Elbuz finds your catalog products on source sites and copies descriptions, photos, specifications and videos into their cards. Hundreds of products go in one batch, and AI checks it found your product.
When you need this:
- The price list has only name, SKU and price: the search tries cascade sites until a match.
- Some cards are proofread by hand: content can go into empty fields only.
- The product is not on your sites: the “Google” mode finds sites and writes the description.
- New products arrive daily: background search fills them by preset.
Where to start: filter products in the base catalog and open “AI & content” → “Product content search”.
What content search can do
- Three modes: a single site, a cascade of your sites or an internet search through Google.
- The “Google” mode finds sites with the product, merges them into one card in the card language and writes the description in its own words.
- The site cascade goes through sources from top to bottom and stops at the first one where content is found.
- AI verification and content requirements keep someone else's product and poor cards with a single specification out of your cards.
- The “What to save into the product” panel decides whether saving touches what you have already proofread by hand.
- Run history shows which site in the set is unnecessary and which one is missing.
- Content for a single product copies the description and photos from a page you opened into the card.
- Background search finds content for new products after a catalog update without your involvement.
How to choose products for the search
The product set comes from the catalog; you do not define it in the search window. There are two conditions: “Checked in the catalog”, the products you ticked, or “By catalog filter”, everything in the current filtered set. You can change the set only in the catalog itself: close the window, adjust the filter and open the search again.
Here you also choose the category and the field to search by. By default the search uses the name, but searching by the manufacturer's part number is often more accurate.
If no filter is set in the catalog at all, the whole catalog goes into the run, and the window warns you about it directly. Start with a category or a filter, not the entire catalog at once.
One run takes no more than 5000 products from the filtered set, and 1 in demo access. If the filter catches more, the extra products simply do not go into the run, with no separate message about them.
Products that have already been checked on the selected sites are excluded from the set, and the window says how many. This protects you from spending requests twice; they are excluded from those 5000, so fewer products may go into the run. The “Check them again” button lifts this restriction.
If products are ticked or the window is opened on a single product, the “Do not search again for” rule does not apply to them: the program searches exactly what you selected and only warns how many of them have already been checked.
Which mode to choose: single site, cascade or Google
You set the mode in the “Where to search” step of the “Run” tab. The rest of the window is the same for all three: product selection, “Progress”, “Results”, “History” and the “What to save into the product” panel.
| Mode | How it works | When to use |
|---|---|---|
| “Single site” | searches only on the chosen site and copies its text unchanged | you need content from one specific store or supplier |
| “Site cascade” | goes through your sites from top to bottom and stops at the first one where content is found and confirmed by AI | you have proven sites; requests through the browser are free |
| “Google” | finds sites with the product on its own, merges up to 10 sites into one card in the card language and writes the description in its own words; queries and pages are paid on the Elbuz server and free through your browser | there are no sites for your products, and you need the fullest possible specification table and your own description |
Google search in brief
In the “Google” mode you do not choose sites. The program searches Google for the product, AI picks this product's pages and checks each site, then builds one combined card from the confirmed ones. 0.05 € is charged only per saved product. “Run the search on” sets who opens Google and the pages (comparison):
- “Elbuz server” searches in the background, you can close the window; queries and pages are always paid.
- “My browser” searches through the Elbuz extension in your Chrome without paying for queries and pages; the window must stay open.
The same place also offers collecting a specification table from the text of the card itself, without a search. Learn more: how to fill cards using Google search.
How to build a cascade of source sites
| Mode | How it works |
|---|---|
| “Single site” | searches only on the chosen site |
| “Site cascade” | sites are tried from top to bottom; the cascade stops at the first one where content is found and confirmed by AI |
- Action or result
- Decision: the path depends on the answer
A cascade is built from presets, saved sets of sites. The “Suggest by category” button offers a set whose categories match the categories of the selected products; such a preset is marked “Recommended”.
Each site row shows:
- “Success rate”: the share of products for which the site returned content over the last 90 days. The “Sort by success rate” button moves the most productive sites to the top.
- “Browser” or “Server”: where the pages are fetched. Browser sites need the tab to stay open and work through the Elbuz extension for Chrome.
- the “⚡ Fast” label: the site returns data in bulk through its own API, so pages are not loaded and the run is many times faster.
