GTIN, EAN and MPN explained: help shoppers find the right product

MPN means Manufacturer Part Number: a reference a manufacturer assigns to identify one product within its range. The phrase “MPN number” is common, but the reference can hold letters as well as digits, and Google Merchant Center accepts up to 70 characters.

An MPN is different from the stock code your shop creates. Your internal SKU helps you manage inventory; the manufacturer's part number identifies the manufacturer's product. Two references that look alike are not proof they do the same job.

Merchant Center requires an MPN for a product with no manufacturer-assigned GTIN, and its MPN guidance says that unless you are the manufacturer you should not submit a value you created. Use the manufacturer's value, check the requirements for your product type, and do not treat every internal SKU as an MPN.

GTIN stands for Global Trade Item Number, and EAN is one of the numbering formats inside that system. Merchant Center's GTIN page lists the formats:

• EAN, the European format, is a 13-digit GTIN-13.
• UPC is the 12-digit format used in North America.
• ISBN is the format for books.
• ITF-14 is the 14-digit format for multipacks.

So an EAN is a kind of GTIN rather than an alternative to it.

An MPN belongs to a manufacturer's own numbering, so read it together with the brand and the rest of the product record. A retailer SKU belongs to your stock system. Keep the three in separate fields, even when a supplier file puts them under a vague heading such as “product code”.

Google's GTIN guidance asks you to submit a GTIN only when you are sure it is correct, and never to guess or make one up. Some products have none: a store-brand item, or one you are the only seller of, goes to Merchant Center with its brand and MPN instead. An empty field needs investigating; it is not permission to invent a plausible barcode.

A value can have the expected length and still belong to a different product. Format checks catch a mistyped or truncated code, but they cannot tell you whether your listing describes the item that code was assigned to.

Take an illustrative retailer, not a client: it sells a 500 ml insulated bottle in black and in green, plus a two-bottle bundle. The supplier file holds one code labelled “bottle”. Before copying it to all three listings, the team needs to know what that code identifies: a product family, one colour, or one packaged offer.

Ask the supplier for a mapping between each sellable item, its manufacturer reference and any assigned GTIN, and keep the source document with the date you checked it. Google's guidance expects each colour or size variant to carry its own GTIN, so the black bottle's code does not belong on the green one just because they share a description.

The bundle needs its own answer. A code printed on one bottle does not identify the pair you sell. Merchant Center, for instance, treats a manufacturer's bundle and a bundle you put together differently, so confirm the rule for your offer and each destination rather than applying one rule everywhere.

The same discipline matters wherever a feed is read by someone else. SO-ME, the consumer shopping platform Sighted powers, receives feeds from ASOS, REVOLVE, Vinted, M&S, New Look, Debenhams and EGO, and the attributes each carries and the way each categorises the same kind of product differ from one retailer to the next. A field meant to identify the item has to be right at source, because a system reading the feed has nothing to check it against.

1. Export the records customers can buy

Include the internal SKU, brand, title, variant options, pack quantity, supplier reference, current identifier fields and landing URL. Work at the level of the sellable item, since a parent row can hide errors in its variants.

Keep identifiers as text: spreadsheet software can strip leading zeros or show a long number in scientific notation. Save a raw copy before cleaning the file, so you can trace any change you did not expect.

2. Separate missing, invalid and conflicting values

These are three different jobs, so do not fold them into one generic enrichment request:

• Missing means no value was supplied.
• Invalid means the value fails a format check.
• Conflicting means two records disagree about which item a value belongs to.

Log each exception with the SKU, the issue, the evidence needed and an owner: a missing supplier reference goes to supplier or merchandising review, and a value damaged during export goes to whoever owns the integration.

3. Verify each identifier against an accountable source

Use manufacturer documentation, the product's packaging or an approved supplier record. If the evidence disagrees, hold the change until the disagreement is resolved, and never pick whichever value appears on more websites.

Record whether the evidence covers this exact variant and pack: “confirmed against supplier mapping for black, 500 ml, single item” is a useful approval note, where “looks correct” is not.

4. Follow the value through to publication

Check the approved record against the store, the page's structured data and every feed you send. Look for a transformation or default rule that could swap in another field, such as a SKU copied into the MPN column. The guide to feeds, schema and product pages explains what each of those carries.

OpenAI's feed rejects a GTIN that fails its check digit, one of the checks in the ChatGPT troubleshooting guide. Then check the output again after publishing: a corrected source file is not a finished repair while the old value is still live in a destination.

AI can organise supplier records, flag inconsistent labels and propose matches between your products and a supplier's references for someone to investigate. Ask it to return the source for each suggested match and to list any differences it could not resolve, such as a pack size that appears in one record and not the other.

Do not approve an identifier because a model recognises a product name. Near-identical titles can belong to different revisions, regional models or pack sizes, and a confident match is still a guess until a document confirms it. Suggested matches belong in a review queue until the evidence establishes the connection.

The product data optimisation guide shows how to tie each source fact to an approved change. Identifiers deserve the same discipline as material, dimensions and compatibility claims.

Accurate identifiers make your product record more dependable, because a system can tell which item an offer refers to and match it with other information about that item. They do not guarantee that an assistant will recommend your offer: whether the product suits the request, what else is known about it and the competing offers still decide that.

Treat identifier correction as an information-quality fix first. Confirm the mismatches are resolved, then watch visibility and sales separately, with the AI shopping measurement guide as the method, so a change in recommendations is not credited to identifiers when price or stock also moved.

Is an MPN the same as a SKU?

No. The manufacturer assigns the MPN; your SKU is your own stock reference. Store them separately unless you have checked that your business deliberately uses the manufacturer's reference as its SKU.

Can you make up a GTIN for a product without one?

No. Find out whether the manufacturer has assigned one, then follow the destination's rules for a product with no GTIN. In Merchant Center that means leaving the GTIN blank and sending the brand and MPN, or, for a custom-made item with no identifiers at all, setting identifier_exists to “no”.

Should every variant use the same code?

Do not assume so. Check the identifier for each sellable variant. Google expects each colour or size to have its own GTIN, while clothing sizes often share one MPN, so a family reference and an item identifier can do different jobs.

1. Export the records customers can buy, at the level of the sellable item, with identifiers kept as text and a raw copy saved.
2. Sort the exceptions into missing, invalid and conflicting, and log each with the SKU, the issue, the evidence needed and an owner.
3. Verify each identifier against manufacturer documentation, the packaging or an approved supplier record, and note which variant and pack the evidence covers.
4. Check the approved value in the store, the page's structured data and every feed, then check again after publishing.
5. Let AI propose matches with their sources, and hold every suggestion in a review queue until a document confirms it.
6. Once the mismatches are resolved, watch visibility and sales separately with a defined method.

Sighted reads each variant's GTIN from your store (the barcode field on Shopify) and never makes one up: on Shopify, a barcode removed from a variant is cleared on the next sync. Whether a value belongs to the item you sell is a check against your supplier's records, and that stays with your team.

• Import flags a product GTIN with the wrong length or a failed check digit, and the feeds leave out any GTIN that is not 8, 12, 13 or 14 digits rather than send an invalid item.
• Enrichment reads GTIN and MPN as facts about the product and does not write them.
• Sighted's feeds send each variant's SKU from your store as its MPN, so if your SKUs are your own stock codes rather than the manufacturer's part numbers, check that column before you publish.
• The free AI visibility audit shows which of your product pages carry no GTIN in their structured data, and the enrichment that runs on the rest of the record is the technology behind SO-ME.