A new pair of running shoes reaches a retail site with its paperwork already done. Brand, model, colourway, upper material, a size grid, a barcode, a category assignment, and a dozen filterable attributes underneath. Somebody was paid to enter all of that, and an entire enterprise software category exists to keep it consistent across every channel the brand sells through.
The same shoes, worn twice and photographed on a kitchen floor, reach a resale marketplace with none of it. They arrive as four photos and whatever the seller could be bothered to type into a box.
That gap is the most underrated constraint in the secondhand economy. Not demand. Demand for used goods has been healthy for years and keeps getting healthier. The constraint is that resale hands the data entry to the least equipped participant in the whole chain: one person, with a phone, at nine on a Sunday night.
The Data Entry Nobody Budgeted For
Retail solved this problem by paying for it. Product information management is an unglamorous line item, but every brand of any size carries it, because search, filters, recommendations, comparison engines and now shopping assistants all run on the same structured fields, and those fields do not populate themselves.
Resale inverted the arrangement without anyone deciding to. Every used item is effectively its own product record, created once, by an amateur, for free. There is no catalogue to inherit from. The drill in the garage has a model number rubbed half off the housing. The jacket has a brand tag, a fabric composition tag, and a small hole in the lining that has to be disclosed. Somebody has to look at all of that and turn it into text.
Most people quit long before they finish. Ask anyone who resells casually and they will describe the same thing: a corner of a room with things in it that are genuinely worth money and have been sitting there since spring. Not because the seller doesn’t know where to sell them. Because listing item number seven, on a Sunday, is worse than owning item number seven.
Why the Listing, Not the Item, Decides the Sale
Here’s what makes that abandonment expensive rather than merely untidy.
eBay’s own seller guidance is direct about item specifics: they are what lets buyers narrow a search, and a listing that leaves a field blank simply will not appear when a buyer filters on it. Etsy’s tags and attributes behave the same way. The marketplace cannot see the object. It sees the fields.
So a listing that reads “black jacket size M good condition” is not a slightly worse listing than one that reads “Barbour Bedale waxed cotton jacket, medium, olive, 1990s, some wear at cuffs.” It is a listing that a large share of buyers will never be shown at all. The item was fine. The record was empty.
It is the same failure mode TechBullion has covered at the brand end, where AI shopping tools keep hitting a wall built out of inconsistent product data. Resale has the identical problem at the opposite end of the scale, spread across millions of people who were never going to fill in an attribute schema for a $22 sweater.
And nearly all the advice aimed at these sellers is about sourcing. Where to find inventory, what to buy, which thrift stores restock on Tuesdays. Very little of it is about the twenty minutes after the photo, which is where most of the value actually leaks out.
Reading the Product Off the Photograph
The interesting technical development of the last two years is not that models can write product copy. They could do that in 2023, and the results were fluent and useless, because fluency was never the bottleneck.
What changed is extraction under a schema. Listing AI tools now read a batch of photos of a single object as one input, pull the brand off the label in the second shot and the part number off the sticker in the fifth, judge condition from what is visible including the scuffs, and then check the marketplace’s own category tree and required field list before writing anything into it. That last step is the part that matters and the part that is genuinely hard. An invented attribute value gets a listing rejected. The output has to validate, not just read well.
The practical effect is that the seller no longer has to know what the thing is before they start. Photograph the drill, get back a title, a category, a condition assessment, the item specifics, and a suggested starting price range, in roughly a minute, with every field editable before anything goes live.
Photo handling has quietly converged on the same logic. Marketplace grids reward plain backgrounds, and a garage floor is not one, so these tools cut the background to white, grey or transparent, straighten a crooked shot, and adjust exposure. What they do not do, and must not do, is regenerate the item. A resale photograph is evidence. The buyer is purchasing that specific object with that specific worn corner, and anything that redraws it crosses from cleanup into misrepresentation. Cutting away a garage floor is fine. Producing a picture of a drill that does not exist is fraud.
Distribution is where the current generation is still uneven. Publishing directly into a seller’s shop requires a real integration with that marketplace, and most of these tools have one or two at best. Etsy is the common one. For eBay, Poshmark, Mercari, Craigslist and Facebook Marketplace, the workflow is still generate, then paste, which is unglamorous but takes seconds and sidesteps a category of account-permission problems that sellers are right to be cautious about.
What This Does Not Fix
A suggested price range is a starting point produced at the moment of listing. It is not market monitoring, and it does not reprice anything a week later when three identical items appear. Sellers who treat it as research will get burned on the items where it matters most, which are the rare ones.
Extraction also fails in the ordinary way that extraction fails. A worn model number gets read as a similar model number. A reproduction gets read as the original. This is why the field-by-field review before publishing is not a formality, and why any tool in this category that removes the review step is doing its users harm.
The deeper limitation is that none of this creates the intent to sell. It only lowers the cost of acting on it, which turns out to be the number that governs everything downstream. At twenty minutes an item, anything under about thirty dollars is not worth listing, so it doesn’t get listed, so it stays in the corner of the room and eventually goes to landfill or a donation bin. At one minute an item, that whole tier of goods comes back into circulation.
The circular economy has spent a decade being described as a demand problem and a logistics problem. It was mostly a typing problem.