- an origin mark: a ready-made source (set up in advance and updated automatically) or your own.
The checkbox to the left of a site turns it off for this run only; you change the order by dragging the number. The “Own site” button adds a source that is not in the list: the site address, its search address and the page parsing rules.
What to collect and when a product counts as found
First, the language: the “Write into language” field sets which language version of the card receives the found content. “Single site” and “Site cascade” do not translate text; only the “Google” mode does. What to collect from a site (photos, specifications, description, video) is set in that site's settings.
Next come three settings that affect the result the most.
- “Do not search again for”
- 7, 30, 90 days or “always search”. A product recently checked on this site does not go into the run
- “Park the product”
- after how many sites without a result the product gets the “Content not found” status and leaves the working set. Set as many as the sites you actually use: with a larger number the mark will never appear
- “If content is found nowhere”
- set the “Content not found” status or leave the status unchanged
The “Search criterion cleanup” group hides two things that most often make the search “find nothing”:
- Remove special characters: brackets, quotes and hyphens are removed from the query. They confuse the search on most sites.
- Remove Cyrillic: Cyrillic words are removed from the query, so «холодильник Samsung RB34» (refrigerator Samsung RB34) is searched as “Samsung RB34”. Careful: for a product with a fully Cyrillic name the query becomes empty and the site is skipped.
The “Allow paid requests to the Elbuz server (such sites in the set: N)” checkbox is right here, in the “How to run” step of the “Run” tab. It enables requests to the Elbuz server, which are billed separately; the number in the label is how many server sites the selected set has.
Without this permission the run does not start if the cascade has at least one site loaded by the server. On start you see “Server-side loading is a paid feature. Allow paid requests in the run settings or switch to the browser.” The “run settings” are this very step, not the settings of the parser source.
The free alternative: in the same step, in the “How pages are loaded” field, choose “By browser (Chrome extension, free)”. You then need Google Chrome with the Elbuz extension and an open tab. Sites labeled “⚡ Fast” always run on the server: if the cascade has one, allow paid requests or remove it from the set.
What counts as found
When a site is done, the program decides for each product whether to stop the cascade on it. Two filters stand in the way:
- “AI check”. With “Verify that the right product was found” on, AI rates each candidate from 0 to 100%. If none reaches “Minimum confidence, %” (80 by default), the finding is deleted and the product moves to the next site.
- The “Search result” group: the “Minimum attributes” and “Minimum photos” fields. The variant that will go into the card is checked. 0 in a field means the threshold is not checked, and both fields are 0 by default.
- Action or result
- Decision: the path depends on the answer
Rejected findings do not disappear silently. After each site the log on the “Progress” tab writes a summary: how many were confirmed, how many AI rejected, how many fell below the quality threshold and how many were left without content.
1
Minimum confidence, %2
Minimum attributes and photos- Minimum confidence, % — candidates scored below this threshold are rejected and the product moves on to the next site.
- Minimum attributes and minimum photos — a card poorer than the threshold counts as not found; 0 means no check.
Save without review
A separate switch writes found content straight into the card when AI confidence reaches the set value. It needs AI verification and does not work in the “Google” mode. It follows the “Save” rules (what is overwritten), so found photos fully replace the old ones. Turn it on once you trust your sources; otherwise keep operator review.
How to track a run and assess it
Pre-flight summary
Before the start the window shows four numbers: products, sites, requests and minutes. In the “Google” mode “€ search maximum” replaces sites; with “My browser” it is 0. The run works in the background: you can minimize the window and follow progress in the operations center.
The minutes are a rough estimate: request count × 1.5 seconds. The first cascade site gets all products, each next one what is left after the previous one, based on its “Success rate” (50% if zero). “Pages at a time” sets how many pages the browser loads in parallel: 6 by default, 10 at most.
Before the start the program also compares the cost of the run with your balance. If the money is not enough, the start is rejected before the first request with a message showing the amount and the balance, instead of breaking off in the middle of a batch.
- Action or result
- Decision: the path depends on the answer
The “Progress” tab
The tab shows the current site and how many products are done, queued, found and rejected by AI. Next to it is the operation log with the “All”, “Errors only” and “Not found only” filters. You can pause or stop a run. The “Google” mode has only “Stop”, no pause; when the run is over, these buttons are gone.
