedule according to the speed of your warehouse plus checking that the site really took the file, and not yesterday's
Analyzing the report on deviations: we remove mass causes, single ones are given in a list with an indication of what needs to be corrected
Connection of the ready-made feed to trading campaigns and Performance Max together with the account manager
Transfer: accesses, a brief instruction in case of a new deviation, a guarantee of 30 calendar days
When this service isn't right
What's not included — so there are no surprises at delivery
Running advertising campaigns is a separate service
Account verification and correspondence with site support
Reshooting product photos
Writing titles and descriptions for the entire catalog
Legal texts about delivery and returns on the website
Who it's for
Situations where this service delivers results
Scenario 1 of 5
There is nowhere to launch trade campaigns: the items do not pass
The budget is approved, the campaign is created, but there are no products in it. We take the report and divide it into two: reasons that are removed by one rule for the entire catalog, and single ones that are ruled in cards. A long list usually boils down to three or four sources.
We'll review your situation in a free audit
Scenario 2 of 5
A smaller part of the catalog is passing, and it is not clear why
Half of the items are accepted, half hang in warnings, the campaigns do not show what is being sold. Here it is useful to count not the number of errors, but the share of the catalog for each reason: one reason for 60% of goods and forty reasons for the rest are different tasks and different estimates.
Scenario 3 of 5
Clothes, shoes, accessories — categories with mandatory fields
Size, color, gender, and age group must be separate values. If they live in the product name or in the description, the catalog is formally complete, and the items are not passed. We see if this data is in the database at all, and we say honestly when we have to start it manually.
Scenario 4 of 5
Prices and balances move faster than the feed is updated
The product ran out in the morning, it is in delivery until the evening. Each such click is paid and ends with a blank page. Resolved either by reading more often, or updating price and availability separately, depending on what your server can handle on a full regen.
Scenario 5 of 5
Feed gives the box module, and no one looked into it
Such a feed usually works and even transmits something. The question is where it gets the addresses and identifiers from: in the measured cosmetics store next to the working site map, there is an outdated vein, where there are 1,240 fewer products and every third address is repeated three times.
The rule in feed versus editing in the catalog: what is cheaper when
How this option differs from the alternative
Edit in the directory itselfOur approach
When it happensfrom a day to several weeks - depending on how many items we touchfrom the next reading of the file, that is, in hours
What fixesthe content of the data itself, including what is visible to the buyerformat, value substitution, position clipping, category template
What can't be fixednothing from this list remains, but also costs moredisagreement with the page and address that leads to another position
Where else will it workin all discharges at once, because the source is the sameonly in this cabinet - the rule is not transferred to Rozetka or Meta
What will happen when the account is movedremains with you along with the catalogthe rules remain in the old office, they are introduced anew
Free analysis of the feed and deviation report
Merchant Center rejects products not because of the file format, but because of the content of the data and because the page says otherwise. Analysis shows which errors are mass, which are isolated, and which of them can be corrected on your end.
What we measure
Unique identifiersSeller's article number and international manufacturer's code. At the store we measured, the EAN format code (sku 5904765561811) was in the card markings - this is a level that is accepted without question.
Product marking on the pageMerchant compares the feed with the page. In the measured store, the card gave 5 markup blocks and 21 types, among them MerchantReturnPolicy, OfferShippingDetails, and ShippingDeliveryTime — exactly the fields that are read from the page.
Product addressesOut of 150 random product addresses of the site map of the measured store, 50 (33%) returned 301 and all led to another position. For the feed, these are direct deviations.
Price, currency, availabilityThe three values must match in the feed, in the markup, and on the page. The most common reason for warnings is a promotional price that the page shows but does not show in the feed.
Categorization and mandatory fields by categoryClothing requires size, color, gender and age group; without them the position does not pass. Let's see if this data is in the catalog at all.
Current deviation reportWe divide into system errors, which are treated by one rule, and individual errors, which must be corrected by hand.
What you get
Analysis of the report on deviations with a division into mass and individual errors.
Evaluation of the share of the catalog that will pass without editing.
A list of fields that are missing from the data and where to get them.
Conversation for 30 minutes on analysis.
Timeline: 2-4 working days
Why is it free
Because analysis of the report takes two to four days, and the feed collected on uncorrected data gives an account with thousands of errors and campaigns that do not run.
What's next
Next, an estimate by stages: assembly or reconstruction of the feed, markup on the page, correction of addresses, rules of the feed. You see exactly what you're paying for.
Short form: your contact and site URL
Did not find your case?
Describe how it works on your side — we will tell you whether “Google Merchant Center and product feed” fits and what it means in your situation. No brief and no call: one question, one answer.
