AI Image Matching for Shopify: Match Messy File Names to Products
How AI image matching works in Shopify: it reads messy file names like a person would, pairs them with SKUs or barcodes, and you review every match.
By Craftshift Team · · 5 min read
AI image matching reads the file names of your product photos the way a person would and pairs each one with the right Shopify product or variant, even when the names are inconsistent. In Smart Bulk Image Upload it works on file names, not on what is in the picture, and you review every match before anything is uploaded.
What is AI image matching in Shopify?
Normal matching is rule-based: the file name must equal the SKU (AB-1234.jpg), contain it, or be contained in it. That is predictable, but it breaks the moment names are messy.
AI matching (currently in beta) replaces those fixed rules with a reader. You still choose the field to match by: SKU, barcode, product title or metafield. The AI then decides which product’s value a file name refers to. Variant ID has no AI option, because a number leaves nothing to interpret.
It is a file-name reader, not image recognition. It never looks at the photo itself.
What kind of messy names can it handle?
These are the situations where exact and “contains” rules usually fail:
| File name | Why rules struggle | What AI can do |
|---|---|---|
cct_navy_01.jpg for SKU CCT-NAVY-S | Different separators, an image number | Reads it as the navy variants of product CCT |
NAVYCCT.jpg | Words run together and out of order | Splits it into words and matches the navy CCT variants |
Wilcats-pink.jpg for a “Wildcats” product | A typo | Reads it as a close spelling of a word in your catalog |
43443-86-1,.jpg for SKU 43443-86-1-S | Stray punctuation and a number glued on | Finds the SKUs that share its code |
maxidresssummer.jpg | No separators at all | Splits it into the words of the product title |
It also understands image numbers and views. _2, -02, (2), front, back and detail are treated as the order of the photos, not as part of the identifier, so the main image ends up first.
How does it work, step by step?
- Clear-cut names are matched instantly. If a file name equals exactly one identifier, apart from case and separators, there is nothing to decide.
- Other names get a shortlist. The app finds the likely candidates in your catalog for each file.
- The AI picks from the shortlist. It chooses one candidate or “none”. It cannot invent a product that is not in your store.
- Low-confidence picks stay unmatched. A match is only kept when the AI is confident.
- You review the result. Everything appears in the same preview as any other matching type.
The AI reads your store’s own data, such as SKUs, barcodes, titles and variants, plus the file names. Your photos are not analysed.
Does one photo go to all sizes?
It can. Many catalogs name photos after a style or colour (8946N86-B807_04.jpg) while the store has one variant per size (8946N86-B807-0, -2, -4). AI matching recognises that the name stands for the whole group and can attach the photo to every size of that product. For a name that is genuinely ambiguous, it asks the AI instead of guessing.
AI matching or rule-based matching: which should you use?
| Rule-based (exact, contains) | AI matching (beta) | |
|---|---|---|
| Names follow a convention | Best choice | Not needed |
| Names are inconsistent or hand-typed | Needs renaming first | Designed for this |
| Results | Identical on every run | Judged per name, answers are reused on re-runs |
| Speed | Instant | Clear names instant, others judged in batches |
| Cost | Included | Free stores: 1,000 images per day and 5,000 per month. Paid plans: no cap |
If you control the naming, rule-based matching is simpler and fully deterministic. Rename once with a clear convention and use Match by exact SKU. If the photos come from many suppliers or years of history and renaming thousands of files is not realistic, AI matching saves the clean-up.
A third option is to store each photo’s existing code in a product metafield and match on that. See all matching options.
How do I review AI matches before uploading?
Matching is read-only. Nothing touches your store until you press Start upload. On the results page you can:
- See every pairing with its thumbnail, file name, variant, position and the identifier it matched on.
- Unmatch anything that looks wrong with one click.
- Match leftover images manually, or export the unmatched list as CSV.
- Fix a SKU in Shopify, press Clear cache and match again.
Spot-check a sample of matches, especially for products with similar names, and look at the unmatched list. If a pairing is wrong, unmatch it. While an AI run is in progress, unmatching, manual matching and upload are disabled until it finishes. The preview and upload guide covers the rest of the flow.
What should I watch out for?
- It is a beta. Always read the preview. The AI is good at messy names, but a review is what makes the upload safe.
- Replace mode removes existing images. If you replace rather than add, check the replace vs add guide first.
- Keep the folder clean. Stray files such as documents show up as unmatched entries.
- Re-run after renaming. Use Match again after changing file names, and Clear cache after editing products in Shopify.
Summary
AI image matching lets you upload photos with inconsistent names without renaming them first. It reads file names like a person would, pairs them with your SKUs, barcodes, titles or metafields, leaves uncertain ones unmatched, and shows everything for review before upload. When your names follow a convention, rule-based matching is still the simplest choice. When they do not, AI matching saves hours of clean-up.
Try it on your own folder: install Smart Bulk Image Upload from the Shopify App Store. Matching is read-only, so you can see the results before you upload anything.