If a site asks you to pass a check (a captcha), the search on it is paused and the window offers three actions: open the page and pass the check by hand, continue from the same place after that, or skip the site. The run does not bypass captchas on its own.
In the “Google” mode the log shows every step for a product: the query, the selected sites, page loading with links and the reason for a failure (what the reasons mean).
At the end the program writes a summary: how many were found and how many were left without content. If sites often responded with errors, it says so separately; this usually means the site changed its layout or a stronger loading mode is needed.
The “History” tab
“History” answers the question “is this work worth the cost”. For each run you see how many products were taken, found, saved and rejected, and how much time and money it cost.
A separate breakdown of the cascade by step shows how many products reached each site, how many it closed and how many moved on. This table shows which site in the set is unnecessary and which one is missing. You can export the statistics to Excel.
How to review results and save them to cards
Found content does not go into cards on its own. The “Results” tab is where the operator decides what to keep.
| Results filter | What it shows |
|---|---|
| “Found” | all products of the run that currently have found content |
| “AI confirmed” | AI compared the page with your product and confirmed it is the same |
| “Needs review” | AI did not confirm the match or the check was off, so a person decides |
| “Saved” | the content has already been written to the cards |
| “Content not found” | nothing was found on any of the sites tried |
A row shows your product next to the source finding, the confidence percentage and what arrived: photos, specifications, description. Difference highlighting marks words and codes missing from the other name: that is where “similar, but not the same” products hide. If the site found several pages, pick the right variant in the panel on the right.
The “Save” button writes found content to the cards of the ticked products; “Reject” removes the finding from the results without changing anything, and such products can be searched again. The “Verify the results with AI” item in the tab menu runs the check manually for findings AI has not looked at yet.
Each saved product charges your balance for writing content (except plans where writing is already paid for). After saving, the product disappears from the results: you can no longer see in the search window what arrived, only in the card.
Each new search run clears the results: unreviewed findings disappear. The window warns you about this before the start. Review and save before you start the next run.
There is no undo for saving: it does not record previous values anywhere, and old photos are deleted if new ones are found. What you can do after a wrong save is described in “If the wrong content was saved”.
When saved content reaches your site
Saving by itself sends nothing to your site or marketplaces. It writes content to the base catalog and clears the “Exported (flag)” mark on the product: this is how the program notes that the product data has changed. The content goes to your site with the next export, whether you start it manually or it runs on a schedule.
- If “Only export new data” is on in the export, saved products get into it precisely because of this cleared flag.
- If it is off, the export takes all products that pass its selection conditions, saved ones included.
How to keep your own data when saving
What “Save” writes is set by the “What to save into the product” panel, the slider button next to “Reject”. The same settings apply to “Save without review”. The main thing decided here is whether saving touches what you have already proofread by hand.
| What is in the card | What happens on save |
|---|---|
| Full description, short description, manufacturer, SKU, model, barcode, name | if “Do not overwrite filled fields” is on, found content goes only into empty fields and yours stays. If it is off (the default), a filled field is replaced with the found value. An empty value from the site never erases a field |
| Specifications | “Do not overwrite filled fields” does not apply to them. If “Delete previous attribute values” is on, a product that received at least one specification first has all its specification values erased, except attributes marked “Read-only (value is never removed automatically)” in the attribute list, and only the found ones remain. Your manual specifications that the source site lacks will disappear. If it is off, values of specifications present on the site are replaced (except “Read-only” attributes), new ones are added and the rest of yours stay |
| Photos | found photos replace the old ones entirely: the old ones are deleted. If no photos are found, the old ones stay, and the preview shows, for example, “photos 18 → 18” |
| Video | if a video was found on the site, the product's videos are deleted and the found ones are written |
| “External link” | filled with the source page address only if it was empty |
1
Full description2
Short description3
Product name- Full description — the site's long text replaces the description in the product card.
- Short description — the short text is written to the card's short description field.
- Product name — off by default, so the card keeps its own name unless you tick it.
To lock your own value, untick that field or its whole group in the panel: an unticked field is not written at all. For a single product you can untick a field right in the preview panel on the right: this applies only to saving that product and does not change the general settings.
For a specification driven by a catalog field formula marked “Read-only (do not replace value when parsing)”, the value is not erased, appended to or replaced with the found one, whatever the “Delete previous attribute values” checkbox says.
