Here’s the part of this work nobody warns you about: the selling is the easy part. It’s Sunday night, the sale is Saturday, and you’re staring down 340 lots that each need a title and a couple of honest lines. That’s what eats the evening. Not the calling, not the crowd — the typing, one dresser at a time, until you’re doing it tired and sloppy at midnight and hating a job you actually love.
A free AI chatbot won’t call the sale for you, but it will turn “oak dresser, mirror, four drawers, one pull missing” into a clean paragraph in a few seconds. You read it, fix what’s wrong, move on. That’s the pitch — not magic, just a fast set of hands that never gets tired and never squints at your handwriting.
But there’s a catch, and it’s the whole reason to read the rest of this: hand one of these tools a photo and it will mislabel your merchandise with total confidence. It’ll look at a plain Pyrex bowl and announce, in a firm and pleasant voice, that it’s Fire-King. It is never unsure. So the skill you actually need isn’t prompting — it’s knowing when to believe it and when to overrule it. Here’s how to do both.
What the free chatbots are good at — and what they’re not
As of 2026, both ChatGPT and Claude have free tiers you can use in a browser or on your phone. Either works for this. You type in what you know, attach a photo if you like, and it drafts. Nothing to install. If you can send a text message, you can do this.
One thing up front, because everything below rides on it: the words are the machine’s strong suit. The photo is not. Lean on your own notes and treat anything it claims to “see” in a picture as a guess waiting to be checked. It writes well. It does not know things.
The basic move: tell it what you see
The single biggest mistake is treating the chatbot like a search box and typing two words. Give it the facts you already know as the auctioneer — the ones it can’t get from a blurry photo — and it’ll do the writing part well. Here’s a starter prompt you can keep in your notes app and reuse all day:
Write a short auction lot title and a 2–3 sentence description for this item. Item: solid oak chest of drawers, four drawers, beveled mirror on a swing frame, circa 1910s, one wooden drawer pull missing, surface has honest wear and a water ring on top. Plain factual tone, no hype, don’t guess at anything I didn’t tell you. Give me a title under 12 words and a description under 60 words.
Notice what’s in there: the material, the count, the era, and the flaws. And notice the two instructions that matter most — “don’t guess at anything I didn’t tell you” and a length limit. Without those, these tools pad. They’ll call your dresser “a stunning museum-quality masterpiece of American craftsmanship,” which is exactly the kind of talk your bidders have learned to scroll right past.
Two more prompts worth stealing
For a firearm or anything where you want the facts and nothing else:
Write a factual auction lot description. Item: Winchester Model 94 lever-action, .30-30, serial dates it to the 1960s, bore is bright, some bluing wear on the receiver edges, stock has a few honest dings. State only what I gave you. Do not estimate value, do not comment on rarity, do not add safety or legal language. Under 50 words.
For a box lot or a “smalls” grouping where you’re trying to make a pile of stuff sound worth bidding on without lying:
Write a one-sentence lot title and a short description for a mixed box lot. Contents: assorted kitchen glassware, a few pieces of Pyrex, a stack of vintage cookbooks, two enamel pots, one chipped bowl. Group it appealingly but honestly — mention the chip. Under 40 words.
The pattern never changes: what it is, roughly when, what shape it’s in, what’s wrong with it, how long you want the answer. Fill those five blanks and the writing takes care of itself.
Good draft vs. bad draft
Here’s what you’re steering toward, and away from. Same dresser, two outputs.
Bad (what you get from a lazy two-word prompt):
“Absolutely stunning antique oak dresser — a rare and breathtaking piece of history! This gorgeous Victorian masterpiece features exquisite craftsmanship and a beautiful mirror. A must-have showstopper for any collector. Don’t miss this once-in-a-lifetime opportunity!”
It’s all air. “Rare” and “Victorian” are guesses the machine made up — you never said Victorian, and a 1910s dresser isn’t Victorian. Worse, it buried the missing pull and the water ring, which means an unhappy bidder standing at your pickup table on Saturday.
Good (from the fuller prompt above):
“Oak Chest of Drawers with Beveled Mirror, c. 1910s. Solid oak, four drawers, swing mirror on a turned frame. One wooden drawer pull is missing and the top has a water ring; overall honest wear consistent with age. Sturdy and functional.”
Shorter, truer, and it tells the bidder exactly what they’re walking up to. That second one sells better over a year of sales, because the people who buy from you come back.
