Elbuz suggests a marketplace category for each of your unmapped categories: AI reads the category path, its sample products and your existing links to other marketplaces.
When you need this:
- Hundreds of categories whose names don't match the marketplace — AI matches by meaning, not by exact name.
- You're entering a new marketplace — Allegro, eBay or Kaufland get suggestions hinted by your Google or Rozetka mapping.
- You're afraid of ruining manual work — applying writes only to empty rows; manual choices stay.
- No time to wait — matching runs in the background; you can close the window.
Where to start: “Marketplace mapping” → marketplace tab → “✦ AI match”.
What AI matching can do
- Matching by meaning — the marketplace category is chosen by the category path, its products and other marketplaces, not by a name match.
- Suggestions with a confidence score — those at 85% and up are ticked for applying; the rest wait for your decision.
- Applying without overwriting — what you mapped manually doesn't change.
- Background work — up to 3,000 categories per run; after a failure it continues where it left off.
- Nineteen marketplaces — Rozetka, Prom, Epicentr, Ceneo, Google and marketplaces with a downloadable dictionary.
How to run AI matching for a marketplace
Where: Directories → Categories → the “Marketplace mapping” tile → marketplace tab.
- Go to the marketplace tab. For eBay, Amazon and eMAG, choose a country in the list next to the tabs — matching runs for that country.
- To match only some categories, set a filter in the category table: AI then takes only the rows loaded into the table — up to 250, as on one table page.
- Click “✦ AI match”.
- Elbuz counts the categories and asks: “Match Google categories for 190 unmapped categories?”. Confirm — matching runs in the background.
- Review and apply the suggestions (how).
1
Google tab2
✦ AI match button3
AI suggestions button4
Matching status line- Matching runs for the marketplace of the open tab — Google here.
- ✦ AI match counts categories without a mapping and, after you confirm, runs matching in the background.
- AI suggestions opens the list of suggestions for review at any time.
- Status line: AI matching finished, suggestions to review: 10.
Matching takes only active categories that aren't mapped to this marketplace. It doesn't touch mapped rows — yours or those set by auto-matching by name.
Without a filter, matching includes all such categories, including those with no products: the window's table doesn't show them. Disabled categories appear in the table but aren't included in matching.
The number in the confirmation also counts categories that already have a pending suggestion: matching skips them. So after the first run, apply or reject the suggestions first — then a repeat run handles the rest.
One run takes the first 3,000 unmapped categories — including those that already have a suggestion. Until this batch's suggestions are applied or rejected, a repeat run takes the same 3,000 and matches nothing new. Match a large catalog in parts: with a table filter by branch or in batches of “run → apply or reject → run”.
1
How many categories go to matching2
Start matching- The app counts active categories without a mapping on this marketplace in advance.
- After you confirm, matching runs in the background; Cancel starts nothing.
What AI uses to choose a category
First, Elbuz finds categories close in meaning in the marketplace dictionary — up to twenty candidates along with their parent sections. Then AI picks one from this shortlist that fits all products in the category, sets a confidence score and briefly explains the choice. What it takes into account:
- the full path of your category (“Computer components / Cooling systems / Air cooling”) and its names in the store's other languages;
- up to five latest active products for which this is the main category;
- mappings of the same category to other marketplaces — as a strong hint about what it contains.
A broad store category with products of different types gets a broad marketplace category with low confidence. If the dictionary has nothing similar, the suggestion is “No matching category found”.
Start with Google. Its category tree is the same for all countries, and a finished Google mapping then becomes a hint for Allegro, eBay, Kaufland and others — they're matched more accurately.
How to review and apply AI suggestions
The result of matching is a suggestion, not a link. Until you apply it, the mapping in the table doesn't change and the export uses the old data. The category's modification date is still updated, so an export of “only new data” will re-export these categories.
Suggestions are kept until you apply or reject them — you can review them anytime with the “AI suggestions” button. If you started matching from this window, the dialog opens by itself when matching finishes.
1
85% threshold hint2
Ticked confident suggestion3
Green confidence pill4
Yellow 72% pill5
AI explanation under the marketplace category6
Apply ticked- Suggestions from 85% are ticked for applying right away.
- You can clear the checkbox on any suggestion or set it yourself.
- Green pill — a confident suggestion.
- Yellow — AI is unsure: the marketplace category is broader or narrower than yours.
- Under the marketplace category name — a short explanation of why it was chosen.
- Applying writes only to empty rows; manual choices are not changed.
| What's in the column | What it means | Ticked when opened |
|---|---|---|
| Green badge, 85% and up | the same type of products; you can take the suggestion | yes |
| Yellow badge, below 85% | the marketplace category is broader or narrower, or AI is unsure — read the explanation under the name | no |
| “—” and “No matching category found” | none of the candidates fit | no |
Check the yellow rows, tick the ones that fit and click “Apply ticked”. Elbuz writes the marketplace category only where there's no link yet and shows a summary: “Applied: 9. Skipped (the category is no longer in the marketplace dictionary): 0”.
Skipped are suggestions whose category is no longer in the marketplace dictionary: the dictionary was updated after matching. Categories you mapped by hand don't appear in the suggestion list at all. For Prom, the category's ad bids from the dictionary are saved along with it.
