↑↓ selectEnter openEsc close

    1. Elbuz

    How to create and clean up product attributes with AI

    18 min readUpdated

    Elbuz uses AI to build a category's attribute set with types, find duplicate attributes and values, and pick attributes for filters and short descriptions.

    When you need this:

    • A new category has no attributes — AI suggests 30+, you untick extras and save.
    • “Diagonal” and “Display diagonal” exist separately — a confirmed group merges values into one attribute.
    • The filter shows four “black”s — the value check unifies spelling and adds a replacement rule (cleaning up values).
    • Forty attributes, filters needed — AI flags the top ten as filters.

    Where to start: select a category in the “Attributes” tree, then “AI tools” → “Create attributes for the category”.

    What AI can do in the attribute directory

    Some tools build a draft for you to review, others change the catalog right away.

    “AI tools” menu itemWhat it doesWhen it changes the catalog
    Create attributes for the categorysuggests a set of attributes for the selected categoryafter you click “Save”
    Create attributes for all categoriesthe same for the whole storeright away, with no operator review
    Check valuesbrings different spellings of one value to a single spellingafter you click “Apply”
    Find duplicate attributes using AImerges attributes that mean the same thingafter you confirm the groups
    Pick attributes for filterspicks the TOP 10 attributes for storefront filtersright away, with no confirmation
    Pick attributes for the short descriptionpicks the TOP 5 attributes for the product's short descriptionright away, with no confirmation

    Three items work without a draft: once you click, the catalog has already changed. These are “Create attributes for all categories”, “Pick attributes for filters” and “Pick attributes for the short description”. The rest build a list that you review and confirm.

    How to create attributes for a category

    The tool takes the category name with its full path (“Household appliances / Washing machines”) and the list of attributes the category already has, and asks AI to add what's missing.

    Rules AI follows when building the list

    The rules are built into the software; you don't set them. They're worth knowing because they explain why the list looks the way it does:

    • A compound attribute is split in two. Not “Quick wash function”, but “Quick wash” (yes/no) plus “Quick wash time, min” (a number). That's how filters work: a checkbox filters by presence, a slider by value.
    • The unit of measure goes at the end of the name after a comma: “Power, W”, “Weight, kg”.
    • Technologies get their own attributes: “Drum technology”, “Motor technology”.
    • Internal fields are left out: SKU, name, price, availability.
    • No words like “function”, “availability”, “support” in names — the attribute type already says that.
    • At least 30 new attributes — the list is deliberately excessive, on the assumption that you'll untick what you don't need.

    Step by step

    1. Select a category and click “Create attributes”. Expect tens of seconds: it's one AI request for the whole category.
    2. Review the list. New suggestions are already confirmed (ticked), while rows marked “Already in category” are shown for the full picture and aren't created again. An attribute that's in the directory but not linked to the category counts as new: after “Save”, Elbuz links the existing attribute with that name instead of creating a second one.
    3. Untick what you don't need, fix names, and mark which attributes are required. You can add your own attribute with the “Add manually” button.
    4. Check the “Type” column: in each row it's a drop-down from the attribute type directory (types). For a filter with a list of values (brand, color, material), choose “List”; for numbers, a numeric type. The filter slider isn't turned on here but by the filter type in the category (how). The “Required” checkbox is also set right in the table.
    5. Click “Save” — confirmed attributes go into the directory and are linked to the category. If an attribute with that name already existed in the directory, that one is linked. The exception is when “Uniqueness includes the block” is turned on in settings: then only an attribute with that name in the same attribute block where Elbuz puts new attributes counts as existing; an attribute with the same name in another block isn't picked up, and a new one is created.
    The «Create attributes via AI for the selected category» window
    1Create attributes button
    2Category Information block
    1. Create attributes — starts AI generation for the selected category (not pressed while shooting).
    2. Category Information — shows the name, path and number of products of the selected category.

    What green rows and the “Confidence” percentage mean

    A green attribute name means the attribute is already linked to this category: the name has the caption “Already in category” under it, and confidence is 100%. New AI suggestions have names in the regular color.

    • “Save” doesn't create a new attribute from a green row, even if it's ticked in the “Status” column.
    • If you change the “Required” checkbox in a green row, “Save” writes it to the category.
    • The type of an attribute that's already in the category can only be changed in the attribute directory.

    For new suggestions, the “Confidence” column is AI's own score from 0 to 100: how much, in its view, the category needs the attribute. If AI gave no score, it shows 85%. Attributes already in the category and ones added manually have 100%. The score is rough and the list is deliberately excessive, so pick by name.

    To run it again, use “Clear all”. After saving, the category gets the “attributes created” mark, and a second attempt replies: “Attributes have already been created for this category. To start over, click “Clear all”.” The “Clear all” button removes the results and clears the mark — after that, you can run generation again.

