In the “Google” mode, Elbuz finds sites with your product, has AI confirm the match, and merges specifications, photos and descriptions from several sites into one card. No site setup needed.
When you need this:
- New products have no sites: up to 10 sites per product.
- You need your own description: AI rewrites site text without inventing numbers.
- Photos have watermarks: “Do not take photos” drops a site's photos.
- No charge for queries: “My browser” searches through your Chrome.
- Which site is worth paying for: “Statistics” shows data, costs and rating.
Where to start: filter products in the catalog, open “AI & content” → “Product content search” and click “Google”.
What Google search can do
- Finds sites itself: AI picks pages of this exact product from Google results and rejects articles and other models.
- Merges several sites into one card: one specification table in the card's language and a description in its own words.
- Limits where to take data from: the lists “Search only on these sites” and “Do not take content from these sites”, two site lists for photos and a list of individual photos not to take.
- Searches through your browser: Google queries and site pages cost nothing while the search window is open.
- Builds a table from the card text: no search, when the description is there but specifications are empty.
- Takes the largest product photos: from the whole page, without placeholders or photos of neighboring products.
- Shows where each item came from: the site of every photo and specification, plus a photo slideshow captioned with the site.
- Measures how useful sites are: how many pages gave data, what they cost, photo size, site status and rating.
What the program does with each product
The steps are the same for both options of the “Run the search on” field: the only difference is who opens Google and site pages, the Elbuz server or your Chrome. With “Elbuz server”, up to 15 products are processed at once; if you close the window, the search continues, but the queue picks up new products once a minute.
- Searches Google in the search country. The country is set by the “Search country” field; by default it's the country of the catalog language: Ukraine for Ukrainian and Russian, the USA for English. The first query is by barcode, if there is one, then up to two queries by name with the “Word added to the name” (by default “buy” or its equivalent in the country's language). The next query runs only if the previous one gave no usable pages, so a country takes up to 3 queries.
- Picks pages. AI reads the result titles and snippets and keeps possible pages of this product: one page per site, up to 5 sites from one country (with “Search only on these sites”, up to the “Sites per product” limit). Articles, reviews, forums, categories and other models are rejected. Marketplaces without their own product pages (OLX, AliExpress, Allegro and the like) are dropped unless the site has a system parser or driver or you listed it in “Search only on these sites”. Sites in the .ru zone are never used.
- Loads pages in parallel and waits up to 90 seconds for each. If a page doesn't arrive, the program tries another loading service; a page that doesn't exist (404, 410) is not retried. Rozetka and Hotline go through their fast drivers, without loading the page. In “My browser” mode your browser opens the pages: it retries a page it couldn't open once more, and there are no fast drivers.
- Parses the page with the site's parser: the system one if it exists, otherwise yours. For a site with no parser, AI sets one up from this page and remembers it. A site whose parser you disabled is dropped before loading and is not charged.
- Checks each site. AI verification is always on here: a site with a different product is rejected. The threshold is the “Minimum confidence, %” field; 0 works as 80. A site with no specifications and no description is rejected too.
- Adds other countries only if data is scarce. If no confirmed site gave 15 specifications and the “Sites per product” limit is not used up, steps 1–5 repeat in turn in the countries from “If data is scarce, also search in” (by default Poland, then the USA). The query uses the word in their language. Sites already visited are not repeated. If you untick “Attributes”, one confirmed site is enough.
- Builds the merged card even from one site. A card that fails “Minimum attributes” and “Minimum photos” is deleted, and the product counts as not found.
- Action or result
- Decision: the path depends on the answer
| What | Result | If the AI answer is rejected |
|---|---|---|
| Specifications | One table in the language from the “Write into language” field. One specification under different names (“Weight” and “Mass”) becomes one row and gets the name from your catalog, if it is there. Only values from the sites are taken; AI adds nothing | The table is merged only by identical names and stays in the sites' language |
| Name | Only the word naming the product type is translated. Brand, model, codes, color and sizes stay as on the site | The name from the site stays |
| “Full description” | AI writes it anew in its own words from the fullest text on the sites, roughly up to 1,500 characters. Numbers, codes and models are not changed. Price, delivery, warranty, promotions and store names are not allowed | No description: the program does not keep a copy of the site text |
| “Short description” | The same from the fullest short text on the sites, as plain text, roughly up to 300 characters | No description: the program does not keep a copy of the site text |
To avoid duplicates, AI gets your catalog's specification names: those filled in products of the same category, pinned to the category, or close in meaning to the site names. A row that means the same is named exactly as in the catalog. AI does not add catalog names that are not on the sites.