The panel checkboxes are remembered. Every change in “What to save into the product” is saved to your window settings on the server right away: next time, from any computer, the panel will be as you left it. Each user has their own settings.
If the text generator has active product templates (without AI) that use specifications, after saving these templates are applied to the saved product and overwrite the fields they fill: descriptions, meta tags, , tags. “Do not overwrite filled fields” has no effect on this.
New specifications and where they go
Along with the content come specifications your catalog does not have yet. Before saving, match them with your own: if the source site has “Screen size” and you have “Diagonal”, matching keeps the program from creating a duplicate.
Copying specification values into catalog fields works separately: if a field and an attribute are matched, the value goes into the field too.
New specifications appear in the attribute list as regular ones, and you can delete them there like any other. Deleting an attribute removes all its values in all products, its category links, the column in the catalog grid, formulas on this attribute and the field in export templates. The program cannot restore these values.
If such a specification is essentially the same as yours, it is safer to match it with yours first and delete only the one you really do not need.
Scenarios
Fill new products without touching cards you already proofread
Task: the catalog has cards you filled in by hand next to new products from a price list with no description.
What to turn on: in the “What to save into the product” panel, turn on “Do not overwrite filled fields” and untick “Delete previous attribute values”. If the cards already have your own photos, untick “Photos” too.
What you get: the description, manufacturer, SKU, model and barcode are written only where the card is empty. Specifications present on the source site replace the values of the same specifications in the card (except “Read-only” attributes), new ones are added and the rest of yours stay.
What you do not get: “Do not overwrite filled fields” does not protect specifications from replacement; it applies only to card fields. If you leave “Photos” ticked and photos are found, the old ones are deleted and replaced with the found ones.
Find products that were not found on the first set of sites
Task: after a run some products were left without content, and you want to try other sources.
What to turn on: in the “Where to search” step, choose other sites or add new ones to the set. Leave “Do not search again for” as it is, with the “By catalog filter” selection.
What you get: a product already checked on every site of the new run within the “Do not search again for” period is not processed. If at least one site is new for the product, it goes into the run and passes the set's sites from top to bottom, including those where it was searched before.
“Park the product” counts different sites without a result across all runs together: with three in this field, a product not found on three different sites gets “Content not found” at the end of the run, even if the sites were tried in separate runs.
What you do not get: “Do not search again for” does not apply to products ticked in the catalog: they go into the run no matter how many times they were searched before.
Skip empty product pages from a site
Task: a source site often finds the product but returns a page with one specification and no photos, and the cascade stops there.
What to turn on: in the “Search result” group, set “Minimum attributes” and “Minimum photos”.
What you get: if the variant that would go into the card has fewer specifications or photos than the threshold, the content counts as not found: the finding from this site is deleted, the product moves to the next cascade site, and the log shows a “rejected” line with the actual number and the threshold.
What you do not get: the program will not take another variant from the same site instead of the poor one: only the variant that would go into the card is checked.
Search by direct links to supplier pages
Task: the products already have the address of their page on the supplier's site, so searching by name is not needed.
What to turn on: in the “Which products” step, choose “External link” in the “Search by field” field.
What you get: only products with a filled “External link” go into the run. A site in the set without a search address is not dropped in this mode: the address is taken from the card.
What you do not get: products with an empty link do not go into the run at all. A site missing server-side fetch settings is skipped in this mode too.
