A good AI prompt for an auction lot description gives the model every photo of the lot, asks it to describe only what the photos show, sets the title length and layout your platform wants, and tells it to write "unknown" instead of guessing a maker, metal or date. Gavelist, an auction cataloging tool, runs that kind of instruction on every lot of a sale at $0.15 a lot: one Google Gemini model with one prompt that carries a checklist for each of 18 auction categories. Below are seven copy-ready prompts for ChatGPT, Claude or Gemini, and the point where a prompt stops being the practical way to do it.
By Ben Cope, founder of Gavelist. The prompts below work in any chat assistant that accepts photos; none of them needs Gavelist.
Before you prompt: what makes a lot description good
Bidders online cannot pick the item up. A useful description answers what they would check in person:
- What it is: maker, model or pattern, material, size.
- What is marked: stamps, signatures, labels, serial plates, written out exactly as they appear.
- Condition: chips, cracks, wear, missing parts, "untested".
- What is included: pieces, accessories, box.
One way a chat assistant goes wrong is stating something the photo does not show, such as a maker it assumed or "sterling" without a visible hallmark. Every prompt below tells it not to.
Prompt 1: the full lot description
Use this one for most lots. Attach every photo of the one lot first.
You are writing an auction catalog entry for ONE lot. All attached photos
show the same lot.
Write:
1. TITLE: under 80 characters. Start with maker or brand if one is visible,
then the item type, then the key detail (pattern, model, size, era if marked).
2. DESCRIPTION: 2 to 4 plain sentences a bidder can act on. Include material,
construction, style, and any marks or labels, quoted exactly as they appear.
3. CONDITION: list visible damage, wear, or missing parts. If the photos do
not show a side or detail, say "not shown".
4. MEASUREMENTS: write "[measure]" as a placeholder. Do not estimate sizes.
Rules:
- Describe only what the photos show. If you cannot read a mark, say so.
- Do not name a metal (gold, sterling) unless a hallmark or stamp is legible.
Otherwise write "gold-tone" or "silver-tone".
- Do not name a gemstone unless a certificate is shown. Otherwise "clear stone".
- Do not guess a maker, artist, or date. Write "unknown" instead.
- No sales language ("stunning", "rare find", "must-have").
Prompt 2: condition notes only
For lots you will title yourself but want a careful second look at.
Look at every attached photo of this one item. List each visible condition
issue as a short bullet: location, type (chip, crack, scratch, stain, loss,
repair, rust), and size relative to the item (small, medium, large).
Then list anything you cannot assess from these photos (for example: "base
not shown", "interior not shown", "working condition unknown").
Do not describe the item itself and do not grade it overall.
Prompt 3: read the marks, do not interpret them
A guessed mark can put the wrong maker in your title.
Transcribe every mark, stamp, signature, label, or serial number visible in
these photos, exactly as written, one per line, with which photo it appears in.
If part of a mark is unreadable, show the unreadable part as [?].
Do not tell me what the mark means or who made the item.
Look the transcription up yourself, or in a reference you trust, before the maker goes into the title.
Prompt 4: write it in my house style
Paste three of your own past descriptions so the assistant copies your voice instead of its own.
Here are three descriptions from my past catalogs:
[paste example 1]
[paste example 2]
[paste example 3]
Write the description for the attached lot in the same style: same length,
same order of details, same way of writing condition. Follow the same rules:
describe only what the photos show, quote marks exactly, and write "unknown"
instead of guessing a maker or date.
Prompt 5: titles for a specific platform
Rewrite this title for [platform name]. Keep it under [N] characters.
Order: brand or maker (if known), item type, model or pattern, key detail.
No abbreviations a buyer would not search for. No punctuation except commas.
Title: [paste]
Check your platform's own help pages for the current title length and fields. Limits differ by platform and change.
Prompt 6: turn your notes into descriptions
If you walked the house with a voice memo or a notepad, a text-only prompt works without photos.
Below are my rough notes, one lot per line, in the form "lot number: notes".
For each line, write a title (under 80 characters) and a 1 to 2 sentence
description using only the facts in my notes. Do not add details I did not
write. Return a table with columns: Lot, Title, Description.
[paste notes]
Prompt 7: check a description against the photos
Use this as a last pass on your higher-value lots.
Here is a draft description and the photos of the lot. List every claim in
the description that the photos do not support, and every visible detail
(mark, damage, part) that the description leaves out. Do not rewrite the
description.
