Elbuz takes the name, prices, SKUs, photos, files, and specifications from the supplier's product page and puts them into the source price list: no manual copying.
When you need this:
- Can a field be collected? The list has 33 product page, 21 listing, and 6 specification fields.
- The site and price list differ: the cleanup order shows where the value changed.
- The price came as two numbers: the app converts the price to a number; listing traps are covered separately.
- Specifications aren't collected: usually “Selector #2” is empty in the “Attributes” group.
Where to start: find the field in the list and add its selector on the “Import settings” tab.
What the parser can do on a product page
- 64 rule rows: the full list of what you can collect; an empty selector means “don't collect”, and there are no default values.
- Value cleanup: markup, replacements, the regular expression, and the delimiter are applied in a fixed order.
- The price is turned into a number without regular expressions.
- Photos, videos, and files: placeholders and duplicates are dropped, and a preview on the product page is scaled up to full size.
- Specifications: the parameter table is split into “name — value” pairs.
- Without a selector, the app fills in the category, cleans the SKU, and takes meta tags.
- “Auto-detect selectors” picks selectors for 14 fields.
- What happens to the price list on a repeat run or when a run breaks off is in How the parser updates the price list; how specifications become catalog attributes is in Site specifications in the catalog.
The value's path: what happens to what's found
Between what you see on the site page and what ends up in the price list there's a chain of transformations. Its order is fixed, and confusion about this order causes most of the “strange” results.
- Action or result
- Decision: the path depends on the answer
- What to take. By default, the element's text is taken. The “Get HTML” checkbox keeps the markup: a description needs it to keep its paragraphs and lists.
- Addresses are completed. For link fields and photos, a relative address becomes a full one;
#,javascript:,tel:,callto:, andmailto:are dropped entirely;//site.combecomeshttps://site.com; a?click_track_data=tail is cut off; a Cyrillic domain is converted to punycode. - Automatic cleanup. HTML comments, the tag attributes
class,rel,style,id,name,title,data, emoji, non-breaking spaces, and line breaks are removed from the result; several spaces in a row collapse into one. - “Text to clean up” cuts the specified text out of the result. Separate several values with
|||. - “Find” / “Replace” do three different things depending on what's filled in. Both fields: a normal replacement. Only “Replace”: the text is appended to the end of the value. Only “Find”: the text is cut out.
- “Remove links” removes link tags from the result. What to remove (your own domain, other domains, or all) is set by two checkboxes in the “Additional” accordion at the bottom of the tab. If neither is checked, all links are removed, and the addresses themselves are cut out of the text too.
- “Allowed HTML tags” keeps only the listed tags in the markup, for example
<p><br><ul><li>. - The price is turned into a number; there's a separate section on this below.
- “Regular expression” is applied after everything above: use it to extract the part you need from the already cleaned value.
- “Value delimiter” joins the matches if the selector found several elements. By default it's a comma and a space; for the category it's always
/, for photos|||.
“Allowed HTML tags” does nothing without the “Get HTML” checkbox. Without it, the value gets the element's plain text, which has no tags left, so there's nothing to keep. These two columns only work as a pair.
1
Get HTML2
Allowed HTML tags3
Text to clean up4
Find / Replace- Get HTML keeps the markup of the value instead of plain text.
- Allowed HTML tags lists the tags that survive when HTML is kept.
- Text to clean up is a fragment removed from the collected value.
- Find and Replace work as a pair: the found text is replaced with the text from the next column.
The four selector columns are fallbacks
Each field row has “Selector #1” … “Selector #4”. The app tries them in turn and stops at the first one that returns a value. This helps when sale products are marked up differently from regular ones: put the main variant in #1 and the sale variant in #2.
There's one exception: photos. In the image row, all four columns run, and everything found goes into one list of photos.
A fixed value instead of a search
If you write text in curly braces instead of a selector, it becomes the field's value, and nothing is searched on the page. {New collection} in the “Manufacturer” field sets this text for all products of the source.
Which fields you can collect: the full list
When a universal parser is created, it gets exactly 64 rule rows, grouped by crawl operation. This is the complete list of what can be collected: the grid won't get more rows, and you can't add your own field. If the field you need isn't there but the site shows it in the specifications table, collect it as a specification.