Applied and skipped suggestions disappear from the list. If you mapped a category manually in the table, the suggestion is only hidden: remove the link and it returns to the list, and a new run skips this category until you reject the suggestion.
How to reject a suggestion
Tick the rows and click “Reject ticked” — the suggestions are deleted (“Suggestions rejected: 3”), and the categories stay without a link. The next “✦ AI match” run matches a category for them again. A repeat run skips suggestions that weren't rejected, so to have AI reconsider a row, reject it first.
You can't apply a “No matching category found” suggestion: choose a category for such a row by hand with the arrow in the marketplace column (manual matching) or reject the suggestion.
What happens while matching runs
Matching runs as a job in the queue. You can close the window and come back later: to the right of the buttons, a status line shows what's happening on the selected marketplace.
| Message | What's happening |
|---|---|
| “AI matching: preparing the marketplace category list, N entries left” | Elbuz prepares the marketplace dictionary for search by meaning; after a dictionary update, this step repeats only for new and renamed categories |
| “AI matching: 40 of 190” | how many categories have already got a suggestion; updates every few seconds |
| “AI matching finished. Suggestions to review: N” | everything is matched; the suggestions are waiting in “AI suggestions” |
| “AI matching stopped. Run it again — it continues where it left off.” | the AI service didn't respond; matching doesn't retry by itself. If AI access is gone (no key, inactive account), the job is removed from the queue without this message |
After a stop, click “✦ AI match” again. Categories that already have a suggestion aren't matched again — matching continues with the remaining ones.
Scenarios
You're preparing a Google Shopping feed
What to do. Google tab → “✦ AI match” → confirm. When “AI suggestions” opens, review the yellow rows and click “Apply ticked”.
The result. Your categories get a Google category, and the Google Merchant feed template inserts it into every product. Google mapping also adds points to catalog AI readiness.
You're entering Allegro, and Rozetka is already mapped
What to do. Allegro tab → “Load marketplace dictionary” if it isn't loaded yet → “✦ AI match”.
The result. AI sees that the category is already mapped to Rozetka (and to Google, if any) and uses this as a hint. The link isn't copied directly: marketplace trees differ, so the Allegro category is still matched on its own.
You need to match only one branch of the catalog
What to do. In the category table, filter by path, for example “Home appliances”, → “✦ AI match”.
The result. The confirmation shows the number of only the displayed unmapped categories (up to 250); the rest of the catalog stays as it was. If the filter leaves no rows, matching takes the whole catalog — check the number in the confirmation.
What to do when AI matching doesn't start
| What you see | Why | What to do |
|---|---|---|
| “The marketplace dictionary is not loaded yet — click “Load marketplace dictionary”.” | for this marketplace, the dictionary is downloaded from the marketplace itself | click “Load marketplace dictionary” on this tab; the marketplace must be connected |
| “All categories are already mapped to this marketplace.” | there are no unmapped categories among the active ones (or among those shown by the filter) | clear the table filter; remove a wrong link first (how) |
| You confirmed matching, but no new suggestions appeared | all unmapped categories already have suggestions, or these are the same 3,000 categories from the first batch | apply or reject the existing suggestions and run matching again (batches) |
| “No AI access: AI service API key is not configured.” | AI access isn't set up in the account | set up the AI service key and run matching again |
| “The account is inactive — AI features are off until the balance is topped up.” | the account is inactive | top up the balance |
| A category you expected isn't in “AI suggestions” | it's disabled or already mapped, matching is still running, AI returned no answer for it, it was outside the table filter, or beyond the 3,000-category limit | enable the category, wait for matching to finish, or run it again — for this category or as a batch without a filter |
FAQ
Will AI matching overwrite categories I mapped manually?
No. Matching takes only categories without a link, and applying writes only to empty rows: if you mapped a category manually while matching was running, applying skips it (applying).
Does AI match attributes and values?
No, only categories. Attributes and values for Rozetka and Epicentr are matched by hand or with auto-matching by name (auto-matching).
How is “✦ AI match” different from “Run automatic linking”?
Auto-matching sets a link right away, but only on an exact name match. AI matching finds a category by meaning and gives a suggestion that you apply yourself.
Can I see a suggestion in the category table?
No, there's no separate column: suggestions are visible only in the “AI suggestions” dialog. The marketplace category gets into the table after you apply it.
Can I copy mappings from one marketplace to another?
No: marketplaces have different category trees. AI uses mappings to other marketplaces as a hint (scenario).
Where to find it and who has access
- Menu
- Directories → Categories → the “Marketplace mapping” tile → marketplace tab
- Other entry points
- “AI readiness” → the “Map categories to Google” action
- Access right
- the “Marketplace category and attribute matching” window
- Prepare
- AI access; for eBay, Allegro, OLX and others — a loaded marketplace dictionary
Related topics
- If you need the big picture of mapping and manual attribute setup — How to match categories and attributes with marketplaces.
- If the “Google category” tile takes away AI readiness points — How to check whether your catalog is ready for ChatGPT and other AI agents.
- If your categories are mapped and it's time to send products to Rozetka — How to export your catalog to Rozetka.
- If your Google categories are ready and you need to build a feed — How to build a Google Merchant feed for Google Shopping.
