Field Guide

Eighty Percent of What? Whatnot's Shill-Bidding Claim and Your Comps

Bayley Coleman · 2026-08-03 · 7 min read

Whatnot published an encouraging set of trust-and-safety numbers: shill bidding down "nearly 80%" in six months, user reports of suspected shilling down 45%, and a fivefold increase in the signals it uses to spot suspicious activity, with a Trust and Safety team that "more than doubled in size last year." If those figures mean what they sound like, comps from the platform just got cleaner, which affects every card you price off them.

The problem is that you cannot tell what they mean. And a number you cannot interpret should not change what you pay for a card. Here's how to read a claim like this, and what it costs you to get it wrong.

What did Whatnot actually claim?

The reported figures, as published:

• Shill bidding activity "declined nearly 80% over the last six months."
• A "45% drop in user reports of suspected shill bidding."
• A "fivefold increase in the number of signals used to identify suspicious activity."
• The Trust and Safety team "more than doubled in size last year."

And the reporting on it is explicit about what came with those numbers, which is nothing underneath them. No totals for how many accounts, auctions, or transactions were affected. No statement of whether the 80% refers to bids, auctions, accounts, sellers, or an internal metric. No baseline rates and no absolute numbers.

To be fair to Whatnot on two points. It did disclose one useful denominator elsewhere, that sellers running breaks are about 4% of its seller base. And the direction of all four figures is the direction you'd want. This is not an accusation that the numbers are false. It is an observation that they are not checkable, which is a different thing and still decision-relevant.

Why an undefined percentage can't move your max-buy

Four pieces are missing, and each one changes what the 80% would mean:

The third row deserves care, because it cuts both ways honestly. Fewer user reports alongside far more detection is consistent with the platform catching problems before users notice them, which would be genuinely good. It is also consistent with users reporting less for unrelated reasons. Both readings fit the disclosed facts, and that ambiguity is the whole point: when two opposite stories fit the same numbers, the numbers aren't evidence yet.

What a bad comp actually costs you

Here's why this isn't academic. Comps are the input to every decision the engine makes. If the sold prices you're reading are higher than what cards genuinely clear, your max-buy is too high by roughly the same margin, and you overpay on every unit you buy.

Put a number on it. You keep about 87% of a sale price after fees, plus a few dollars of flat costs, which is the same assumption our break-even math runs on. So on a card you believe comps at $100, you expect to net about $84. Now suppose the real clearing price is lower than the comp you read:

The pattern is a rule worth memorizing: every $1 of phantom comp costs you about 87 cents of real money. On a thin flip, a 10% inflated comp is the whole margin. And notice what you cannot do with Whatnot's disclosure: you cannot pick a row. Without a baseline rate you have no way to estimate how inflated any given comp might be, before or after an 80% improvement.

What would make the claim checkable

This is the constructive version, and it's a short list. To turn that 80% into something a flipper could actually use, a platform would need to publish:

1. The denominator and the unit. 80% of what, counted how.
2. The baseline. The rate it fell from, in absolute terms, not just the change.
3. Detection held constant. The decline measured against a fixed detection method, so behavior change is separable from measurement change.
4. A category split. Trading cards broken out from the rest of the platform.

None of that is exotic. It's the standard any of us would apply to a seller's claim about a card.

What to do until then

• Prefer fixed-price sold data for your primary comps, and use live-auction results as a secondary check rather than the anchor. Auction dynamics add noise even when everyone is honest.
• Read the whole sold range, not the top. A single high sale is the least reliable data point in any set, which is the same reason we warn against comping to headline sales in the signed-vintage piece.
• Demand a cushion. The engine requires a margin before it says buy, precisely because inputs are imperfect. Comp uncertainty is one of the things that cushion is for.
• Treat unverifiable good news as neutral. Not false, just not actionable. If the improvement is real, you will see it in tighter, more consistent sold ranges over time, and that you can measure yourself.

The broader point outlives this one announcement. Platforms will keep publishing self-reported progress, because they should, and some of it will be real. Your job is narrower than judging them: decide whether a number is specific enough to change a price you pay. Nearly 80% of an undisclosed base isn't. A guide, not gospel.

Frequently asked questions

Sources

• Whatnot says shill bidding fell nearly 80%, Value Added Resource (published July 28, 2026, updated August 1, 2026) (all four claimed figures, the explicit note that no totals, denominators, or baseline rates were provided, the 4% breaker share, and the Lesko arbitration claims with Whatnot's response that gambling is prohibited and card breaking is a longstanding collecting format)
• Whatnot says shill bidding fell nearly 80% but doesn't define what the figure counts, Shopifreaks