A selector searches only the markup the site returned: it won't find data the page loads after a button is clicked.
| Operation | Rows | What it is |
|---|---|---|
| List of product category links | 1 | only “Operation selector”, no fields |
| List of product links | 22 | “Operation selector” and 21 fields; they work in “without opening the product page” mode |
| Product card | 34 | “Operation selector” and 33 fields |
| Attributes | 7 | “Operation selector” and 6 specification fields |
Exactly one selector in the whole grid is filled in out of the box: h1 for the “Name” field in “Product card”. The other 63 rows are empty.
Product card: 33 fields
This is the main set. Each field is a separate row in the grid, and the labels are exactly as in the interface. One row has no label: photos are listed in the grid under their own code, image.
| Group | Fields |
|---|---|
| Name and description | Name, Model, Short description, Full description, Instruction, Variant family name |
| Prices | Price, RRP, Old price, Currency, Delivery cost |
| Availability | Stock status (from price list), Quantity, Product condition |
| Identifiers | Supplier SKU, Manufacturer SKU, Manufacturer, Barcode, , ISBN Barcode |
| Dimensions and weight | Height, Width, Depth, Weight, Dimension, Unit, Color, Warranty |
| Category | Category from price list, Category link |
| Media | image, Video link, File link |
1
Name with a CSS selector2
EAN barcode with an XPath selector3
Warranty without a selector- The name is read with a plain CSS selector.
- The EAN barcode is read with an XPath selector (the XPath box is ticked).
- Warranty has an empty selector, so this field is not collected.
Listing: 21 fields
If the “Products in this operation” checkbox is selected in the “Operation selector” row of the “List of product links” group, product pages aren't opened at all: names and prices are taken straight from the section page, the .
It has its own, shorter set of fields.
These are the same 20 product page fields, without dimensions, weight, color, warranty, product condition, variant family, instruction, delivery cost, category link, video, and files, plus one field the product page doesn't have: “Product link (URL)”. On a product page the address is already known; in a listing it has to be collected with a selector.
In listing mode, a product without a name or an address isn't saved at all. So the “Product link (URL)” field is as mandatory here as “Name”: without it, the run finishes cleanly, but the price list has no rows.
“Product card” mode has no such check: a product without a name is saved as an empty row.
Attributes: 6 fields
“Attribute name”, “Attribute value”, “Attribute block name”, “Attribute description”, “Attribute image link”, “Attribute block image link”. They have their own section below.
The list of operation types has a fifth one, “Photo”, but no rules are created for it. Photos are collected by the image field in “Product card” and in the listing.
Which fields people actually set up
The grid has many fields, but in practice people set up seven or eight; the rest are there just in case. Here they are, from most to least needed:
- name, category from price list, and price: without them a price list row is useless;
- manufacturer, supplier SKU, and manufacturer SKU: matching with your catalog relies on them;
- stock status and quantity;
- full and short description, RRP, old price: when you also pull content from the supplier's site.
An empty field doesn't mean “not supported”: each one is turned on with a single selector, exactly like the price or the name.
How the parser takes the price: three fields and two traps
A product page has three price fields: “Price”, “RRP”, and “Old price”. Plus “Currency”, a separate row with text from the page.
A price-type value is turned into a number: everything except digits, commas, and periods is dropped. Then the app applies the separators set for the price list (by default, a period for decimals and a comma for thousands) and rounds to two decimal places.
So 1 499,50 грн becomes 1499.50 without any regular expression.
The saved price goes into three price list columns at once: the price itself, the original price, and the cost price. After that, the usual price list operations work with it: markup, rounding, currency conversion.
There are two traps, and both apply only to listing mode; in “Product card” they don't work.
Two prices in one element. When a site shows the old and new price side by side, both get into the value. A listing handles this: if a currency mark (грн, ₴, $, USD, €, EUR, PLN, and others) appears twice, both numbers are taken and the lower one is saved. On a product page, refine the selector instead.
Cents in a separate element. Some sites put the whole units and the cents in different tags. In a listing, put the units selector in “Selector #1” and the cents selector in “Selector #4”: the app combines them, dividing the second by a hundred. A product page doesn't do this; there, “Selector #4” is just one more fallback.