    Three outcomes leave the catalog unchanged: the “attributes created” mark, an AI failure, and saving with no confirmed new row. The category gets the mark only in the bottom-right branch.
    yesnonoyesnoyes“Create attributes”Category has the“attributes created” mark?“Attributes have already beencreated for this category”first “Clear all”AI request: name, pathand the category's existing attributesAI returned a listthat could be parsed?“AI did not return a response” or“Failed to parse the AI response”List in the window: new rowsconfirmed,“Already in category” rows notYou untick extras, fix names andtype,click “Save”Any confirmednew attributes?Nothing saved,no mark setAttributes in the directory and linkedto the categorycategory gets “attributes created”
    • Start or end
    • Action or result
    • Decision: the path depends on the answer

    “Clear all” deletes only this window's draft list. Attributes you've already saved stay in the directory and linked to the category, and values in products don't change either. On the next run, AI receives them as existing, and they appear in the list marked “Already in category”.

    The only way to remove an unwanted attribute that AI has already saved is to delete it in the directory — along with all its values in products. If the values are already filled in and you need them, merge the attribute into the right one with the “Link and delete” button (merging duplicates).

    Check the type before saving. AI suggests one of three types — text, number or checkbox — and that type goes into the directory. You can change it in the “Type” column to any type from the type directory: for example, choose “Decimal number” where you expect fractional values.

    The type doesn't decide whether the filter has a slider: that's set by the filter type (how), and the storefront takes the slider range from the products' numeric values. If the type isn't set or isn't recognized, the attribute gets “Text”.

    Names in other languages

    A new attribute gets a name only in the default language. In the catalog's other languages, its name stays empty. If your store is multilingual, run bulk translation after creating attributes — otherwise the attribute will have no label on the other-language storefront.

    How to create attributes for all categories at once

    The same “AI tools” menu has a “Create attributes for all categories” item. It does the same thing, but for the whole store and without your review: for each category, AI builds a list and saves it to the directory right away.

    It takes only active leaf categories with products where attributes haven't been created with AI yet. Categories with the “attributes created” mark are skipped — to run such a category again, open its window and click “Clear all”.

    Before the run, Elbuz shows a warning: each category means a separate AI request, which takes a long time and spends credits. After you confirm, the log window opens and Elbuz checks AI access:

    • in the demo version, AI works with a daily limit — up to €3 a day for AI, parsing and media generation combined; once the limit is used up, the run isn't queued — “The daily demo limit is used up: up to €3 a day for AI, parsing and media generation. You can continue tomorrow or on a paid plan.”;
    • without an AI service key — “No AI access: AI service API key is not configured.”

    If access is available, the work runs as a background process:

    • the log shows progress by category; a failure in one category becomes a warning and doesn't stop the rest;
    • the cancel button stops the run at the next category;
    • a category where AI suggested only already-linked attributes gets the line “Attributes created: 0”, counts as “Successful”, and is skipped by the next run;
    • at the end, there's a “Processed / Successful / Errors” summary; the operation is visible in the operations center, so you can close the window.
    The run loops until it runs out of categories or you click cancel. A failure in one category goes to “Errors” and back into the loop instead of stopping the run.
    no: demo or no keyyesyesnoyesnoyesnoyesno, or AI failedno categories leftConfirm “Run”AI access available?Red message with the reasonthe run isn't queuedNext active leaf category withproducts“attributes created”mark?Cancelclicked?Run stoppedAI request, all new attributessaved without reviewAt least oneattribute saved?Log line, “Successful”category gets “attributes created”AI suggested onlyalready-linked attributes?“Attributes created: 0”, “Successful”category gets “attributes created”Warning in the log, “Errors”no mark setSummary “Processed / Successful /Errors”
    • Start or end
    • Action or result
    • Decision: the path depends on the answer

    This tool is for a catalog that's just being filled. In a catalog where attributes are already maintained by hand, it means hundreds of new attributes nobody has reviewed. First run one category the regular way and look at the result.

    How to find duplicate attributes with AI

    This is for when your directory has “Diagonal”, “Screen diagonal” and “Display diagonal” — three attributes instead of one. It usually happens when attributes come from several sources: price lists, parsers, exports.

    The tool takes the attributes of the selected category and shows groups: a main name plus variants that mean the same thing. Attributes that already have a synonym aren't analyzed — they count as sorted out.

    When you confirm a group, Elbuz does three things at once:

    1. Moves product values from the duplicates to the main attribute — across the whole catalog, not just the selected category;
    2. renames the duplicates to the main name and removes those that became identical;
    3. creates a synonym for each old name. From then on, a price list or that brings “Display diagonal” automatically puts the value into “Diagonal” — you won't have to sort out the same thing twice.

