What is citation bidding?

What is being bought

In search, the paid unit is a position above the organic results — visibly separate, and competing with them for attention. In an assistant's answer there is no list to sit above. The unit that has value is being named at all.

Citation bidding is the auction for that. The advertiser is buying answer insertion: appearance as a cited source or a named option in the response the user actually reads.

It sits deliberately at the boundary between paid placement and organic GEO, which is what makes it both valuable and the format requiring the most careful governance.

Why it prices so differently

• Supply is one, not many. An answer names two or three options. There is no equivalent of positions four through ten.
• Timing is exact. The placement occurs at the moment of decision, not somewhere upstream of it.
• Implicit endorsement. Appearing inside an answer carries weight a banner does not, which is precisely why disclosure is non-negotiable.
• No arbitrage. The moment cannot be reached any other way, so there is no cheaper substitute to bid against.

The governance problem

Every mechanism here has to defend against the same failure: an assistant whose recommendations can be bought outright is worthless, and quickly known to be worthless.

The constraints that make it defensible are consistent across the platforms doing it seriously. Paid inclusion is disclosed. Payment selects among genuinely suitable candidates and cannot introduce an unsuitable one. Factual content is not for sale. And there is a limit on how much of an answer can be paid for at all.

A buyer should treat the absence of any of these as a signal about the surface rather than as an opportunity.

Paid and organic together

Citation bidding and GEO reach the same outcome by different routes and on different timescales. Paid inclusion works immediately and stops the moment spend stops. Organic inclusion is slow, compounding, and cannot be bought — it is a function of the trust graph.

Run as one programme they reinforce each other; run separately they compete for the same budget on incompatible measurement. The reporting that reads both is AI share of voice.