Why Automated Systems Can't Tell Real Handmade From Fake

In 2024, Etsy banned 3.5 million accounts and removed more listings than any previous year, a 22% increase over 2023. The vast majority of those decisions were made by automated systems, not human reviewers.

Some of those removals caught genuine resellers and counterfeiters. Many others caught real makers who did nothing wrong.

I worked at Etsy. I watched these systems being developed. And I can tell you that the fundamental challenge isn't a lack of care about real makers. It's that the technical approach required to moderate at this scale is structurally limited in its ability to distinguish a genuine handmade seller from a sophisticated reseller. That's not a bug anyone can fix with better tuning. It's a limitation inherent to automated moderation across every major marketplace.

Let me explain what I mean.

How the detection systems actually work

The enforcement relies on several automated detection methods running simultaneously. I won't reveal specifics that could help bad actors game the system, but I can give you the general picture.

The most notorious one, what sellers call "the AliExpress Bot," performs reverse image searches across major marketplaces. It looks for Etsy listing photos that also appear on AliExpress, Temu, Alibaba, and similar sites. A match triggers a flag. The limitation: the system can't determine who posted the image first. When a scammer steals a maker's photo and posts it on AliExpress, the system sees a match and flags the Etsy listing. The maker ends up caught in a net designed to catch someone else.

Other systems analyze seller behavior for patterns associated with reselling. Shipping directly from certain regions, listing large volumes of similar items rapidly, pricing patterns that suggest wholesale sourcing. These signals are useful in aggregate (a shop exhibiting many of them is probably a reseller), but individual signals can match legitimate sellers too. A maker who buys materials from overseas, lists seasonal inventory in batches, or prices competitively can look like a reseller to an algorithm.

Certain product categories, keywords, and listing attributes carry higher risk scores based on historical patterns of abuse. If your products fall into a category that's been heavily exploited by resellers, your listings face more automated scrutiny regardless of whether you're genuine.

Even 99% accuracy means thousands of wrong calls

The core challenge is statistical. Etsy has over 7 million active sellers. Even a system that's 99% accurate produces tens of thousands of false positives at that scale. And for edge cases, accuracy is well below 99%. This is the fundamental tension of automated moderation: any system aggressive enough to catch the flood of resellers will inevitably sweep up legitimate makers too. Every major marketplace faces this same tradeoff.

These systems struggle with legitimate makers for reasons that feel almost obvious once you see them. An algorithm looking at a listing photo has no idea whether that item was made in your garage studio or a factory in Shenzhen. The photo looks the same either way. A hand-thrown ceramic mug and a mass-produced one look identical in a product photo. The difference is in the making, and no image-matching algorithm can see the making.

And there's no mechanism in the current system for a seller to pre-establish themselves as the original creator. When the image matching system finds your photo on AliExpress, it doesn't ask "who created this image?" It just sees a match and takes action.

There's a real frustration sellers express in forums that captures something important: once an automated system makes a decision, reversing it through the appeals process can feel inconsistent. The automated systems handle the initial decision, and by the time a human reviewer sees your case, you're starting from a position of having to prove your innocence. Sellers report talking to different support agents and getting different answers. This isn't because anyone is being unreasonable. It's because the volume of cases makes consistent, nuanced review incredibly difficult. Etsy's support team is genuinely trying to help, but they're working within the same constraints the automated systems create.

You can't fix the system, but you can be ready for it

So you can't change how the automated systems work. But you can make yourself much harder to wrongly flag, and much easier to reinstate if you are.

Use unique, high-quality photos that clearly show handmade details. Tool marks, slight imperfections, your workshop in the background. Write descriptions that emphasize your process, not just the product. "Hand-carved from locally sourced walnut" is harder for a system to confuse with a factory product than "beautiful wood sculpture." Avoid stock-style photography that looks like it could come from a manufacturer's catalog. Show your hands. Show your tools. Show the mess.

The single most important thing you can do is document your creative process before you get flagged. When (not if) an automated system makes a mistake, you want to be able to respond with organized, timestamped proof within hours, not weeks. Process photos with original metadata, supply chain records, design files with creation dates, and a system for keeping all of it organized and accessible.

If scammers steal your photos and post them on AliExpress, you want to know about it before the detection systems do. Regular reverse image searches of your product photos can give you a head start on filing takedowns and preparing your defense.

When I worked at Etsy, I kept thinking about something that seems so obvious to me: a way for genuine sellers to proactively verify themselves. A system where makers could submit documentation of their process, have it reviewed, and earn a "verified maker" status that automated systems would check before removing a listing. Etsy explored this idea but never shipped it. The reasons had more to do with prioritization and resource constraints than with the idea itself (I wrote more about that in another post). But the need for it has only grown.

That gap is why ProvenMaker exists. It gives you the documentation layer that automated systems don't have. Timestamped proof of your creative process, image monitoring for theft detection, and organized evidence packages ready for appeals. It's free for early users, because I believe every real maker deserves tools that work as fast as the systems that flag them.

The automated systems can't tell that your work is real. But your documentation can.

Want to understand how your specific shop might be at risk? I'm building tools to help assess that. Reach out and I'll keep you posted as they roll out.