What to do when the search finds nothing or saved the wrong content
| What you see | Why | What to do |
|---|---|---|
| “…so none will be processed” | products were taken “By catalog filter”, and all of them have already been checked on these sites | choose other sites or click “Check them again” |
| “The selected products have no “External link” — search by link skips them” on start | “External link” is selected in “Search by field”, and this field is empty in the ticked products | fill in “External link” in the cards or choose another field in “Search by field” |
| Nothing is found on any site | the query contains the full name with the product type and symbols | turn on special character and Cyrillic cleanup or search by the manufacturer's part number |
| A site is skipped without explanation | after Cyrillic cleanup the query became empty | turn off Cyrillic cleanup for such products |
| On start: “Server-side loading is a paid feature. Allow paid requests in the run settings or switch to the browser” | the set has a site loaded by the server, and paid requests are not allowed | “Run” tab → “How to run” step → the “Allow paid requests to the Elbuz server” checkbox, or browser loading for the site (details) |
| A browser site returns nothing | the tab is closed or the extension is not working | keep the tab open and check the Elbuz extension |
| The results disappeared | a new search was started before the previous findings were saved | save before the next run |
| The “Google” mode did not find the product or returned text in the site language | reasons of its own: auto-pause, the product name, site protection | troubleshooting for the “Google” mode |
| The wrong photos were saved | a similar product was found and AI let it through | use the “Needs review” filter and difference highlighting |
| Duplicate specifications appeared in the catalog | source attributes are not matched with yours | match attributes before saving |
If the wrong content was saved
Content saving has no undo button, and it does not record previous values anywhere. What you can get back:
- Photos: no. The old ones are deleted when the new ones are written; the path to a deleted file stays only in an internal deletion log, and you cannot restore photos from the interface. A backup will not help either: the database backup does not contain photos (backups). Photos gone after saving are not a failure but a replacement under the “Photos” checkbox: if the cards already have your own photos, untick it before saving.
- Descriptions and specifications: only by filling them in again, since the program does not remember replaced values.
Two base catalog columns help you find products that received content from a specific site:
- “Attribute source”: for products where saving added specifications, it shows “Site: ” and the source site name. Filter the column by the site name.
- “External link”: if it was empty before saving, it now holds the address of the page the content came from. A filter by the site's domain shows these products.
To replace someone else's content with the right one, tick these products in the catalog and run the search on another site or with another search field. The “Do not search again for” rule does not apply to products ticked by hand.
The new save overwrites fields by the same rules from the “What to save into the product” panel. Specifications missing on the new site are removed only with “Delete previous attribute values” on. Photos and descriptions that exist nowhere have to be added to the card by hand.
FAQ
Do I need the Chrome extension for content search?
For sites in the “Browser” mode and for Google with “My browser”, yes, and the tab must stay open. Sites in server mode and Google with “Elbuz server” work without it, but server requests are billed separately (paid requests).
Which should I choose: cascade or Google?
The cascade uses only your sites, stops at the first with content, and browser requests are free. Google finds sites itself, merges specifications into one table in the card language and writes its own description; paid on the Elbuz server, free in your browser. Saving costs the same (mode comparison).
Why is a site cascade better than a single site?
The cascade goes through sources from top to bottom and stops at the first one where content is found and confirmed. A single site covers part of the products, a cascade covers most of them, with the same clicks.
What does AI verification do?
It compares the found page with your product and gives a confidence percentage. Confirmed findings can be saved in bulk, the rest stay under the “Needs review” filter.
What does the “Content not found” status mean?
The product was tried on the set number of sites or was not found through Google. It leaves the working set so that requests are not wasted. Next, either add a new source to the set or fill in the card by hand.
Old product photos disappeared after saving. Can I get them back?
No. The “Photos” checkbox in “What to save into the product” means replacement: found photos delete the old ones with no undo; otherwise the old ones stay. You can only upload them again. To keep future saves off your photos, untick “Photos” (if the wrong content was saved).
Are the checkboxes for what to save into the card remembered, or do I set them every time?
They are remembered automatically, with no separate button: every change in the “What to save into the product” panel is saved to your settings on the server and works from any computer. Only a checkbox unticked for a single product in the preview panel is not remembered.
Can the search write a product description on its own?
In the “Google” mode, yes: AI writes the description in its own words in the card language, using only facts from the sites. “Site cascade” and “Single site” copy the site text unchanged; for them, your own text is written by the text generator.
Where to find it and who has access
- Menu
- Base catalog → “AI & content” → “Product content search”
- Access right
- the “Product content search” window in “Group permissions”
- Prepare
- a filtered set in the catalog: the products that need content
1
Single site / Site cascade / Google switch- Search mode switch — the cascade goes through the sites top down and stops at the first one that returns confirmed content.
The automatic search window has four tabs: “Run” for the run settings, “Progress” for the progress and log, “Results” for what was found, and “History” for statistics of past runs.
Related topics
- If browser sites in the cascade return nothing: the Elbuz extension for Chrome.
- If you want to add your own source and need to set page parsing rules: CSS and XPath selectors.
- If found specifications create duplicates: the attribute list.
- If there are no source sites for your products and you need the program to find data on the internet itself: Google search.
- If you need to understand what happens to card images: product photos.
