Introducing AI-assisted catalog drafting

Cataloguing is the most expensive bottleneck in a modern auction pipeline. A single painting takes a specialist ~45 minutes of research, writing, and comparable-lookup before it's ready for the sale page. Multiply that by 400 lots per sale and you see why Q4 always slips.

We've been quietly beta-testing AI-assisted drafting with four Auction Rabbit clients since November. Today we're making it generally available.

How it works

1. Upload the photoset and any consignor-supplied notes.
2. The model pulls comparables from your own past sales (plus public benchmarks if you opt in) and generates a structured draft:
   • Lot description (in your house voice — trained from your last 1,000 catalogue entries)
   • Dimensions + medium extraction from EXIF + vision pass
   • Estimate band (low / high / reserve suggestion)
   • Condition report skeleton with flagged areas
   • Marketing long-copy and social snippets
3. Your specialist reviews and edits. Everything remains a draft until a human approves it.

Safety rails

• The model cannot publish a lot. Every draft is reviewed.
• All training uses your data, stays in your tenant, and is never shared with other houses.
• Provenance fields are never auto-filled — those require source documents.

Early numbers

Across 3,200 drafts reviewed so far, specialists kept ≥80% of the generated description in 71% of cases, accepted the suggested estimate band in 58%, and cut per-lot cataloguing time by 64%.