Audit a Shopify catalog before a Google Merchant Center review

The first reader is a merchant whose Merchant Center account was just suspended for "misrepresentation" or whose products were disapproved in a wave. The email names a policy. It does not say which of the six hundred products triggered it, and the appeal form asks what you changed. You need the list before you can change anything.

The second is a merchant about to connect the Google & YouTube channel for the first time, or about to scale spend on Shopping ads, who would rather find the problems than have Google find them. A catalog imported from a supplier has supplier text in the descriptions, promotional words in the titles and watermarked images, and every one of those is a known disapproval reason.

The third is an agency or a freelancer running the same check across several client stores before a campaign. The instruction is the same on every store; the report format is the same; the fixes are scoped per store. This is the job that benefits most from being a sentence rather than a spreadsheet macro.

Merchant Center itself has a diagnostics page: Products, Needs attention, filter by issue. This is the list of what Google has already rejected. It is accurate and it is late: it tells you nothing about the products Google has not looked at yet, and during a suspension the whole feed is offline, so the per-product list stops updating.

In the Shopify admin, you can filter Products by "Sales channel: Google & YouTube" to see what is published to the channel. There is no filter for "has a barcode", "description mentions a supplier" or "title contains an exclamation mark". Each of those is an export, a spreadsheet, and a formula per column. Image checks are not possible in a spreadsheet at all; someone opens the images.

Feed management apps sit between Shopify and Google and can rewrite fields on the way out, which hides the problem from Google without fixing it in the store. Some flag missing GTINs and brands. None of them read the picture. The audit here reads the store, so what it finds is what the customer and the reviewer see, and the fixes land in the store, not in a feed layer.

A skincare merchant wrote the template after a disapproval wave: audit all active products on the Google channel for image compliance, do not make changes yet, go through every product, report totals and the first 50 flagged. That prompt has the four things an audit instruction needs: the scope, that nothing changes, that every product is read rather than a sample, and what the output looks like.

Say "do not make changes" even though a read cannot write. It tells the tool to plan a report, not a fix, and it keeps the audit and the fixes as separate steps with separate approvals. The fixes come later, one check at a time.

Each row is a Merchant Center reason, what in the Shopify data predicts it, and the job that fixes it. Run them as one instruction or one at a time; the report is the same shape either way.

Totals first: scanned, published to the channel, passed, flagged, unable to verify. "Unable to verify" is honest; it means the data was not there to check, for example a product with no images at all cannot fail the image check, and it is counted separately so the passed number is not inflated.

Then the first 50 rows per check, with the product title and the specific reason: "title contains 'FREE SHIPPING'", "description mentions AliExpress in paragraph 2", "image 3 has text overlay". Fifty is enough to see whether the flags are real before you ask for the full list, and the full list is one more instruction away.

Promotional text in titles and supplier names in descriptions are the two fastest fixes and the two most common suspension reasons. "FREE SHIPPING", "SALE", "50% OFF", three exclamation marks and ALL CAPS in a title are all flagged by Google's title policy, and all are found by a plain text check. The fix is a title rewrite by rule, with a preview showing the old and new title per product, and it is a job merchants approve in one sitting.

Descriptions that mention AliExpress, Alibaba, "dropship", "ships from China mainland" or a competitor's brand are misrepresentation flags. The fix is a per-product description write that removes the sentence and keeps everything else, previewed as the whole new description so you can see the formatting survived. The imports cleanup case runs that job.

Thin descriptions, under about forty words, are not a policy violation but they are a quality signal Google uses, and they are the same products that fail the SEO check. Fix them with the description job after the policy items are done.

A missing GTIN (the barcode field on the variant) does not get a product disapproved by itself, but it puts it in "limited performance", which in practice means fewer impressions. The audit lists the variants without one. The numbers come from your supplier; the tool does not make them up, and a made-up barcode is a worse problem than a missing one.

Brand is read from the vendor field. A vendor that is your store name on a product you did not make, or an empty vendor, is a missing-brand flag. Filling vendor by rule ("every product with SKU starting KS- is Kestrel Supply") is a small job with a preview.

A compare-at price lower than the price, or a price of zero, is a price mismatch. Active products with zero stock at every location are an availability mismatch: Google sees "in stock" in the feed and "sold out" on the page. Both are found by the audit and fixed by the price and inventory job, or by setting the product to draft until stock arrives.

Image checks run through a vision model in batches. The question the model answers is narrow: is there promotional text, a watermark, a supplier or marketplace logo, or a collage on this picture. Text that is physically part of the product, a brand name on a box or a slogan on a shirt, is not a flag. The output is one line per flagged image with the reason and the image position on the product.

This is the check nobody does by hand on a large catalog, and it is the one Google's reviewers do look at. It is also the one where the tool is a screen, not a judge: a flag is an image to open. Expect some flags on product photos with legitimate packaging text, and expect to clear them by looking. What you get is a short list instead of every image in the store.

The fix for a flagged image is a new image. That is the merchant's job, or the supplier's. The audit tells you which products need it, in order of how many flags they carry, so the photo work is scoped before it starts.

The report has an order of operations built into it. Follow it and the fixes take a few sessions instead of a few weeks.

Each fix is its own job with its own preview. Do them in the order above, one approval each, and re-run the audit at the end. The audit is cheap; the fixes are where the approvals go, and keeping them separate means a wrong title rewrite is reverted on its own without touching the description job.

Re-run the same audit instruction. The report should show the flagged counts at zero for the checks you fixed, and unchanged for the ones that need the supplier or a photographer. That second report is what goes into the appeal: what was found, what was changed, what remains and why.

Google re-crawls the feed on its own schedule, usually within a day for a connected channel. Merchant Center's diagnostics page updates after that. A suspension appeal is reviewed by a person and takes longer; the report gives them a specific list of changes, which is what the appeal form asks for.

The audit writes nothing, so there is nothing to revert. Each fix job is its own history entry with its own saved values, and each is reverted independently. A title rewrite that went too far is put back without touching the description job that ran after it.

The audit reads Shopify. It does not read Merchant Center. If Google has already disapproved a product for a reason the Shopify data does not show, such as a landing page that fails to load in Google's region, the audit will not see it. Merchant Center's own diagnostics page is still the source for that.

Barcodes, images and brand names are facts about the product. The tool reports that they are missing and does not invent them.

The vision pass is a screen, not a judge. Google's reviewers make the call; the audit tells you where to look first, and it will flag some images that are fine.

Policy names change. The eleven checks map to the reasons Google gives most often; a new policy is a new check, added when it shows up in merchants' reports.

These resolve to the same report. Add "and the first 50 flagged" if you want the rows, not just the totals.