Contract, act and 30-day warranty
Every project gets a written contract: scope, deadlines, amount, acceptance procedure. After delivery — act and invoice, then 30 calendar days of warranty.
Sole proprietor & bank transfer
The contractor is a registered sole proprietor. Payment by invoice with closing documents.
Rights & access — yours
Code, design and materials transfer to you after full payment. Domain, hosting, repository and analytics are registered to you.
Client portal instead of email chains
During the project you get access to a portal: contracts, invoices, acts and project status in one place.
European clients
Among our work — projects for Norway, Bulgaria, Moldova and Spain.
Verifiable numbers
Every case in the portfolio comes with a link to a live site and a technical measurement.
Audit first, then pricing
There is no price list on the site intentionally: the scope of the same work differs multiples between clients.
We say "no" when unsure
If the task isn't ours or the deadline is unrealistic — we tell you upfront.
What affects the price
Why two seemingly identical tasks are priced differently
How many different reasons in the reportThe main variable. A thousand rejected items with one system reason is a few days and one rule. A thousand items with ten reasons — ten separate analysis, and the service is called the same.
Whether identifiers exist in the dataThe manufacturer's article number and code, which are already recorded, are simply mapped into the fields of the feed. If the article is repeated in different items or is not there, it is an adjustment in the database, not work on the feed, and it is counted separately.
Product categoryThe technique is almost based on what is already in the catalog. Clothing and shoes require size, color, gender, and age group as separate values — and if they're inside the name in the database, that's the longest line of the estimate.
Where does the data source come from?Export directly from the base is clean. Assembly from the site map or from the box module requires verification: in the measured cosmetics store, the parallel outdated map gave 2,706 products against 3,947 in the canonical one.
How the promotional price is arrangedOne discount for the entire catalog is transmitted by a separate field and closes in an hour. Prices by quantity, personal prices for authorized users and promotions with a timer — separate logic plus a check that the same amount is visible on the page.
Catalog size and update frequencyOn several thousand items, the feed run is imperceptible. It takes tens of thousands of minutes and loads the server, so we choose the schedule together with the person responsible for hosting.
Cases
Tasks and results in numbers — all metrics measured by us
online store of an official seller of garden equipment and spare parts
Task
Assess whether the catalog of tens of thousands of items is suitable for feeds to the product feed and what needs to be done in it.
Solution
OpenCart for HTTP/2, the address of the feed is explicitly declared in the robots.txt line Sitemap. Checked from the outside: submaps of the site page by page, markup, return speed, checkout.
Result
Measurement 07/31/2026: Commodity subcards yield 55,321 addresses (18 files of 3,000 records plus 1,321), categorical another 3,868. TTFB 840.5ms at 454.5KB HTML — slowest response of the lot, which for 55k items is expected. The main markup is rich: 6 types of JSON-LD, including Store, Place, GeoCoordinates. It is critical for the feed itself: there is no Product type, that is, there is nothing to compare the data of the feed with the page; PHP 7.3.33 is unsupported as of December 2021.
wholesale store of cosmetics for the beauty industry, catalog of about 4,000 items
Task
Find a data source to collect the product feed from and verify that it is complete.
Solution
OpenCart, a canonical sitemap at a separate address. They took apart both cards that the site gives and compared the number of goods between them.
Result
Dimension 07/31/2026: canonical map — 3,947 products, 166 brands, 47 categories, a total of 4,167 addresses. In parallel, the outdated map is alive: 7,782 entries, only 2,755 unique addresses, 2,454 of which are repeated three times, and about 2,706 products in it. The feed collected from this map would undercount about 1,240 items and bring duplicates. Another defect: there is no Russian-language address in the map, although two language versions are installed and the switch works.
online retail underwear store, catalog of over 110,000 items
Task
Understand why a large catalog with connected advertising gives less than it could in commodity formats.
Solution
OpenCart, the sitemap is made by an index of 42 submaps. The catalog was calculated using two independent methods — a site map and regular search.
Result
Measured on August 1, 2026: the site map gives 116,420 product records and 112,646 unique addresses, standard search — 115,456 items, method discrepancy 2.4%. There are 3,774 duplicate entries (3.2%) in commodity subcards, individual addresses are repeated up to 55 times. Advertising is connected: GTM, GA4 and Google Ads, the remarketing module gives the product_id. At the same time, the card has 5 marking blocks and 21 types, the price is UAH 204, the EAN format identifier, the presence of InStock — there is data for the feed, but the address source for it is unusable without cleaning.
What we need from you
We can't start without this — best to prepare in advance
1Access to Merchant Center or consent to create an account for your company.
