Game Theory · GT-22
Costless, unverifiable communication — sometimes it changes outcomes, sometimes it's worth exactly what it cost to say.
Communication that carries no direct cost and isn't binding — statements of intent, non-verifiable claims, pre-play chat before a game begins. Whether cheap talk can meaningfully change equilibrium outcomes depends entirely on whether both parties' interests are sufficiently aligned; when interests diverge sharply, cheap talk carries no credible information at all.
Formal treatment developed by Vincent Crawford and Joel Sobel in their influential 1982 paper 'Strategic Information Transmission,' establishing when and how much costless communication can convey.
The Mechanism
Aligned interests vs. conflicting interests — the same words, different credibility
The Crawford-Sobel result: the amount of real information cheap talk can convey is a direct function of how aligned the two parties' interests actually are — from fully informative (aligned) down to pure noise ('babbling equilibrium,' fully opposed).
01 · CHEAP TALK IN BATTLE OF THE SEXES-LIKE GAMES IS FULLY CREDIBLE
When both sides want to coordinate, talk works
In games like Battle of the Sexes, where both players prefer any coordinated outcome over any uncoordinated one, simply announcing a preference is credible and genuinely resolves the coordination problem — there's no incentive to lie, so nobody does.
02 · IN CONFLICTING-INTEREST GAMES, CHEAP TALK CAN COLLAPSE TO PURE NOISE
The 'babbling equilibrium'
When interests diverge sharply (think Prisoner's Dilemma-style incentives layered onto communication), any pre-play promise ('I'll cooperate') carries zero credible information, because it would be said whether true or false — rational listeners correctly ignore it entirely, an outcome Crawford and Sobel formally labeled the babbling equilibrium.
03 · PARTIAL ALIGNMENT PRODUCES PARTIAL, COARSENED INFORMATION
Real communication is often exactly this middle case
Most real-world messages (a supplier's claim about delivery timelines, a job candidate's claim about their skills) fall between the two extremes — Crawford-Sobel shows the equilibrium message typically conveys coarse categories ('roughly on time' vs. an exact date) rather than either full precision or total noise, because interests are partially but not fully aligned.
Where It Fails / Inversion
Where it fails / inversion
Treating any verbal assurance as meaningful evidence, regardless of the underlying incentive alignment, is a reliable way to be misled — 'I promise I'll deliver on time' from a counterparty with every incentive to say that whether or not it's true carries essentially the babbling-equilibrium level of information, and should be weighted accordingly, however sincere it sounds.
How To Use It
Worked example · reading vendor promises during procurement
When evaluating a vendor's verbal assurances during a sales pitch ('we'll definitely hit your deadline'), the correct question isn't whether the statement sounds sincere — it's whether the vendor's incentives are aligned enough with the truth that the statement carries real information. A vendor whose contract has no penalty clause for delay has weak incentive alignment; their cheap talk about timelines approaches the babbling equilibrium. Adding a real penalty clause changes the underlying game and makes their subsequent statements meaningfully more credible.
How to use it
Before weighting any unverifiable promise or assurance in a negotiation, ask the Crawford-Sobel question directly: are this person's incentives aligned enough with telling me the truth that their statement carries real information, or would they say the same thing regardless of what's actually true? If the latter, discount the statement toward zero — it's babbling, not information — and look for a costly, binding signal instead.
See Also