Your data source has errors.
Fix them once, for good.

Before you upload

Does anything change in my shop?

No. FeedMender writes one new file and hands it to Merchant Center. Your CMS, your export and your primary data source are untouched.

Do you need access to my Merchant Center account?

No. You give us a file, we give you a file. There is no permission to grant, which also means we cannot see what Google finally decided about your products. We tell you what your data says. Merchant Center tells you what Google did with it.

Where does my file go?

To our server, where it is scanned and then kept for up to six hours, long enough for you to read the report and decide, because the repair needs the file again. After that it is deleted automatically, and there is a button on the report to delete it sooner. We keep no copy of the corrections either: those go into a spreadsheet in your own Google account, and Merchant Center reads them from there. Image URLs are fetched from our server, so your own shop sees those requests rather than you.

Couldn't I do this myself with an LLM?

Partly, and we use one ourselves. Rewriting a thin title is a language job and a model is good at it, so that step does go to one. The rest is not a language job. A GS1 check digit is arithmetic, and a model will wave through a barcode that fails it. Whether an image is 480 by 480, or answers at all, is only knowable by fetching every URL in the file. Whether AdsBot-Google is disallowed sits in a file on your server. Our scan itself never calls a model at all.

The bigger problem is that it becomes a loop. A model corrects what you thought to ask about and leaves untouched the ninety-nine faults you did not know were there, so the way you find the rest is to upload and let Google refuse them. That verdict arrives on Merchant Center's schedule rather than yours, and by the time it does, your shop has exported the file again over the top of your work. Every round waits on Google. We run the whole list in one pass and then check that the corrections did not break each other.

Being right is harder than it looks, too. Ours once proposed "Blue" for a television, because the colour column read Blue on most of the catalogue. Grounded in the merchant's own data, and still wrong. What stopped it was a relevance filter and a word by word check against the source row, both written after we watched it happen.

And a correct rewrite still has to arrive somewhere: as structured_title with the generated content declared, knowing Google uses title when both are present, in a supplemental source that cannot add products and expires 30 days after its last refresh. The scan is free, so the quickest way to settle all of this is to run it and see whether what comes back is what you would have found.

Do you invent product information?

No. Every generated value is checked word by word against your own row. If a claim is not in your data it is discarded and the original is kept. Rewritten titles are declared to Google as generated content.

What formats do you read?

csv, tsv, xlsx, xml, json, and zip or gz archives of any of those. Dutch, French, German and English column headers are recognised without configuration.

Are you part of Google?

No, and nothing here is endorsed by them. What we do have is seven years of Google Ads work in house, across a great many Shopping accounts and a great many feeds with errors in them. FeedMender is the checklist we kept rebuilding by hand.

After the scan

What if my feed turns out to be fine?

Then you pay nothing. The scan says so, there is nothing to download, and that is a perfectly good outcome.

What do I have to do in Merchant Center?

Put the file we generate into a spreadsheet in your own Google account, and add that sheet in Merchant Center as a supplemental data source. Google reads it every 24 hours from there.

It does not have to be public and we never see it. Merchant Center connects to your sheet through your own Google account rather than through a link you publish, so the corrections stay inside your workspace, under your retention and your access rules. On a tightly managed Workspace an administrator may first have to allow the Merchant Center service account, which is one setting rather than a project.

The report comes with it: the attribute rules to paste for the fields we deliberately do not pin, and the primary source, feed label and language your sheet has to match.

What happens when my feed regenerates tonight?

Google applies the supplemental source on top of your primary source every time it processes it, matching on id. Your export can regenerate as often as it likes.

My prices have commas in them. Do you fix that or not?

We fix it, with an attribute rule rather than with a value. A value would pin the price we saw: drop the product to 99.95 next month and the supplement would keep serving 119.95. A rule replaces the comma and appends the currency in whatever your feed says tonight, so it holds for every sync after that. The report gives you the rule and where to paste it. Stock wording works the same way.

I add new products every week. Are those covered?

Partly, and it is worth knowing which part. The rules live in your account rather than in a file, so they run over every row your shop sends from then on, including products that did not exist when we scanned. The corrections that had to be written per product, a rebuilt title or a repaired barcode, cover the products that were in the file we read.

The report is the other half of the answer. Almost nothing we find is a one-off: your export makes the same mistake in the same field every time, and the report says which faults are systematic and where they come from, so they can be fixed at the source and new products arrive clean. Scanning again is free, so you can check whenever you like, and you only pay if there is something new worth repairing.

How soon will I see the difference?

Not immediately. Google refetches your sources on its own schedule, reprocesses the products, and lifts the disapprovals it agrees with. Your part is done the moment the data source is in place. The rest is Google's timing, not ours or yours.