Photos, videos, and files: what the app does with the addresses
Photos are collected by one field; in the rules grid it's labeled with the code image. All four selector columns run, and everything found goes into one list in the order it was found: the first photo becomes the main one in the catalog.
What the app does with each address it finds is the same in both modes:
- Drops placeholders. If the address contains
placeholder,no-photo,empty,no_image,no-image, or/brand/, the image isn't saved at all. - Sends YouTube to video. A
youtube.com/watchlink that ends up inimageis saved as a video, not a photo. - Skips embedded
data:imagepictures; in “Product card” it also skips.swgfiles. - Drops duplicates: the same address is saved only once per product.
- Escapes separators in the address. Commas, semicolons, and vertical bars become
%2C,%3B,%7C: the photo list is stored as a single line, and an address like…/image/width=1063,height=1063/…would otherwise be confused with the list separator.
Two things exist only in “Product card”:
- The largest size from
srcset. If the selector text contains the wordsrcset(for example,img.photo::attr(srcset)), the address with the largest width is chosen from the set of variants;2xcounts as larger than1x. - Scaling a preview up to full size. The address loses the parameters
w,h,width,height,resize,size,thumb,thumbnail; the file name loses size suffixes like_100x100,-150x150and-thumb,-thumbnail,_small,_mini,_tiny,-previewbefore the extension. If the result stops looking like an address, the original returns: a working preview beats a broken “large” file.
In a listing, on the other hand, the app can get the address from a CSS background: background-image: url(...).
The srcset trick doesn't work in a listing. There, the selector img.photo::attr(srcset) returns the first variant in the set, usually the smallest, and a listing doesn't scale previews up to full size either. If you collect without opening product pages, get the full size from another attribute: ::attr(data-src), ::attr(data-large), or wherever the site keeps the large photo.
There's no limit on the number of photos from one product page. The “Max. number of results” column is read only from the “Operation selector” row and limits link lists, not fields.
The doesn't download the files themselves; it saves addresses. An image becomes your copy in the cloud during the base catalog update; what happens then to the order, the main photo, and whose copy it is, see Product photos. No watermark is applied along this path at all: watermarks exist only in exports.
Videos and files
“Video link” and “File link” exist only in “Product card”. The video value is split by commas, and duplicates are dropped.
Files have a quirk worth knowing: the app takes the address from the href attribute and saves the link text itself as the file name. So the selector a.instruction gives both the address and the caption “User manual”, while a[href*=pdf]::attr(href) gives only the address, with an empty name.
How to collect product specifications
The product parameter table is collected by a separate operation, “Attributes”. It goes through the table row by row and saves “name — value” pairs.
It's defined by four selectors:
- “Operation selector”, Selector #1: the specifications block, for example
table.specs. Empty means search the whole page. - “Operation selector”, Selector #2: the selector of one row of specifications, usually
tr. - “Attribute name” and “Attribute value”: selectors of the cells inside the row, for example
td.nameandtd.value.
Two more rows, “Attribute block name” and “Attribute description”, are needed for sites where specifications are split into sections.
An empty “Selector #2” stops the operation completely: without a row selector it exits without collecting anything, no matter how much is filled in below. This is the most common reason for “specifications aren't collected even though everything is set up”.
Two exceptions. <dl> markup: if Selector #1 is dl or starts with dl. and Selector #2 is empty, the app converts the list to a table and sets the row and cell selectors. And a source set to instead of : there, an empty “Selector #2” doesn't stop it (JSON is a separate topic; this article covers HTML).
1
Selector #2 of the operation row2
Attribute name3
Attribute value- The operation selector finds the block of characteristics, and Selector #2 picks one characteristic row inside it.
- Attribute name has its own selector, applied inside each row.
- Attribute value has its own selector, applied inside each row.
What the app parses on its own
- “Name: value” in one cell. If name and value share one line (“Color: yellow”) and both selectors take the same text, the app splits it at the first colon; a line without a colon is skipped. A value in its own element inside the name line (
<li>Color: <b>yellow</b></li>) is cut off the name automatically. Another separator (“Color — yellow”, “Color | yellow”) goes in the “Value delimiter” column of the “Attribute value” row. - A value without a selector. If no value selector is set at all, the text of the row's last nested element is taken. On simple tables this is enough.
