Keeping bids fair: how we're taking on shill bidding
Behind every bid is trust. You're trusting that the price in front of you is real, set by people who want the item just like you do. It's easy to take for granted, but it matters every time. That's why we've always worked to keep bidding fair, and why we’re sharing more about the progress we’ve made.
For those of you who may not be aware of the term, shill bidding is when someone places a bid with no real intent to buy, only to push the price up. Sometimes that looks like a seller bidding on their own item. Other times, it’s a friend, employee, or mod bidding on their behalf. Either way, it's a bid that isn't real, and it chips away at trust for everyone else.
Shill bidding isn't new. Our signals are.
Shill bidding is as old as auctions themselves. Long before the internet, it happened in the room: a nod to someone in the audience, an understood arrangement before specific items came up. In a crowded room, it was easy to miss and hard to prove.
Online auctions change that. Every bid carries information. The account behind it, its device fingerprint, how it connects to other accounts, and how it lines up with the way that bidder usually behaves. We can map relationships between accounts, score bidding and cancellation patterns against what's normal, and flag anomalies at a speed no auction house could ever match. That gives us something no IRL auctioneer ever had: the ability to see the patterns behind manipulation and act before they do damage.
What progress looks like
Tackling shill bidding takes people and machines working together. Last year we more than doubled the size of our Trust & Safety team, making it the largest function at the company. That means more data scientists and machine-learning engineers designing new ways to detect manipulation, more engineers building automated systems that act on those signals around the clock, and more expert investigators digging into the cases automation alone can't resolve. When we confirm shill bidding, we take action, including restricting or permanently banning accounts that manipulate auctions.
And it’s paying off. Over the past six months we have:
Cut shill bidding activity by nearly 80%.
Increased the number of signals we use to proactively detect shill bidding by 5x.
Brought reports of shill bidding down by 45% overall.
That last number is the one we watch most closely. Our community is sharp — buyers watch every auction live, and they notice when something feels off. That makes their reports one of the best independent measures we have. It isn't our models alone grading their own work; it's millions of perceptive buyers telling us whether auctions feel fair. When fewer of them flag suspicious bidding even as more people bid than ever, that's real progress.
What we can’t show you
Writing this post means walking a line. We want to be open about shill bidding, but there's a lot we can't fully talk about. The more we explain about the models we've built, the behavioral signals that feed them, how we detect anomalies in bidding patterns, or how we map hidden connections between accounts, the easier we make it for bad actors to design around us.
This work never stops
Fair bidding, receiving what you paid for, and real community are the standard Whatnot was built on, and it's the standard we hold ourselves to. We've made real progress, and we're not done. We'll keep listening, investing, and building the best experience we can – sharing updates here as we go.
