I built an AI tool that turns photos into eBay/Etsy listings in seconds

Hahaha they actually copied the whole thing, couldn't even come up with a new name!

I have the documentation how this was built using Claude Code, only using prompt engineering. It was originally built for my wife from the ground up, it's not a novel concept. BTW Underpriced AI is actually copyrighted.

Thanks for the thoughtful questions - these are exactly the right things to dig into.

Pricing accuracy: We actually show ranges, not point estimates, for exactly the reason you mentioned. Claude provides initial identification and a value range, but we validate against live eBay data (both active listings and sold comps). When the eBay market data differs significantly from the AI estimate (>40%), we automatically adjust to market-based pricing. So for that vintage item where Claude says "$50" but similar items are actually selling for "$150", users see the real market price.

Category performance: You nailed it. Electronics and branded goods are easy mode - model numbers basically solve the problem. The interesting challenge is exactly what you'd expect: handmade items, regional pottery, unsigned art, vintage fashion. We've invested heavily in the prompt engineering here - the system now has category-specific identification guides (hallmarks for silver, backstamps for ceramics, union labels for dating vintage clothing, etc.). Confidence scores reflect this too - we'll show 0.9+ for a clearly marked Wedgwood piece but 0.6 for "probably 1970s studio pottery."

Feedback loops: This is the moat we're building. We capture the listing price users choose, and when they connect their eBay account, we can see actual sale prices. Still early days on the data volume, but the flywheel is there. The goal is exactly what you're describing - proprietary pricing data that improves over time.

Appreciate the good wishes on launch!

Great question on edge cases - that's honestly where the product gets most interesting.

When comparable data is scarce, a few things happen: the confidence score drops (we'll show 60-70% instead of 90%+), and the market saturation analysis will explicitly surface "low active listings" so users know they're in uncharted territory. The AI also provides a quick tip for these situations - often suggesting what details to photograph or research further (e.g., "check for maker's marks on the base" or "this pattern was only produced 1965-1970").

And yes - everything is fully editable. The AI suggestions are a starting point, not a mandate. Experienced resellers often know their niche better than any model, so they can adjust title, description, price, and category before listing. We've found the sweet spot is handling the 80% of grunt work (writing descriptions, keyword optimization, category selection) while letting sellers apply their expertise where it matters.

Thanks for the encouragement on cross-posting - that's where we think the real time savings stack up!

Thanks, appreciate the kind words!