- A list inside a value. When the value cell has several
<li>items and the text has no delimiters, the items are joined with the “Value delimiter”; by default it's a slash with spaces. - Question marks and colons are removed from the attribute name.
What the app throws out
These pairs aren't saved: pairs with an empty name or empty value; pairs where the name equals the value (this almost always means both selectors point to the same element); a repeat of the previous row; and full duplicates by the “block + name + value” triple.
A real setup: block selector #pdpPropertiesContent, row .pdp-main-properties-row, name .pdp-main-properties-name, value .pdp-main-properties. One TV page gave 42 specifications, such as «Діагональ / 65"» (Diagonal / 65"), «Частота розгортки / 165 Гц» (Refresh rate / 165 Hz), «Рік випуску / 2026» (Year of release / 2026), in site order, plus 18 photos and one file.
How the collected names become attributes in your catalog, and how to keep the directory from bloating, is in Site specifications in the catalog.
What the app fills in on its own, without a selector
A few things are filled in without your involvement, and you don't need to spend rule rows on them.
Category. If the “Category from price list” field is empty, the app reads the breadcrumbs from the page markup meant for search engines and joins them with / .
Then service links are cut from the start of the line («Головна / », «Каталог / », «Магазин / », «Всі товари / », that is, Home, Catalog, Shop, All products, and about a dozen more variants), and the product name itself is cut out of the line: the last breadcrumb is usually the product page's name.
The SKU is cleaned automatically. From both SKU fields, “Supplier SKU” and “Manufacturer SKU”, the app cuts the words «Артикул товара», «Артикул», «Арт.», «Код товара», «Код товару», «Код», “SKU”, and all colons. This happens in both modes, and the selector .pdp-menu-code on a page that reads «Код: 7265989» puts 7265989 into the price list.
Availability, SKU, manufacturer, and unit in a listing. When “Auto-detect selectors” is on and a selector didn't work, the app searches for the value in the page text:
- availability: by a list of words («В наявності», «Немає на складі», «Під замовлення», «Очікується», that is, in stock, out of stock, made to order, expected, plus about twenty more Ukrainian, Russian, and English variants);
- SKU: by the labels «Код товара:», «Код товару:», «Код:», “SKU:”, «Артикул товара:», «Арт.:»;
- manufacturer: by «Бренд:» and «Производитель:» (Brand, Manufacturer);
- unit: by the words «ящик» and «/ящ» (box).
If availability still isn't found, a listing gets +.
The same auto-detection, when on, adds «Бренд:» and «Производитель:» to the cleanup of the “Manufacturer” field; this one works in both modes.
SKU auto-cleanup doesn't ask permission. The SKU Код-114 arrives as -114, SKU12 as 12, and A:B as AB. If the supplier's numbers contain these words or a colon, don't collect them into a SKU field; use, for example, “Model”, where nothing is cut.
Currency and stock status are finished after the crawl. When the last task is closed, the usual price list operations are applied to the collected products. They turn the text «В наличии» from the “Stock status (from price list)” field into the “In stock” status using the stock recognition rules, fill in the currency, and calculate the markup.
Meta tags and a text copy of the page
Two checkboxes on the “Additional” tab, in the “What data to collect” group. Both are off by default.
“Save meta tags”: the app reads the page's <title> and meta tags and puts them into the product's meta fields: title, description, and keywords. In listing mode, the product name is prepended to each of the three values; in “Product card” they're saved as is.
“Save a text copy of site pages” saves the page text, stripped of markup, to the “Website page text” field of the price list row. This works only in “Product card” mode: in a listing the page text isn't saved, however long the checkbox stays on.
Which selectors “Auto-detect selectors” picks
The “Auto-detect selectors” checkbox on the “Additional” tab turns on auto-detection: where a selector column is empty, the app tries its own set of typical selectors and writes what it finds straight into the rules grid, up to four columns per field.
Not every field has such a set. Here's the full list of what auto-detection can look for:
| Field | On the product page | In the listing |
|---|---|---|
image | yes | yes |
| Price | yes | yes |
| Old price | yes | yes |
| Stock status (from price list) | yes | yes |
| Short description | yes | yes |
| Supplier SKU | yes | yes |
| Manufacturer SKU | yes | yes |
| Manufacturer | yes | yes |
| Category from price list | yes | yes |
| Category link | yes | no such field |
| Video link | yes | no such field |
| File link | yes | no such field |
| Name | — | yes |
| Product link () | no such field | yes |
Auto-detection never looks for the rest: “Full description”, “Warranty”, “Weight”, “Color”, dimensions, barcodes, “Model”, “Currency”, “RRP”, “Quantity”, “Product condition”, “Variant family name”, “Instruction”, “Delivery cost”, “Dimension”. Enter these selectors by hand.