- Translation. Names and values are translated into the card's language: color, material, country, “yes/no”, warranty. Brands, models, series, codes, abbreviations (USB, Wi-Fi, LED), units of measurement and trademarked technology names stay as on the site.
- Merged rows. “Waga: 9,9 kg” written as one text is split into a name and a value. A row with no value (a heading, a question) is skipped.
- Not specifications. The store's SKU or product code, EAN, price, availability, delivery, payment and store phrases such as “Why buy from us” do not go into the table.
Close names are matched once the attribute list is indexed for matching. After the run, findings are matched against the list again if the “Map specifications to the catalog” checkbox is ticked in the “How to run” step (it is on by default) or “AI attribute matching” is on in the general settings. More: matching in content search.
The AI answer is checked automatically. A specification with a number that is not on the stated site is rejected, and a table that keeps less than 70% of the rows of the fullest site is not accepted at all. Text values are not checked this way, so review them before saving.
A description with a foreign number, one that is too long or too short, or an exact copy of the site text is not accepted; if the second attempt fails too, the description is removed from the result so that no copy of the site gets into the card.
AI does not write a description from its own knowledge: no text on the sites means no description. Nothing is written to the card until you click “Save”.
A product already searched via Google is not searched again within the “Do not search again for” period. Earlier attempts on cascade sites don't affect this. Products ticked in the catalog are always taken.
What to set before you start
Where: the “Run” tab → the “Where to search” step → “Google”. Every field change is saved to your window settings right away.
1
What data to get2
Search only on these sites3
Do not take content from these sites4
Take photos from these sites first5
Do not take photos from these sites6
Sites per product, at most7
Do not take these photos- What data to get — the types of content the search collects.
- Search only on these sites — the search is limited to the listed sites, one per line.
- Do not take content from these sites — sites excluded from the search.
- Take photos from these sites first — the preferred photo sources, the top one matters most.
- Do not take photos from these sites — sites whose photos are never taken.
- Sites per product, at most — how many sites are loaded per product.
- Do not take these photos — account-wide list of individual photo addresses to skip.
In all site lists, put each site on a new line, with or without the address; subdomains are included with the site. Sites in the .ru zone are not used, even if listed.
| Field | What it does | Default |
|---|---|---|
| “Run the search on” | “Elbuz server — in the background, paid” or “My browser — no charge for queries”. The difference is explained below | “Elbuz server” |
| “Sites per product, at most” | From 1 to 10. Every loaded page takes a slot, including an extra page from the same site, and so does a card through a fast driver. Fewer sites cost less but give less data | 10 |
| “What data to get” | The “Photos”, “Attributes”, “Video”, “Full description”, “Short description” checkboxes, the same as in the “What to save into the product” panel. AI spends no tokens on unticked items. Pages are loaded and charged the same way | all on |
| “Search country” | Google searches as a shopper from this country and shows its stores. Nine countries: Ukraine, Poland, Germany, Austria, France, Italy, Spain, the USA, the United Kingdom | the country of the catalog language |
| “Word added to the name” | Added to the product name so Google shows stores rather than reviews. Empty — search by the name only. As long as you haven't changed the word, it switches to the new country's language when you change the country. Fallback countries always use the word in their own language | “buy” or its equivalent in the country's language |
| “If data is scarce, also search in” | Checkboxes for the countries the search goes to when your own country gives little data. Your own country isn't on the list. Each ticked country adds up to 3 Google queries per product | Poland and the USA, except your own country |
| “Do not take content from these sites” | The site is dropped right in the results: the page is not loaded or charged. This list overrides all the others, even “Search only on these sites” and a ready site parser. The “Do not take from site” icon also adds to it | empty |
| “Search only on these sites” | A strict filter, not a priority: Google searches only on these sites, by product name, without the barcode. Other countries are not used, and a product that isn't there stays not found. Every 5 sites is a separate paid query | empty: searches everywhere |
| “Take photos from these sites first” | Line order is priority, the top one matters most. Photos are ticked from the first listed site where they were found, even if there are few. The field does not limit the search itself. The “Use these” switch adds to it | empty: the site with the most photos |
| “Do not take photos from these sites” | The site's photos don't go into the merged card; specifications and description are collected as usual. Good for sites with watermarks. The “Do not take photos” link adds to it | empty |
| “Do not take these photos” | Addresses of individual photos, one per line. Such a photo is not loaded for any product. The cross on a thumbnail in the results adds to it; erase a line and the photo will be taken again. One list for the whole account | empty |
When to untick extra countries. Fallback countries can easily eat up most of the queries: a run on 71 products used 61 queries in Ukraine, 70 in Poland and 69 in the USA. If you sell only in your own country or foreign sites give nothing useful, untick them in “If data is scarce, also search in”. Each unticked country means up to 3 fewer queries per product, and in “My browser” mode also fewer “I’m not a robot” checks.
