Understanding AI research confidence scores and data sources

Understanding AI Confidence Scores

When ListForge’s AI researches your items, it provides confidence scores that indicate how certain the system is about its findings. Understanding these scores helps you make better decisions about when to trust AI recommendations and when to review manually.

What Confidence Scores Mean

Confidence scores are expressed as percentages:

90-100% (High Confidence)

• AI found strong, consistent evidence
• Multiple data sources agree
• Direct product matches with abundant comparable sales
• Action: Quick review and approve

75-89% (Good Confidence)

• Solid evidence with minor gaps
• Primary identification likely correct
• Some fields may be inferred
• Action: Standard review, verify key details

50-74% (Moderate Confidence)

• Partial evidence found
• Identification is plausible but uncertain
• Limited comparable data
• Action: Careful manual review required

Below 50% (Low Confidence)

• AI is uncertain
• Item may be rare or unusual
• Conflicting or missing data
• Action: Manual research essential

Field-Level Confidence

Beyond overall confidence, you can see confidence for individual fields:

• Title: How certain is the AI about the product name?
• Brand: Was the brand clearly identified?
• Model: Is this the exact model or a guess?
• Condition: How well could condition be assessed from photos?
• Price: How reliable is the pricing data?

This helps you focus your review on uncertain fields rather than re-checking everything.

Data Sources

ListForge uses multiple data sources for research:

Visual Recognition

• Analyzes product photos using computer vision
• Identifies brands, categories, and product features
• Detects text and logos in images

Barcode Database

• UPC/EAN lookups for exact product identification
• Links to manufacturer specifications
• Highest accuracy source when available

Product Catalogs

• Cross-references against product databases
• Matches against known product listings
• Provides specifications and details

Marketplace Data

• Analyzes sold listings from eBay, Amazon, etc.
• Tracks current active listings for competition
• Provides pricing trends and history

Viewing Evidence

For each item, you can view the evidence supporting the AI’s conclusions:

1. Click on an item in your Review Queue
2. Look for the “Research Evidence” panel
3. See which sources contributed to each finding
4. Review the specific comparable sales used for pricing

Improving Confidence Scores

You can improve AI accuracy by:

• Better photos: Clear, well-lit, multiple angles
• Barcode scans: Always scan barcodes when available
• Brand visibility: Ensure brand markings are photographed
• Complete items: Photograph all components and packaging

Using Scores for Workflow

Smart sellers use confidence scores to prioritize:

• High confidence items: Batch approve quickly
• Medium confidence: Quick individual review
• Low confidence: Detailed attention required

This tiered approach maximizes efficiency while maintaining quality.