“Unit” is a special case: auto-detection picks no selector for it, but in a listing it can fill in the value by the words «ящик» and «/ящ».
The default name selector is h1, and on a page that isn't a product it also finds something: the section heading. Auto-detection then fills in a description, the first block with a class like description.
This happened on a real setup: the crawl caught city pages, and the price list got 223 “products” named like «Посуд в Ізмаїлі», «Посуд в Білій Церкві», «Посуд в Чернігові» (Cookware in Izmail, Bila Tserkva, Chernihiv) with one description, «з 8:00 до 21:00 (пн-нд) Безкоштовно по Україні…» (8:00 to 21:00, Mon–Sun, free across Ukraine…): a delivery banner.
In 38 of the 223 rows the price was zero, and they collected 727 photos: section banner images. The run still finished without a single error.
After the first run, open the price list's product grid and look at the names and prices. Zero prices and names that look like section headings mean the crawl picked up the wrong pages. The full list of checks: what to check after the first run.
Scenarios
Two typical field setup tasks. Scenarios about repeat runs are in How the parser updates the price list, and about specifications in the catalog, in Site specifications in the catalog.
Collect names and prices from sections without opening each product page
What to do. In the “List of product links” group, select the “Products in this operation” checkbox in the “Operation selector” row. In the same group, fill in “Name”, “Product link (URL)”, and “Price”.
What you get. No product page tasks are created: price list rows are written straight from the section pages, from the 21 fields available in the listing.
What you won't get. A row without a name or an address isn't saved. There are no specifications: the app parses the “Attributes” group only on an open product page. There are no videos or files in a listing either: it doesn't have these fields.
Take only photos and specifications from the supplier's site and leave the price alone
Task. The supplier sends a file with your price and links to product pages, while the site shows the retail price. You need only photos and specifications from the site.
What to do. In the file price list card, “Source” section, click “Create a site parser”. The button is active once the price list has a column mapped to “Product link (URL)” and has been imported.
In the new parser's “Site data import settings” window, fill in the selectors for photos (the “Product card” group) and specifications (the “Attributes” group), and leave the “Price”, “RRP”, and “Old price” rows without a selector.
What you get. The button creates a separate parser price list for this supplier, linked to yours: it opens product pages from your price list links. Only price, quantity, and stock status return to your main price list, and only when filled in. The price returns only if it's above zero, so without a price selector the file price stays.
Photos and specifications don't move to the main price list: they stay in the parser price list and reach product cards when the base catalog update processes that parser price list. Transfer of photos and attributes is on by default for the parser price list.
What you won't get. If you do set a price selector, a non-zero site price overwrites the file price in the main price list. Same for availability: quantity and stock status from the site replace the file's values; empty ones don't. Transferring photos and specifications from the parser to the catalog is paid (details).