A site can't be in the photo priority and the photo ban at once: when “Use these” or “Do not take photos” adds a site to one list, it is removed from the other.
Where to run the search: the Elbuz server or your browser
The “Run the search on” field decides who opens Google and site pages. The rest of the work (picking pages, AI verification, the merged card) is the same.
1
Run the search on: My browser2
0 € search maximum- Run the search on — set to “My browser — no charge for queries”.
- 0 € search maximum — searching in your own browser costs nothing.
| “Elbuz server — in the background, paid” | “My browser — no charge for queries” | |
|---|---|---|
| Who opens Google and sites | the Elbuz server | the Elbuz extension in your Chrome |
| Queries and pages | charged at the rates | free; you pay only for AI tokens and €0.05 per saved product |
| Search window | can be closed, the search continues | must stay open |
| Without the extension | works | the search doesn't start: the window asks you to install or enable the extension |
| Products at once | up to 15 | 3, no more than 3 tabs; a Google search no more than once every 4–6 seconds (pace) |
| “I’m not a robot” check | doesn't concern you | Google may show it: what to do |
| A page failed to load | retry through another loading service | one retry in the same browser if the page didn't open |
Google results contain no site addresses, only Google links: the browser opens such a link and goes to the site itself, and the program takes the real address where the browser ended up.
Google pace. A Google search runs no more than once every 4–6 seconds, and any request to google.com (a search or following a link to a site) no more than once every 2–4 seconds. There's no setting for the pace: the captcha is triggered by the number of queries, not the pauses between them, so to get fewer checks, untick extra countries in “If data is scarce, also search in” (when to do it).
If Google asks “I’m not a robot”. When one browser sends many queries, Google may show a check instead of results. The search pauses, the window plays a sound and shows the “Google asks for a verification (captcha)” banner. The extension opens the check itself in the parsing window — a separate Chrome window where it loads pages — and brings that window to the front.
Pass the check right there: in a new tab with the same address Google won't show it, and the next query will run into it again. A passed check resumes the search from the same product by itself. If it doesn't resume, go back to the app and click “Verification passed, continue”. An old extension (before 2.4.4) behaves differently — update the extension.
If nobody responds for 4 minutes (window closed or check not passed), the search stops with the log line “The browser has not responded for 4 minutes, the search is stopped…”. Products not yet reached get no marks: run the search again. Same with the “Stop” button: a product that was waiting for the browser doesn't count as not found.
How to collect specifications from the card text, without a search
The “No search: card text only” checkbox is for when the description is there but specifications are empty. AI reads the card's descriptions and extra fields in the language from the “Write into language” field, plus the specifications already filled, and builds a table from them. No sites are loaded, and the step's other fields are hidden.
- Condition: the card texts total at least 400 characters. If less, the product stays not found but doesn't get the “Content not found” mark in the catalog.
- Check: filled specifications are kept; a specification with a number that is not in the card is rejected. “Minimum attributes” applies, “Minimum photos” does not.
- Cost: only AI tokens and €0.05 per saved product. “€ search maximum” shows 0.
- Result: after “Save”, photos and description stay unchanged; only specifications are written.
- Repeat search: this collection doesn't count as a Google search, so it doesn't block a real search for the product.