    There's no undo. The values have already moved, and the old attribute names no longer exist. If AI mistakenly merged different attributes — for example, “Width” and “Package width” — you'll have to untangle it by hand. So before confirming, read the groups, not just their count, and untick anything you're not sure about.

    The manual way without AI is the “Attribute matching” window and the “Synonym” button (How to merge duplicate attributes into one).

    How to bring different spellings of a value to one

    The “Check values” item opens the value check: AI brings “Exists”, “+” and “Yes” to one spelling by your rules, or looks for values that mean the same thing. You see each change as “Before → After” and apply it yourself, and Elbuz creates replacement rules for new incoming data. Details: How to clean up attribute values.

    How to pick attributes for filters and the short description

    The last two menu items answer the question “which of the forty attributes to show the shopper”.

    • “Pick attributes for filters” — AI picks the TOP 10 category attributes that make sense for filtering on the storefront and sets the filter flag on them right away.
    • “Pick attributes for the short description” — picks the TOP 5 most important ones, makes up a short name of 2–3 words for each, and orders them from most to least important. This is what shows under the product name in the list.

    Each tool sets its own flags and doesn't touch the other's. If a category is selected in the tree, the selection always runs for it, even repeatedly: old filter (or short description) flags are cleared, and the ones AI picked are set.

    If no category is selected, the tool takes one category — the first one that has products and no configured filters (or short description attributes). To go through the catalog, click the button again or go category by category.

    When this shows up on the storefront and in products

    Both tools change only the category's attribute settings in Elbuz, and the short description selection also adds short names to attributes. After that, the paths differ.

    • Filter flags go to a storefront connected via tunnel with the next export to the storefront. Only attributes with “Export to site (file)” turned on get there; an attribute without this flag won't become a filter on the storefront.
    • The short description text in products doesn't change from the selection. It's built from the selected attributes as “Name: value” separated by “ / ”, in the order AI set, and with the short name where there is one.
    • The text is rebuilt when you save the product card or click “AI tools” → “Create product short descriptions” — for the category selected in the tree, or across the whole catalog without one.
    • A product that has none of the selected attributes filled in gets an empty short description when it's rebuilt — separately in each language. In categories without short description attributes, text you entered manually doesn't change.
    • AI writes a short name only for attributes that don't have one yet. A repeat selection won't overwrite a short name you've already set.

    How much an AI run costs

    You're charged for AI requests, not for created attributes or processed products. A request's price depends on the amount of text: what Elbuz sends to AI and AI's response are counted separately. How many requests each tool makes:

    ToolAI requests
    Create attributes for the categoryone per click on “Create attributes”
    Create attributes for all categoriesone per category the run takes
    Check values“Find duplicates” — one per attribute that has at least two different values; “Apply value standard” — one per every 150 not-yet-checked values of an attribute
    Find duplicate attributes using AIone per category
    Pick attributes for filters, for the short descriptionone per category

    A request is paid as soon as AI responds. If the response then couldn't be parsed (“Failed to parse the AI response”) and no attributes appeared, the request is still charged.

    Costs add up and are deducted from your balance as one amount. You can see them in the “Balance” window, “Transaction history” tab, operation “Text generator (AI)” — together with other AI text generation requests; attributes aren't listed separately there.

    If your balance runs out

    • The balance isn't checked before a request. The “all categories” run doesn't stop when money runs out midway: it keeps going, and charges can take your balance negative.
    • When the balance is zero or negative, the service disables the account. In a disabled account, no AI request is sent and nothing is charged: “Create attributes” replies “AI did not return a response”.
    • If the “all categories” run fails this way in a disabled account, each category gets a warning in the log and ends up under “Errors” in the summary. These categories don't get the “attributes created” mark, so the next run after a top-up picks them up again.

    Scenarios

    A new category has no attributes, and the storefront needs filters

    Select the category in the tree, then in the “AI tools” menu choose “Create attributes for the category” and click “Create attributes”. In the list, set a numeric type in the “Type” column where there will be numbers, untick what you don't need and click “Save”. Then, with the same category selected, choose “Pick attributes for filters”.

    • Confirmed attributes appear in the directory and are linked to the category. On save, only a required attribute that already existed in the directory gets the filter flag; a new attribute comes without it. The filter selection sets the flag on up to ten attributes.
    • No values appear in products: attributes are created empty and get filled by price lists, parsers and imports.
    • For the selected category, the filter selection first clears the filter flag from all its attributes, including those that got it for being required, and then sets it on the attributes AI picked.

    Suppliers name the same attribute differently

    Select a category where this shows, then in the “AI tools” menu choose “Find duplicate attributes using AI”. Read through each group and confirm only the ones you're sure about.