The part nobody can skip: you check every one
Here’s the honest warning, and it’s the reason this article exists. Hand a free chatbot a photo of a Weller vase and it may tell you it’s Roseville — not with a shrug, but flatly, like it just read the bottom. It’ll misread a maker’s mark, invent a pattern name, assign a date it cannot possibly know, and call a reproduction “period.” It is not lying on purpose. It genuinely can’t tell, and it’s built to sound sure of itself either way.
That’s the trap. A wrong answer that sounds wrong, you catch. A wrong answer delivered with a straight face is the one that ends up in your catalog under your name. So the rule is simple and it never bends: the AI drafts, you decide. Anything it claims that you didn’t feed it — a brand, a date, a “rare,” a maker — is a guess until your own eyes confirm it. A few things it flatly cannot do, no matter how sharp the photo:
- Judge authenticity. Real vs. repro, signed vs. stamped-later, married pieces — that’s your call, in person.
- Grade condition or spot hidden damage. It can’t see the crack under the glaze or feel the loose joint. Don’t let it write “excellent condition.” You write the condition.
- Price anything. Especially thin, rare, or regional items with no comps. A made-up estimate is worse than none. “No estimate, opening bid $25” beats a confident fake number every single time.
And one bright line on photos, since people ask: cleaning up lighting, straightening a crooked shot, or dropping in a plain background is fine. Editing out a chip, a crack, or a stain is fraud. It’ll cost you the bidder and eventually the reputation you spent thirty years building. Show the damage. Describe the damage. That’s the whole game.
Where “I don’t know” beats a confident guess
You can catch a confident wrong answer on lot 4. On lot 240, tired, at midnight, you won’t. That gap — the one between what the machine will admit and what it just charges past — is the exact thing we built Hammerwerks to close.
Run a lot’s photos through our cataloging and you get the same drafts a chatbot gives you — a title, a description, a category, a condition note, a value range — but every one of them comes back with a confidence score, so you can see at a glance where it’s sure and where it’s reaching. A second AI reads the first one’s work before you ever see it, hunting for exactly the kind of straight-faced nonsense above. And when it hits a coin, a piece of jewelry, a painting, or anything in a market too thin to call honestly, it tells you “I don’t know” instead of inventing a number. A blank you have to fill in yourself is worth more than a fake you didn’t notice. That candor is the point — it’s what keeps your credibility with the people bidding.
On top of the writing, it runs the things you can’t eyeball: it reads a VIN, an ISBN, a maker’s mark, a UPC and looks them up, and it checks the item against NHTSA and CPSC recall lists, so a recalled crib or a bad space heater doesn’t go up on your block. None of it takes the decision away from you — you still approve every lot before it lands in the catalog, same rule as the free tools. The difference is it drafts where the rest of your sale already lives, and it’s honest about what it doesn’t know. The AI features run on credits (Growth includes 500 a month, Scale 3,000); if you’d rather type a lot yourself, manual cataloging is unlimited on every plan.
Or run the whole thing from ChatGPT
If you already live in ChatGPT, you don’t have to copy-paste between apps at all. Hammerwerks connects to it directly, so you can tell ChatGPT to pull up a lot, draft a description, check a recall, or move an unsold item to the next sale — and it does it inside your account. Same approvals, same bidder list, no new screen to learn. It’s the how-to on this page, minus the tab-switching.
A workflow that actually saves you time
- Set the item down. Look it over the way you always do — material, age, flaws.
- Draft the description — from a free chatbot with your reusable prompt, or from a photo inside Hammerwerks.
- Read it. Cut the hype, overrule any claim you didn’t make, make sure every flaw survived. Trust the high-confidence lines faster; slow down on the low ones.
- Approve it. Next lot.
Do that and you’ll knock out a description in under a minute instead of five, and they’ll read cleaner and more consistent than when you were typing them tired at 11 p.m. You’re not handing the machine your judgment. You’re handing it the typing — and keeping the one thing that’s actually yours.
That’s the whole thing — a fast draft, your knowledge, and a hard rule that you approve every line. You can start tonight with nothing but your phone and a free chatbot; the steps above stand on their own and we mean that. And when you want the drafting to happen where your sale already lives — with a confidence score on every line and an AI that will say “I don’t know” instead of guessing — that’s what we’re here for.