eBay now drafts a whole listing from a photo. If you run a marketplace, decide which fields a seller must confirm
Take a few photos, and eBay's app now identifies the item and fills in the listing for you to review. Sellers will love it, and buyers of every other marketplace will start expecting it. If you run one, the work moves from the listing form to the review screen, and to one question: which details does a person have to stand behind?

On November 11, 2026, eBay expanded its camera-first listing flow: sellers photograph an item, eBay identifies it and fills in the details, and nothing goes live until the seller reviews and publishes. If you run a marketplace and plan something similar, decide field by field what AI may draft, what the seller must confirm explicitly, and what AI should never fill, such as condition, authenticity and defects, and show clearly which fields came from the model.
What eBay launched
On November 11, eBay announced an upgraded camera-first listing flow in its app. A seller takes photos of an item, eBay identifies it and fills in the listing details, and the seller reviews the draft and taps publish.
eBay is careful about the last step. "Nothing goes live without your approval," it says. Sellers can review and edit everything before the listing appears. eBay also says more than 10 million unique sellers have used its AI features and that they create more than 800,000 new listings a day. The new flow is available to consumer sellers in Australia, France, Germany, Italy, the UK and the US.
It builds on a simplified mobile listing flow eBay introduced last year, and eBay describes the shift plainly: the work moves "from setup to review."
That's eBay's news. The lesson is for everyone else who runs a marketplace, whether it sells used equipment between businesses, secondhand furniture in one city or collectibles to a few thousand enthusiasts.
The work moves to review
On a traditional marketplace, the listing form is where the seller does the work: type the title, pick the category, choose the condition, write a description. Bad listings come from tired sellers skipping fields.
When AI fills the form, the review screen becomes the product. Nearly everything arrives pre-filled and plausible. The risk changes from "the seller left it blank" to "the seller accepted something that isn't true of this particular item." The model can recognize a model of camera from a photo. It can't know that this one has a sticky shutter.
Buyers will soon expect the same speed everywhere, and sellers on smaller or specialist marketplaces will ask for it. If you run one, you'll be under pressure to offer photo-to-listing too. Before you build it or switch it on in your platform, decide what a person has to stand behind.
Decision guide
Who vouches for each field
| Consideration | AI may draft | Seller must confirm or write |
|---|---|---|
| Item identity | Category, brand and model suggested from the photo | Seller confirms it's the exact model, with any variant |
| Description | A first draft from the identified item | Seller edits anything that isn't true of this one |
| Condition | Never filled by AI | Seller chooses, from defined grades |
| Defects and wear | May prompt the seller to check | Seller states them, in their words |
| Authenticity or provenance | Never asserted by AI | Seller states it; marketplace rules decide what proof is needed |
When AI fills the form, the review screen becomes the product.
Decide field by field
Go through your listing form and put every field in one of three groups.
AI may draft, seller may accept. Category, brand and model suggestions, a title, measurements taken from a known model's specifications, a first draft of the description. These are useful, low-risk and easy to correct.
AI may suggest, seller must confirm explicitly. The exact model or variant, color names, included accessories, anything that changes price. The seller should have to tap something that means "yes, this is right," not just scroll past.
AI never fills. Condition, defects, wear, authenticity, provenance, and any claim your marketplace rules treat as the seller's promise. These are the fields disputes are about. A model's guess here helps nobody, and it muddies who is responsible when a buyer complains.
Where each field lands depends on your category. On a marketplace for used textbooks, edition matters and condition is simple. On one for vintage furniture or art, provenance and condition are the whole listing. Make the decision per category, not once for the platform.
Write the result down as a short table your team can read: field, group, and why. It sounds like paperwork, but it's what product, trust and safety, and support will each point to later. When a seller asks why the app won't let them publish without choosing a condition, support needs a reason that's better than "the model isn't allowed to." When a buyer disputes a listing, trust and safety needs to know which fields were the seller's promise.
Pricing deserves its own line. Many photo-to-listing tools suggest a price from similar sold items. That's helpful, and it's also a nudge sellers tend to accept. Show where the suggestion came from and make it easy to change, so a seller with a rare variant doesn't underprice it because the model saw the common one.
Design the review screen
Once you know who vouches for what, the review screen follows.
Show what the model filled. A small marker on every drafted field is enough. Sellers should never have to guess whether they typed something. When a seller edits a drafted field, drop the marker, so the screen always shows what's still the model's wording and what's now theirs.
Stop at the risky fields. Don't let publish work until condition, defects and any authenticity question have an explicit answer from the seller, chosen from defined options where you can.
Put the photo next to the claim. If the draft says "includes original box," show the photo of the box beside it, or ask for one.
Keep a record. Store what the model drafted, what the seller changed and when they confirmed each risky field. When a buyer opens a dispute, that history tells you whether the problem was the model, the seller or the listing rules.
The same principle shows up wherever AI now fills in commerce content: our note on why AI menu photos need the restaurant's sign-off makes the case that the business, not the model, signs off on what customers see.
How the work moves
A review screen that earns the publish button
- Mark the drafted fields
Show clearly which details came from the model.
- Stop at the risky ones
Condition, defects and authenticity need an explicit answer before publish.
- Show the photo beside the claim
Put each key detail next to the image it came from.
- Record who confirmed
Keep what the AI drafted and what the seller changed, for disputes.
Watch what happens after launch
AI drafting will change your listing data in ways that don't show up on day one. Watch three things for the first couple of months: how often sellers edit the drafted fields (if almost never, they may not be reading), how dispute reasons shift, and whether categories fill up with near-identical descriptions that make search worse.
Also check the photos. A flow that starts from the camera tends to produce listings with fewer, more similar images, because the seller took them for the model rather than for the buyer. If your buyers rely on close-ups of labels, serial numbers or damage, ask for those specifically on the review screen.
Pay special attention to new sellers. A first-time seller with a fully pre-filled form may not know what they're promising. Our note that your first suppliers need a person, not a form applies here too: early sellers benefit from a human explaining what condition grades mean before a model makes listing feel effortless.
Where we help
We design and build marketplace listing flows, including where AI drafting fits, what sellers must confirm and how disputes get resolved afterwards. If you're weighing a photo-to-listing feature from your platform vendor against building your own, the field-by-field decision above is the first thing to settle either way. That's our marketplace work. Talk to us.
Related work: Paddle8
We build alongside the people doing the work.
Bring us the problem you’re thinking about.