2Access to the site and admin for editing data and markup.
3Manufacturer's articles and codes, if they are kept in the accounting system.
4Delivery and return conditions published on the website.
5List of categories that are not displayed in the feed.
6The person responsible for the advertising account and makes decisions on disputed items.
If something is missing — let us know, we'll help you gather it or do it as a separate task.
FAQ
Most frequently asked questions — with concrete answers
Where does the work begin — with opening an office or with data?
From the data. The cabinet starts up in half an hour and does not decide anything: everything that happens next is already laid out in your catalog. Therefore, first we look at the export — is there an article, is the brand filled, is the price the same in the database and on the page, and which categories need additional fields. When this is visible, it is clear what we are collecting in general: a feed from existing data or a feed plus work on data, and how much of the latter.
What are the rules of the feed and where do they not save?
The rule changes the data at the entrance to the office without touching the site: add a brand to the name, substitute a category according to a template, discard items with a zero price. Effective from the next reading, that is, in hours, and is the cheapest way to remove a mass error. It does not save in two cases: when the feed and the page do not match (the rule does not change what the robot sees on the page) and when the product address leads to the wrong position. One more thing: the rule lives in the cabinet, not in your directory — it won't move when you move the account.
We have clothes, the fields seem to be filled, but the items are still rejected
For clothes and shoes, it is not the presence of fields that are checked, but their content: size, color, gender and age group must be separate values, not text inside the name. "XL black dress" is a name for the system, not three attributes. The second place where it breaks down: sizes in an arbitrary record, when XL, x-large and 50-52 coexist in one catalog. We reduce them to one dictionary — either in the data itself, or as an additional feed, if the catalog is edited for a long time.
How often to update the feed and will the server handle it?
Depends on how fast your stock move. Once a day is enough until the product runs out in a few hours; further or more frequent readings, or a separate update of only the price and availability, without regenerating the entire file. A question that is not usually asked: how much does the generation itself cost. In the garden equipment store we measured, the product submaps give 55,321 addresses - on such a volume, a full run takes minutes and eats up a resource, so we coordinate the schedule with whoever is in charge of the server.
We also need a feed for Meta and Rozetka. Is it the same job?
The data source is the same, the requirements are different. Merchant has its own set of mandatory fields and its own taxonomy of categories, Meta has its own catalog format, and marketplaces have their own directories. We build one layer of export and make separate profiles from it - so the price and availability are taken from the same place everywhere and do not diverge. Each additional site is a separate line in the estimate, but cheaper than the first: the hard part, bringing the data to order, is already done.
Why does it say "price discrepancy" if we have the same price?
Because they compare not with your database, but with the page - and there the amount is different. Three usual sources: promotional price, which is applied already in the basket; discount for registration, visible only to authorized persons; different currencies or different prices in language versions. It is not treated in the office: the price on the page, in the markup and in the feed should be taken from the same field. As soon as it duplicates hands somewhere, it will disperse - the only question is when.
We have 50,000 products. How many of them will actually pass?
No one knows this in advance, and we do not name the figure until the analysis. Our recommendation before starting is different: do not download the entire catalog at once. Take one category for several hundred items, wait for the first reading and look at the report - mass defects are already visible on such a sample, and fixing them on two hundred items is many times cheaper than on fifty thousand. The volume is not the main difficulty here: the catalogs we measured at 55,321 and 115,456 items technically hold a full load.
How long does it last and what does the term depend on?
5-18 working days, six stages. Five — when the articles are kept in the account, the marking on the card is correct, the catalog is homogeneous: it remains to collect the feed, set a couple of rules and analyze the first report. Eighteen — when there are no identifiers, the categories require additional fields, and we start the markup from scratch. We name the amount of changes after a free analysis, because neither we nor you can see it in advance.
What should I do with my campaigns while the feed is fixed?
If the campaigns are already running, we do not stop them for processing. First, what passes the test is output to the feed, campaigns live on that part of the catalog, and the rest are added in batches as they are fixed. So the advertising account does not lose the accumulated data. If there are no campaigns yet, on the contrary, we calmly bring the feed to an acceptable state, and launch it after the first clean reading.
Send the website address and access to the account, if it is already established.
The answer is a breakdown of the deviation report with a division into mass and single reasons, an estimate of the share of the catalog that will pass without changes, and a list of fields that are missing. Then the estimate by stages.
From measured casesDimension 07/31/2026: canonical map
There is no price list on the site on purpose: the same work differs several times over between two clients, and a “from” figure explains nothing in that case. First a free audit — we count your pages, duplicates and speed — then we name the sum and the deadline and fix both in the contract.