Draft: [paste]
Where a chat prompt stops working
A chat prompt is a reasonable way to describe a few lots. It gets harder as the sale grows:
- One lot per conversation. You attach the right photos to the right lot each time. Mixing two lots' photos in one chat can produce one blended description.
- Copying and pasting. Every title and description has to be moved into a spreadsheet or your platform's listing screen by hand.
- No import file. Your platform's import still needs its exact columns and photo names.
- Drift. Over a long session, the format and tone of the answers can wander unless you paste the rules again.
- No record. Nothing tracks which lots are done, which need review, or what you changed.
When to use a tool instead
The rules in these prompts are close to the rules Gavelist's own prompt follows, applied to a whole sale at once. Gavelist runs one Google Gemini model with one prompt that carries a checklist for each of 18 auction categories, reading every photo in a lot together. Gavelist flags lots it cannot identify for review instead of guessing, and it names metals and stones only from a readable hallmark or certification. The differences from a chat window are the parts around the prompt:
- Whole sales, not single lots. You upload the sale's photos once. Gavelist keeps them in camera-roll order, groups them into lots automatically when the filenames carry lot numbers, and otherwise gives you a conveyor-belt screen to split them.
- Measured describing time. Most sales of up to 500 lots finish describing in under 10 minutes once describing starts.
- Your style. Every Gavelist plan includes voice profile settings for your house style, so you are not pasting examples into each chat.
- A file for your platform. Gavelist exports ready-made files for HiBid (through Auction Flex 360), LiveAuctioneers, Proxibid, BidWrangler, AuctionNinja and Auction Flex 360, an import workbook for Auctioneer Software, an eBay drafts CSV, and a generic CSV for AuctionZip, AuctionMethod, Wavebid and any other spreadsheet-import platform.
- A review step. Gavelist flags lots it cannot identify, and you check every lot before export. According to Gavelist production data, 88% of lots across Gavelist clients shipped with no client edit to the AI title or description (96% including Gavelist's largest client), through September 2026.
Gavelist costs $0.15 a lot on Pay as You Go, or $79 a month for 1,000 lots, with 0% of auction sales. A 300-lot sale is $45.
Stick with prompts if you list a handful of items a week, one at a time, and type them into your platform anyway. Use a tool if you photograph whole houses or run sales of hundreds of lots and need an import file at the end.
Gavelist's free trial is 100 lots, no card, 30 days from your first describe. Start the free trial, or see how to write auction descriptions with AI and auction lot description examples.
FAQ
What is a good ChatGPT prompt for auction item descriptions? Attach every photo of one lot, ask for a title under 80 characters, a 2 to 4 sentence description, and a condition list, and tell it to describe only what the photos show, quote marks exactly, and write "unknown" instead of guessing a maker, metal or date. Prompt 1 above is a copy-ready version.
Can ChatGPT write auction descriptions from photos? Yes, if the app you use accepts photos. Keep it to one lot per conversation, give it rules against guessing, and check marks and measurements yourself.
How do I stop AI from making up details in a lot description? Tell it in the prompt: describe only what the photos show, write "unknown" for anything not visible, and never name a metal or stone without a legible hallmark or certificate. Then run a check pass like Prompt 7 on your higher-value lots.
Should I let AI estimate the value of auction lots? Use any AI value as a starting point, not a price. Check it against what the same item sold for, ideally at your own past sales. Gavelist gives every lot a value tier and a low-high estimate, and its optional comps add-on ($0.15 a lot) shows the listings behind the range, with eBay results labeled as asking prices.
How is Gavelist different from using a ChatGPT prompt? Gavelist runs one prompt with a checklist for each of 18 auction categories across a whole sale, reads every photo in each lot together, flags lots it cannot identify, and builds the import file for your platform. It costs $0.15 a lot with 0% of sales.
Sources
- Gavelist, "Pricing" (Pay as You Go $0.15 a lot, Auctioneer $79 for 1,000 lots, free trial), gavelist.com/pricing, November 2026.
- Gavelist, "llms.txt" (one Google Gemini model, one prompt, 18 category checklists; export list), gavelist.com/llms.txt, read November 5, 2026.
- Gavelist production data: describing time for sales of up to 500 lots (August 16 to October 15, 2026); export edits through September 22, 2026.