How much it costs
With “Elbuz server”, queries and pages are always charged, even if the product is not found. With “My browser” they are free: AI tokens remain, and “€ search maximum” shows 0. The €0.05 per product is charged only for products saved to the card, in both modes.
| What | Price | When charged |
|---|---|---|
| Google query | €0.0045 | every query run; “Elbuz server” only |
| Site page | €0.0045, €0.0135 or €0.0225 depending on the site's complexity | every page that arrived with a normal response, including a captcha page or a site AI couldn't set up a parser for; “Elbuz server” only |
| Card through the Rozetka or Hotline driver | €0.0004 | every product through the driver; “Elbuz server” only |
| AI tokens | by usage | picking pages, verification, merging the table, the description |
| Saving to the card | €0.05 | only for a saved product |
Technical failures are not charged: a page with an error code (403, 404 and so on) or an almost empty one. The “Unlimited” plans and fixed pricing cover saving only; queries, pages and tokens are paid separately. Charges go through every 10–15 minutes, and an amount under €0.01 is rounded up to €0.01.
“€ search maximum” is the worst case for the whole batch, without tokens and saving.
Per product that is up to 3 queries for each country — your own and the ticked fallback ones; by default there are three countries, i.e. 9 queries (two countries, 6 queries, for a Polish or English catalog). Plus as many pages as set in “Sites per product”, each at the most expensive rate plus a driver: with 10, that's €0.2695 per product, so 100 products come to about €27.
With “Search only on these sites”, one query per 5 sites is counted instead of 9. Actual costs are much lower.
At start, your balance is checked against exactly this limit plus saving all products. If the balance is short, the search doesn't start and the window shows how much you need. During the run the balance is not checked separately. For a large batch, narrow the filtered set or lower “Sites per product”.
Autopause. If at least 30 products are processed and less than 10% are found, the search stops: the window shows “Under 10% of products found… Continue?”. If the window was closed, you'll see the unfinished run banner. This guards against spending on a batch Google can't find: check the product names before continuing.
How to check the merged card
Findings wait on the “Results” tab. Click a product and the preview panel opens on the right. Saving works as in the other modes (review and saving). Below is what only Google results have.
The “Found on the site” block
The block appears when there are several sites: it has a row for each site with its source text and a separate row for the merged card labeled “Collected from several sites: N”; the same label shows under the product name in the panel.
A site row has the address, an arrow that opens the product page, and the “Do not take from site” icon. With only one site there is no block: the arrow next to the product name opens the page, and you can ban the site on the “Statistics” tab.
1
Collected from several sites2
Site address3
Open the product page4
Do not take from this site- The summary row shows how many sites the result was collected from.
- Each site row shows the address of the site.
- The arrow opens the product page on that site.
- The ban icon stops taking content from this site in future searches.
After you confirm, the “Do not take from site” icon adds the site to the “Do not take content from these sites” field. From the next search, its pages are not loaded or charged. Results already found don't change: to remove the site's data from this card, reject it and search again.
Photos: site groups, “Use these” and “Do not take photos”
The gallery holds photos from all confirmed sites by site, except those in “Do not take photos from these sites”. The first 10 photos of one site are ticked: the first site from “Take photos from these sites first” that has photos, otherwise the site with the most photos. The rest stay unticked, because other stores' photos may carry watermarks.
1
Group header: site and photo count2
Do not take photos3
Use these4
Checked photo5
Photo from another site, unchecked- Each group header shows the site and the number of its photos.
- Do not take photos adds the site to the photo blacklist.
- Use these marks the site's first photos and prefers them in future searches.
- A checked photo goes into the product card.
- Photos from other sites stay unchecked until you mark them.
- “Use these” (shown when several sites have photos) ticks the first 10 photos of this site, unticks the others and puts the site first in “Take photos from these sites first”. It applies to this product at once and to later searches through the list.
- “Do not take photos” removes the site's photos from all found, not yet saved Google search results, adds the site to “Do not take photos from these sites” and removes it from the photo priority. Where only its photos were ticked, the program ticks the site with the most photos left. Saved products and individual site rows don't change.
1
Confirm button- The confirm button adds the site to the photo blacklist; Cancel closes the dialog without changes.
Photos removed with “Do not take photos” don't come back to the results, even if you later remove the site from the list. Only a new search for that product can bring them back.
The cross in the thumbnail corner deletes one photo, and its address is remembered for good. Use it for a banner, delivery badge or site image that shows up in every product. Before deleting, the window asks: “Delete this photo? It will no longer be loaded for any product: this is how you remove site banners and images.”
- What is remembered: the photo address, one list for the whole account. It appears in the “Do not take these photos” field in the “Where to search” step of the “Google” mode. Addresses are compared without “www” and without parameters after “?”, so the same photo with a different address tail isn't taken either.
- What changes right away: the photo disappears from this result. In other results already found and in saved cards it stays; remove it there by hand.
