When you capture an item, ListForge's AI research pipeline processes it through four phases. Understanding what happens at each phase helps you get better results and troubleshoot issues.
Phase 1: Identification
The AI looks at your photos (and any hints you provided) and determines what the product is.
It uses multiple strategies simultaneously:
- Visual recognition — analyzes the images to identify the product type, brand, and model
- Text extraction — reads any visible text in your photos (labels, model numbers, brand names)
- Barcode lookup — if a UPC or EAN barcode is visible, looks it up in product databases
- Product Knowledge Base — searches ListForge's database of previously identified products for visual matches
The result is a product identification with a confidence level. High confidence means the AI is very sure about what the product is. Lower confidence means it made its best guess and you should verify.
The better your photos, the higher the confidence. A clear shot of a brand label or model number can make the difference between "high confidence" and "uncertain."
Phase 2: Field Research
Once the product is identified, the AI gathers the attributes needed for marketplace listings:
- Category assignment — maps the product to the correct category on each marketplace (eBay category IDs, Amazon browse nodes, etc.)
- Attribute extraction — determines key attributes like color, size, material, capacity, compatibility
- Condition assessment — evaluates condition from your photos and notes
- Marketplace requirements — checks which fields are required vs optional for each target marketplace
This phase ensures that when listings are generated, they meet each marketplace's requirements.
Phase 3: Pricing
This is where the AI analyzes the market to determine what your item is worth.
It searches for comparable sales — items similar to yours that have recently sold or are currently listed:
- Sold comparables — actual completed transactions. These are the most reliable data points because they represent what buyers actually paid.
- Active comparables — items currently for sale. These show what sellers are asking, but not necessarily what buyers will pay.
The AI analyzes these comps considering condition, completeness, and recency, then generates three pricing strategies:
| Strategy | Approach | Best For |
|---|---|---|
| Aggressive | Priced to sell quickly | Cash flow, clearing inventory, commodity items |
| Balanced | Market-rate pricing | Most items, optimal revenue over time |
| Premium | Higher than typical, willing to wait | Rare items, excellent condition, patient sellers |
Each strategy includes an estimated price, expected days to sell, and the reasoning behind the recommendation.
Phase 4: Listing Generation
Finally, the AI generates marketplace-specific listings from all the data gathered:
- Titles optimized for each marketplace's search algorithm
- Descriptions that highlight key features and condition
- Item specifics / attributes filled according to marketplace requirements
- Shipping recommendations based on item dimensions and weight
One set of research creates listings for every marketplace you have connected.
How Long Does It Take?
Typical research takes 2–5 minutes from submission to completion. The time varies based on:
- How easily the AI can identify the product
- How many comparable sales are available
- Current system load
When Things Go Wrong
Research can fail or produce poor results for several reasons:
- Bad photos — blurry, dark, or cluttered images make identification difficult
- Very obscure items — if there aren't comparable sales, pricing accuracy drops
- Ambiguous products — items that look like many different things (generic white boxes, common shapes)
If research fails, check the Why Did My Research Fail? troubleshooting article for specific guidance on improving results.