What to do when the price list doesn't match the site
| What you see | Why | What to do |
|---|---|---|
| The price list has names like «Посуд в Ізмаїлі» (Cookware in Izmail), and the price is zero | The crawl picked up pages that aren't products, and the default h1 selector took their heading as the name |
Refine the product link selector or add the section addresses to “Links” (about auto-detection) |
The SKU in the price list is shorter than on the site: -114 instead of Код-114 |
SKU auto-cleanup cuts the words «Артикул», «Арт.», «Код», “SKU”, and all colons from both SKU fields, regardless of mode | Collect such a number into a field without cleanup, for example “Model”, instead of a SKU field (what the app does on its own) |
| Specifications aren't collected, although the name and value selectors are filled in | “Selector #2” is empty in the “Operation selector” row of the “Attributes” group: without a row selector, the operation exits without doing anything | Enter the selector of one specification row in “Selector #2”, usually tr (about specifications) |
| Fewer specifications were collected than the site has | The app drops pairs where the name equals the value, and full duplicates. An identical name and value mean both selectors point to the same element | Refine the “Attribute name” and “Attribute value” cell selectors (what the app throws out) |
| The price has the number from the old price, or two numbers in a row | The selector captured a block that shows both prices. Picking the lower of the two works only in listing mode | Narrow the selector to the element with the current price, or use “Regular expression” (about the price) |
| A placeholder picture became the main photo | The app drops placeholders by the names placeholder, no-photo, empty, no_image, no-image and by /brand/; a placeholder with a different name isn't on this list |
Refine the image selector; as a fallback, enter the placeholder's full address in “Text to clean up” in the image row: the value becomes empty and the image isn't saved (about photos) |
| Photos came in small, like gallery thumbnails | The selector took the preview attribute. Automatic scaling to full size removes only typical markers like _100x100 or ?width= |
In “Product card”, take the address from srcset: img.photo::attr(srcset), and the app picks the widest variant. In a listing srcset won't help: you need another attribute with the full size (about photos) |
| In “without opening the product page” mode the run finished, but the price list is empty | A product without a name or an address isn't saved | Fill in both fields in the “List of product links” group: “Name” and “Product link (URL)” (about the listing) |
| “Category from price list” contains the product name or «Головна» (Home) | The category was collected from breadcrumbs, and your spelling of the service link isn't on the list the app cuts | Add the extra text to the “Text to clean up” column of the “Category from price list” row (what the app does on its own) |
FAQ
Does the parser download the photos themselves?
No. The parser saves only the image addresses on the supplier's site. The file becomes your copy in the cloud during the base catalog update. Until there's a copy, the supplier can remove the picture, and it will disappear for you too.
How many photos will the parser take from one product page?
As many as the selector finds: there's no limit on the number of photos. The “Max. number of results” column limits link lists, not fields. Duplicates of the same address are dropped, and the first photo found becomes the main one.
Why is the SKU in the price list shorter than on the site?
The SKU is cleaned automatically: the words «Артикул», «Арт.», «Код», “SKU”, and others, plus all colons, are cut from both SKU fields. Код-114 becomes -114. Collect such numbers into a field without cleanup, for example “Model” (details).
Can I add my own field to the list?
No. The 64 rule rows are created along with the source, and the grid has no “add row” button. If the value is on the site in the parameter table, collect it as a specification: in the catalog it becomes an attribute.
Does the parser collect reviews?
No. The “Load reviews” checkbox is in the settings, but the universal parser doesn't read it.
Does the parser create product variants from specifications?
No. The parser doesn't mark specifications as “this is a size or color variant”: variants are built from the “Variant family name” field and from separate pages of option products.
Where to find it and who has access
Fields are set up in the rules grid: the “Import settings” tab of the “Site data import settings” window. The tab exists only for the universal parser: in a ready-made module (Rozetka, PROM, Hotline, eBay, Amazon, and others) we wrote the parsing rules, and there's nothing to change.
None of the parser windows are in the menu. You can only reach them from the list of price lists.
How to add a parser source, set up site crawling (sections → product links → product page), write a , and check it with the test button is covered in Site parsing.
- Menu
- Directories → Processing price lists → List of price lists
- Rules grid
- the “Site parser” source row → “Setup” → the “Import settings” tab
- Access right
- the “Price lists” window (in the menu: “List of price lists”); there's no separate right for the grid
- Prepare
- a parser source that's already added, and the address of one product page to check the fields on
1
Product card group2
Field name column3
Selector #1 column- Each group of rules is an operation type; the Product card group holds the fields collected from a product page.
- The Field name column lists what the parser can read: one field per row.
- The Selector #1 column holds the CSS or XPath selector that finds the value on the page; an empty cell means the field is not collected.
These windows have no “Save” button: an edit in a grid cell goes to the server right away.
Related topics
- If the parser source isn't added yet or the crawl doesn't reach the product page: Site parsing.
- If after a run the prices are old, products are duplicated, or the crawl broke off: How the parser updates the price list.
- If specifications were collected but aren't in the catalog, or the attribute directory has grown too large: Site specifications in the catalog.
- If the stock text from the site didn't become your status: Stock statuses.
- If the wrong photo became the main one after the transfer to the catalog: Product photos.
- If you're choosing a parser for your store: what it gives your store.