- How to undo: erase the line in the “Do not take these photos” field. The photo will be taken from the next search, but it won't return to results already found.
Clicking a thumbnail opens a slideshow from that photo. Each photo is captioned with its site, so you can see at once whose watermark it is. Use the arrows to browse and Esc to close.
1
Site address of the photo2
Next photo3
Photo counter- The caption under the photo shows the site it came from.
- The arrow moves to the next photo.
- The counter shows the current photo number out of all.
The program collects each site's photos from the whole page, not only where the parser found them: the gallery, click-to-zoom, page markup, photos the browser displayed. Of several versions of one photo, the largest is taken, including the original instead of the preview on OpenCart and Bitrix sites.
- Taken: photos the parser found, the page's main photo and photos with the product's model code in the address or caption, up to 15 per site.
- Rejected: photos of neighboring products (“Similar products” and “Frequently bought together” blocks), “no photo” placeholders, icons and images with the longer side under 200 px, and a photo the same site returned for your other products: that is what a logo or placeholder looks like.
If a site's photos are consistently small, it gets the “Small photos” label on the “Statistics” tab. Why that happens and how to fix it: checking a site.
Specifications and description
If specifications come from several sites, hover over a row: the “Source: …” tooltip shows which site the value came from. This tells you whose value is worth checking. You can correct values right in the table before saving.
The “Description” block shows what goes into the card: the full description if “Full description” is ticked in the “What to save into the product” panel and it was found, otherwise the short one if “Short description” is ticked. Both ticked: the block shows the full one, and “Save” writes both. Both unticked: no block, no description in the card.
What the run log shows
On the “Progress” tab, the log shows for each product the Google query and country, which sites AI picked, the loading of each page, parser setup for a new site and how many sites the AI check confirmed. Page rows are links: click one to open the page and check it yourself.
If a page failed to load, the “Page failed to load” row gives the reason:
| In the log | What it means | Charged |
|---|---|---|
| “the site did not respond within 90 seconds” | “Elbuz server”: the site hung or is very slow | no |
| “the browser did not load the page” | “My browser”: the tab was silent for 90 seconds, or the page didn't open on the second attempt either | no |
| “the page does not exist on the site (code 404)” or 410 | Google showed a page that no longer exists | no |
| “the site blocked access with bot protection (code …)” | the site returned 401, 403, 429 or 451, or the loading service refused because of the site's protection | no |
| “the site returned an empty page” | a response came, but with almost no content | no |
| “error on the site side (code 5xx)” | the site's server failed | no |
| “response code …” | another code | no |
- “Elbuz server”. Before writing the reason, the program has already retried through another loading service (except for 404 and 410), so running again rarely helps.
- “My browser”. A page the browser couldn't open has already been tried a second time; a page where the tab was silent for 90 seconds hasn't. If Google didn't let the browser past its link, the log shows “the site returned an empty page”.
If a site is regularly blocked or empty, add it to “Do not take content from these sites”; the “Statistics” tab shows this too. If processing broke off for some products, the log says “Products not processed: N. Run the search again to process them.” In “My browser” mode it also shows “Google asks to confirm you are not a robot…” and “The browser has not responded for 4 minutes, the search is stopped…”: what to do about it.
Which sites are useful: the “Statistics” tab
The “Statistics” tab sits between “Results” and “History”; the “Statistics” button on the right of the results filter bar leads there too. It collects data from all your Google searches for a period: “Last 30 days”, “Last 90 days” or “Last year”. Rows appear after the first such search.
1
Period switcher2
Summary tiles3
How site pages end up4
How much data a site gives- Period switcher — Last 30 days, Last 90 days or Last year; next to it, By pages / By rating.
- Summary tiles — sites, sites that need attention, pages loaded, share with data, products saved and spend.
- How site pages end up — per site: pages with data, empty or wrong, failed to load.
- How much data a site gives — average attributes and photos per site, and the share of pages with a description.
At the top are the “Sites”, “Need attention”, “Pages loaded”, “Pages with data”, “Products saved” and “Spent on pages” tiles. The “Need attention” tile counts sites with problems from the latest weekly self-check. Four charts answer four questions:
- “How site pages end up”: the 10 sites with the most pages, showing how many gave data, how many turned out empty or with a different product, and how many failed to load.
- “How much data a site gives”: the average number of specifications and photos and the share of pages with a description.