    • Product values from the whole catalog move to the main attribute, the duplicates get its name, and the old names become synonyms. The next price list with “Display diagonal” will put the value into “Diagonal”.
    • Groups you unticked stay as they were.
    • An attribute that already has a synonym won't end up in groups. A merge can't be undone.

    One value in the filter has split into several spellings

    Select a category or the attribute itself, for example “Color”, then in the “AI tools” menu choose “Check values”. How to check values and apply the changes: One color written five ways.

    You need a short summary of key attributes under the product name

    Select the category, then in the “AI tools” menu choose “Pick attributes for the short description”, then “Create product short descriptions”.

    • AI picks up to five attributes, orders them from most important, and gives a short name to those that don't have one yet. The product text is built as “Name: value” separated by “ / ”.
    • Products with none of the selected attributes filled in get an empty short description — separately in each language that has no values.
    • The second button rebuilds descriptions only for products in the selected category.

    What to do when an AI tool returns no result

    Common problems
    What you seeWhyWhat to do
    “The daily demo limit is used up…” or “No AI access: AI service API key is not configured.” The demo version's daily €3 limit for AI, parsing and media is used up; without an AI service key, no request is sent In the demo, continue tomorrow or switch to a paid plan; without a key, set up the AI service (access check)
    “Attributes have already been created for this category” the category has already been through generation click “Clear all” — the mark is cleared and you can run generation again
    The attribute got the “Text” type, though it should be numeric the type wasn't set or AI didn't recognize it fix the type in the “Type” column before saving, or later in the attribute directory
    The filter has no range slider the slider is set by the filter type, not the attribute type; the attribute has a different filter type set the filter type to slider or range slider (how); values in products must be numbers
    The attribute name is empty in another language the name is written only in the default language run bulk translation of attributes
    AI suggests what already exists existing attributes are shown on purpose, marked “Already in category” nothing: these rows aren't confirmed and won't go into the catalog
    “Pick attributes for filters” changed nothing, though it showed “Operation completed successfully” no category is selected in the tree, and there are no categories with products left that have no filters yet — the tool has nothing to take select the category you need in the tree and run the selection again: old filter flags are cleared and AI sets new ones
    “Pick attributes for filters” processed one category with no category selected, it takes one — the first unconfigured one select a category in the tree or click again
    The button ran, but there's no result AI returned a response that couldn't be parsed, or returned nothing run it again; check the “Operation Log” in the tool's window
    An attribute didn't show up in a duplicate group it already has a synonym, so it isn't analyzed check the attribute synonym directory

    FAQ

    I click “Create attributes” and it says “Attributes have already been created for this category”. How do I create attributes for the category again?

    Click “Clear all” in the generation window: the button removes the previous results and clears the “attributes created” mark from the category. Saved attributes stay in the directory (creating).

    AI suggested the “Text” type, but I need a list of values for a filter. How do I change the type?

    Before “Save”, choose “List” in the “Type” column of that row. Without a set or recognized type, the attribute gets “Text”. The type of a saved attribute is changed in the attribute directory (types).

    Can I undo an attribute merge or a value replacement?

    No. The values have already moved and the duplicate names are gone. You can only turn off a permanent replacement rule: “Maintenance” → “Configure attribute value replacement rules” (replacement rule).

    Will I have to clean up the same duplicates after the next price list import?

    No. An attribute merge creates synonyms for the old names, and a value replacement is a permanent normalization rule. Both keep working automatically for new incoming data.

    Can AI fill in values for these attributes in all products of the category right away?

    No, these tools work with the directory. Values are filled in by content search for products, as well as by price lists, parsers and imports.

    Does AI translate the names of new attributes into other languages?

    No: a new attribute gets a name only in the default language. Other languages are filled in by translation (How to translate product attributes into other languages).

    Where to find it and who has access

    Menu
    Directories → Base product catalog → Attributes → the “AI tools” button
    From the category card
    the “Search for attributes” button — creates attributes for that category
    Access right
    the “Attributes” window in group permissions
    Prepare
    a category selected in the tree on the left — without it, most tools will process the wrong thing

    How the directory itself works: How to add an attribute to the product card.

    The Attributes window with the expanded «AI tools» menu
    1AI tools button
    2Create attributes for the category
    3Check values
    4Find duplicate attributes using AI
    5Pick attributes for filters
    1. AI tools — the button that opens the menu of AI actions for attributes.
    2. Create attributes for the category — opens the AI attribute generation window for the selected category.
    3. Check values — opens the attribute value check window.
    4. Find duplicate attributes using AI — opens the duplicate search window.
    5. Pick attributes for filters — an automatic analysis item that applies immediately, so use it with care.

    Related topics

    Was this article helpful?

    Missing a feature? Tell us.

    We will add it for free if other users need it too.

    If only you need it, we will agree on it separately.

    Article contents

    Suggest an improvement

    Try Elbuz free