A client shows me his back office: fourteen hundred references, three channels, everything marked “active”. We pick twelve products at random and go looking for them by hand on each marketplace. Seven show up on the first, nine on the second, four on the third. Nobody on the team had noticed a thing.
“Active” means the listing was sent. Not that it was published, not that it is visible, not that it is complete.
No catalogue survives being copy-pasted wholesale onto three marketplaces. And no team keeps three listings up to date by hand over time.
The way out sits between the two: one product record, and one translation rule per channel. Here is how I build it, and the four places where it breaks in practice.
1. There is no such thing as a shared title
This is the first illusion to drop, and usually the most expensive one. The three channels I see most often at my clients do not ask for the same length, or the same grammar.
| Channel | Title length | The constraint that catches people out |
|---|---|---|
| Amazon | 200 characters, 125 in some apparel categories | A word cannot appear more than twice, plurals included |
| eBay | 80 characters, plus a paid 55-character subtitle | The field is rejected beyond that, not truncated |
| Google Shopping feed | 150 characters, of which 70 are actually read | No price, no shipping cost, no company name |
A hundred and ninety character title written for Amazon arrives cut off at eighty on eBay. What gets lost in the cut is almost always the end of the string: the size, the colour, the volume. Precisely the one piece of information that tells the listing apart from its six variants.
Amazon tightened its rule in January 2025: a 200-character ceiling, decorative special characters banned, and above all any given word limited to two occurrences, with plurals and derived forms counting as repeats. Non-compliant titles are flagged to the seller, who then has fourteen days to fix them. Google, for its part, refuses price, promotion dates, shipping costs and your company name in the title, and warns that shoppers generally only see the first seventy characters.
So I no longer write “the title”. I write fields, and one rule per channel assembles them: brand, product, distinguishing attribute, variant, in that order. The narrowest channel sets the core, the wider channels append behind it. A valid eBay title becomes the opening of a valid Amazon title. The reverse is never true, which is why you never start from the longest one.
2. A GTIN cannot be invented
The GTIN (Global Trade Item Number), the product barcode, is the only thing that tells three platforms they are talking about the same object. It is also the field I see abused the most, because a feed that rejects empty cells pushes teams to put anything at all in there.
I have found internal references copied straight into the gtin field. The immediate consequence is a publication refusal. The slow consequence is worse: on a marketplace with a shared catalogue, a borrowed identifier attaches your offer to someone else’s product page, with their photos and their reviews. You end up selling under a listing you do not control.
Google’s rule is blunt and holds everywhere: do not invent, do not guess, do not reuse the identifier of a similar product. If the code exists but you do not have it, leave the field empty rather than fill it wrongly. Two details that cost hours when you find them late: every colour and size variant needs its own code, and the prefixes 2, 02 and 04 fall in a restricted range that Google refuses.
An empty cell is fixed with one call to the manufacturer. A wrong identifier follows you for months.
3. The same photo does not pass everywhere
The main image is the most tightly specified field of the lot, and the quietest when it fails. Amazon requires a pure white background on the main photo and suppresses the listing in search until it complies, without deleting it: it stays in the seller’s catalogue, simply unfindable. That is why nobody notices.
On the Google feed side, the list of things that get an image refused is long and very concrete: no promotional banner, no price, no free shipping mention, no watermark, no barcode, no border, and no placeholder visual while you wait for the real shot. The product has to fill between 75 and 90 percent of the frame. The minimum size moves to 500 pixels per side on 31 January 2027, with 1500 recommended.
The trap I fix first appears in no brand guideline. When you replace the content of an image but keep the same file address, Google can take up to six weeks to notice, against roughly three days if the address changes. Making the file name change with every new shoot is the most profitable line in a naming convention.
In practice I have two sets produced: a pack that meets the strictest rule, white background and cut-out product, used as the main photo everywhere, and a lifestyle pack reserved for secondary images. One shoot, two outputs. Cheaper than three shoots, and it avoids the silent suppression.
4. The category tree nobody wants to own
Every marketplace has its own tree, and every node on that tree calls for its own required attributes. The same product does not sit in the same place from one channel to the next, and the wrong node does not throw an error: it produces a perfectly correct listing that the shopper’s filters never meet.
That mapping table is the one deliverable on the project I refuse to hand to developers. Choosing the category means choosing who you are compared against, under which filters, and at which commission rate. It is a merchant’s decision, not a configuration step. It gets made with the person who knows the margins, and reviewed every quarter, because the trees move.
What the record looks like
One row per product, and three families of fields. Identity fields, which are not negotiable and never change from one channel to another: barcode, brand, material, dimensions, weight, warranty. Selling fields, which vary by channel: category, price, lead time, assembled title. Status fields, which are never typed by hand but read back from the platform: published, rejected, reason for rejection.
Every field has a named owner. That is the boring part, and it is the part that survives. Without an owner, the listing goes back to being a copy-paste within six months, because whoever enters a new product on a Friday evening takes the shortest route.
On pricing and the Buy Box, which run on a different mechanism, I set out the reasoning in the article on cannibalisation between channels. And for everything to do with organic ranking once a listing is live, how Amazon’s algorithm actually works covers the ground.
The most expensive mistake
Counting listings sent instead of listings visible. A dashboard showing “1,400 active references” says nothing: it counts what you pushed, not what a shopper can find. The only number that matters is the publication rate per channel, meaning listings actually visible against listings sent, with the most frequent rejection reason next to it.
What I watch once it runs
Three numbers per channel, once a week. The publication rate. The top rejection reason, because it is nearly always the same one and it gets fixed once and for all inside the assembly rule. And the number of published listings that have had no impressions in thirty days, which points at category errors better than any audit.
The rest lives in the analytics setup, channel by channel and never in aggregate. A catalogue growing in total volume can lose an entire marketplace without the overall curve moving.
In the crucible: one substance, three moulds
A product does not change nature when it changes window. The moulds differ, and each one has its own temperature. The work is to melt the substance once, cleanly, then pour it three times without ever remaking the alloy. The catalogues that hold are the ones where there is only one place to fix when a platform changes its rules.
And they do change. That is the one thing you can count on when you open a third channel.
Pushing the same catalogue onto several marketplaces?
Send me your feed export and the list of your channels. I will give you back the real publication rate per platform and the three fields blocking the most listings. At worst you confirm everything goes through, at best you recover references that have been unsellable for months over a formatting error.