- “Spending and saved products”: the 10 most expensive sites, with costs as bars and saved products as a line. A tall bar with no line is a site that costs money but gives nothing.
- “All pages for the period”: the share of useful, empty and unloaded pages overall.
The “Sites” table
1
Low value label2
Status label, e.g. Small photos3
Rating column4
Per saved product column5
Three icons on the right- Low value — the site gives little useful data (here together with Not loading).
- A status label such as Small photos shows what is wrong with the site's results.
- Rating — overall score of the site.
- Per saved product — spend divided by products saved from the site.
- The three icons: check the site, do not take from the site, do not take photos from the site.
| Column | How it's calculated |
|---|---|
| “Rating” | from 0 to 100: page usefulness, photo size, number of specifications and presence of a description against the result requirements |
| “Pages” | how many times a site page was taken, including unloaded pages and cards through a driver |
| “With data” | the share of pages that passed the AI check and gave specifications or a description |
| “Attributes”, “Photos” | on average per page with data |
| “Photo, px” | the longer side of the first product photo, averaged over the site; “—” until the size has been measured |
| “With description” | the share of pages with data that had a description |
| “Saved” | how many products collected with the site's help you saved to cards |
| “Spent”, “Per saved product” | page loading only. Google queries and AI tokens are spent on the product as a whole and are not split by site |
Next to the site address is its status: “Good”, “Small photos”, “Placeholders”, “Few specs”, “No description”, “Not loading” or “Needs attention”. How statuses and the rating are calculated, how to set your own requirements and fix a site with the “Check the site” icon: checking sites and parsers.
The “Low value” label appears when a site has 5+ pages and under 30% of them gave data; banned sites don't get it. It's only a hint: blacklisting is your call. The “Excluded” and “No photos” labels mark sites already in “Do not take content from these sites” and “Do not take photos from these sites”.
On the right of the row are three icons with tooltips: check the site, ban or restore the site, ban or restore photos. Banning asks for confirmation, restoring doesn't. Both apply from the next search; results already found don't change. The change shows up right away in the fields on the “Run” tab.
Scenarios
Fill new cards without setting up sites
Task: new products have no description or specifications, and you have no ready set of sites.
What to turn on: tick the products in the catalog, choose “Google” in the “Where to search” step and check “Write into language”. If cards already have your photos, untick “Photos” in “What data to get”.
What you get: merged cards with a table in the card's language and a description in the program's own words wait on “Results”; “Save” puts them into the catalog.
What you do not get: the product name won't change unless “Product name” is on in the “What to save into the product” panel.
One site's photos have watermarks
Task: the results contain photos with another store's logo.
What to turn on: in the product panel, next to that site's group, click “Do not take photos” and confirm.
What you get: the site's photos disappear from all results not yet saved and won't be taken later; its specifications and description stay.
What you do not get: photos in already saved cards stay; remove them in the product card.
A site is blocked by protection or returns empty pages
Task: in the log, the same site keeps showing “blocked access with bot protection” or “returned an empty page”.
What to turn on: find the site on the “Statistics” tab (often labeled “Low value”) and click the ban icon, or the “Do not take from site” icon in the “Found on the site” block.
What you get: the site is dropped right in the Google results: no page loading or charge, and its “Sites per product” slot goes to other sites.
What you do not get: if the site is blocked only occasionally, the ban removes useful pages too: judge by “With data”, not by one error.
Take data only from a few trusted sites
Task: you trust three stores and don't want content from others.
What to turn on: list them in “Search only on these sites” and set “Sites per product, at most” to 3.
What you get: one Google query and up to three pages: up to €0.018 per product at the base rate, tokens excluded; “€ search maximum” estimates about €0.07 per product for this set. No parser setup needed.
What you do not get: no other countries and no barcode. A product Google didn't show on these sites stays not found, and the query is still charged.
Photos always from one store
Task: you want photos from the brand's official site or one store with clean photos.
What to turn on: put the site on the first line of “Take photos from these sites first”, or click “Use these” next to its photos in any result.
What you get: if the site is among the confirmed ones, its photos are ticked, even if few; otherwise the next site on the list, then the site with the most photos.
What you do not get: the search isn't limited to this site: if Google didn't show it, photos come from other sites.
The supplier's description is there, but specifications are empty
Task: the cards have a detailed supplier text but no specification table.
What to turn on: in the “Google” mode, tick “No search: card text only”.
What you get: a table built from the card text waits on “Results”; you pay only tokens and €0.05 per saved product.
