Web Research
Web Research lets merchi.ai search the web for information about your products before generating content. It finds specifications, documents and ecommerce-grade images from manufacturer and retailer pages, then uses those findings to produce richer, more accurate content — all without you having to supply source material manually.
Web Research is an Enterprise feature. If it is not available on your plan, contact your account manager.
What It Does
When Web Research is enabled for a product, two things can happen before content is generated:
Enrich product data — merchi.ai searches public web sources for the product’s specifications, descriptions and supporting documents (datasheets, user guides). Any facts it finds are shown on the Web Research tab inside the product detail view, each with a link to the page it came from. These facts feed directly into content generation.
Find product images — merchi.ai searches for ecommerce-grade images of the product from manufacturer and retailer pages. It checks each image for quality (resolution, sharpness, watermarks) and match confidence before suggesting it. Approved images appear in the product’s image gallery.
Both options add +1 credit per product on top of the standard generation credit.
Enabling Web Research on Upload
When you import products via spreadsheet, a Web Research step appears before the file import dialog. You have three choices:
- Tick one or both options — “Enrich product data” and/or “Find product images” — then click Next. The selected research runs on every eligible product in your spreadsheet.
- Click Skip — go straight to the file import with no web research. The standard generation credit applies.
- Click Cancel — close the importer entirely.
Which Products Are Eligible
Web Research needs enough information to find the product on the web. A product is eligible when it has:
- A brand name, and
- At least one structured identifier: MPN, GTIN, or EAN
Products without these will be flagged with a warning before import begins. They are still imported and processed using any images or data you provided — they just don’t receive web research enrichment.
Additional columns improve accuracy. Title, category, and any other product data you include are not required for eligibility, but they are passed to the research task as context. The more detail you supply, the more precisely web research can identify the right product and the more relevant its findings will be — particularly for products with common MPNs or brand names that appear across multiple product lines.
Default Selections
Your administrator can set default selections for the two checkboxes (Settings → Web Research). If defaults are set, the checkboxes open pre-ticked. You can change them before clicking Next.
Identifier-Only Products
If a product has a brand and identifier but no image or document, Web Research can process it with no upload at all. When “Find product images” is enabled, merchi.ai searches for suitable images before generating content. If a match is found, it is used for generation.
What happens next depends on the image mode your administrator has configured:
- Use matched images — any image that meets the quality bar is used for generation immediately.
- Wait for approval — merchi.ai parks the product as “Awaiting images” until you approve at least one image in the Discovered Images tab. You can then generate content from the grid using Generate content.
- Never use web images — generation runs on identifier data and any datasheet found, with no image input. Discovered images still appear in the gallery for review.
Viewing Research Findings
Open any product and go to the Web Research tab. This shows:
- Each fact found, grouped by category (specifications, additional information)
- The source page each fact came from, with a link
- The date and time research was last run
If no research has been run for the product, the tab is empty.
Refreshing Research
To re-run web research and regenerate content at the same time, use the Reprompt option and tick Refresh web research in the reprompt dialog. This costs 2 credits (1 generation + 1 research).
Discovered Images
Open a product and go to the Discovered Images tab to see images web research has suggested.
Each suggestion shows:
- A preview of the image
- The source page and domain
- Image dimensions and quality assessment (Sharp, Acceptable, or Blurry)
- Match confidence — how confident the system is that this is the right product
- Any quality flags: watermark, text overlay, or diagram
Approving and Rejecting Images
- Approve — the image is downloaded, processed and added to the product’s image gallery.
- Reject — the suggestion is dismissed.
Once approved, an image appears in the main image gallery alongside any images you uploaded. You can reorder or delete it there in the same way as uploaded images.
Auto-approved images (those that met the automatic quality threshold your administrator configured) are already in the gallery when you open the product.
Source Confidence in the Product Detail View
When content has been generated using web research, fields that drew on research findings show a confidence badge with a globe icon. Click the badge to see:
- The confidence score and level
- Which sources contributed (e.g. S1, S2)
- The specific reasons for the score
Sources Disagree
If web research found a different value for a field than the data you provided (for example, a different product title), the field shows a Sources disagree notice. This lets you choose which value to keep:
- Open the product detail view
- The conflicting field shows the current value (your data) and the web research alternative, each with its source
- Select the value you want and click Use selected value — or click Ignore to keep your current value and dismiss the notice
- Save the product
Resolving a conflict clears the “Sources disagree” flag for that field.
Grid Filters
When Web Research is active, two additional filters appear in the product grid:
- Sources disagree — shows only products where web research found a value that differs from your data. Use this to work through conflicts in bulk.
- Awaiting images — shows only products parked because no image has been approved yet. Select them all and click Generate content to process them once you have approved images.
Reprompting with Web Research
When you regenerate content for a product (via Reprompt), the previous research findings are reused by default — no extra credit is charged.
To run fresh research before regenerating, tick Refresh web research in the reprompt dialog. This costs 1 additional credit.
Bulk Refresh
From the product grid, you can select multiple products and click Refresh web research to re-run research on all of them at once. This triggers fresh data and image research and then regenerates content. Each product costs the standard 2 credits (1 generation + 1 research).
Credits
| Action | Credits |
|---|---|
| Standard generation (no web research) | 1 |
| Generation + web research | 2 |
| Reprompt (reuses stored research) | 1 |
| Reprompt + refresh research | 2 |
| Bulk refresh | 2 per product |
Credits are checked before processing begins. If a product cannot be processed due to insufficient credits, it is skipped and you will see a notification.
Frequently Asked Questions
Why is the Web Research step not appearing when I upload? Either the feature is not enabled on your plan, or your administrator has not switched it on for your account. Contact your account manager if you expect to have access.
Can I use web research on products I have already processed? Yes. Use Reprompt with Refresh web research ticked to run fresh research and regenerate content in one step.
Will web research overwrite the data I provided? No. Your data always takes priority. Where web research disagrees with a value you supplied, it is flagged as a conflict for you to resolve rather than applied automatically.
Why are some products skipped during import? Products without a brand and at least one identifier (MPN, GTIN or EAN) cannot be researched on the web. They are still processed using the images and data you provided.
Why is the Discovered Images tab empty? Image research may not have run for this product, or no suitable images were found. You can trigger a fresh image search using Reprompt with Refresh web research ticked.
What happens to discovered images if I re-run research? Existing auto-approved images remain in the gallery. New suggestions appear in the Discovered Images tab alongside any existing pending ones.