What you do not get: new facts from sites, or changes to photos and description. A product with under 400 characters of text gets no table.
What to do when Google search gives no result
| What you see | Why | What to do |
|---|---|---|
| The search doesn't start, the window shows an amount | your balance is less than “€ search maximum” plus saving | narrow the filtered set or lower “Sites per product”, or top up your balance |
| The “Google asks for a verification (captcha)” banner | “My browser” mode: Google noticed many queries from your browser | pass the check in the parsing window — the search resumes by itself; if not, “Verification passed, continue” (more) |
| “The browser has not responded for 4 minutes, the search is stopped…” | the search window was closed or the Google check wasn't passed in time | keep the window open, pass the check if there is one, and run the search again |
| “Search not started: the Elbuz extension for Chrome is not responding…” | “My browser” is selected, but the extension isn't installed or is disabled | install or enable the extension, or choose “Elbuz server” |
| “Under 10% of products found… Continue?” | the autopause kicked in | search Google by hand for a few products that weren't found; if product pages show up, continue |
| “AI found no page for this product in the results” | the name is too generic or contains internal codes | add the brand and model to the name or fill in the barcode |
| Specifications in the site's language | the AI answer failed the check, so the table was merged without translation | reject the result and run the search for the product again, ticking it |
| No description, though the sites have text | AI failed twice to rewrite it without errors, and the program didn't keep a copy of the site | reject the result and run the search for the product again, or fill in the description with the text generator |
| Empty “Short description” | the sites have no short text, and AI doesn't write a description from its own knowledge | fill the field with the text generator |
| With “Search only on these sites”, the product isn't found | Google didn't show the product on these sites | clear the field or add sites |
| The same site in the log with “bot protection” as the reason | the site blocks loading | add it to “Do not take content from these sites” |
| “No search: card text only”: the product isn't found | the texts have under 400 characters, or the table failed the check | untick the checkbox and search via Google |
FAQ
Which should I choose: a site cascade or Google?
The cascade checks only your sites and stops at the first with content; its browser queries are free. Google finds sites, merges several into one card and writes a description; its queries and pages are paid on the Elbuz server, free in “My browser” mode. More: choosing a mode.
How do I know AI didn't make up a specification?
Every number in a value is checked against the source site, and you can see the source site in the tooltip when you hover over the specification. Text values aren't checked this way, so review them before saving.
Why aren't all the found photos ticked?
The first 10 photos of one site are ticked, so that photos from different stores with watermarks don't end up in the card. Tick the rest yourself or click “Use these” next to the site you need.
How do I restore a site I banned?
Delete its line in the field on the “Run” tab, or click the restore icon in the table on the “Statistics” tab. It applies from the next search.
How do I get rid of a banner that shows up in every product's photos?
Delete it with the cross on the thumbnail in the results. The address is remembered, and the photo won't be loaded for any product again. These photos are listed in the “Do not take these photos” field (more).
Why is the “Statistics” tab empty?
It counts only Google searches for the selected period. Collection from the card text and the site cascade don't go there.
Why are the costs in “Statistics” lower than what the search charged?
It shows page loading only. Google queries, AI tokens and saving are counted per product as a whole and aren't split by site. The run's costs for queries and pages are on the “History” tab.
Can I close the window during the search?
With “Elbuz server”, yes: the search runs on the server. Findings wait on “Results”, an unfinished run on a banner with “Continue”. With “My browser”, no: your browser opens Google and pages, and without the window the search stops within 4 minutes; unprocessed products wait for the next run.
Where to find it and who has access
- Menu
- Base catalog → “AI & content” → “Product content search”, step “Where to search” → “Google”
- Access right
- the “Product content search” window in “Group permissions”
- Prepare
- a filtered set in the catalog and balance for saving; for “Elbuz server”, also for Google queries and pages; for “My browser”, Chrome with the Elbuz extension
Product filtering, the “Progress” and “History” tabs, and the “What to save into the product” panel are shared with the other search modes. More: product content search.
Related topics
- For the overall picture of content search, product filtering and saving, see product content search.
- To have content for new products searched automatically after a catalog update, see background content search.
- If found specifications create duplicates in the attribute list, see matching specifications to the catalog.
- If a site gives small photos or few specifications, see checking sites and re-tuning parsers.
- If AI set up a site parser but the data is parsed wrong, see CSS and XPath selectors.
