The Same AI, a Different Face: How Counterpart Cues Change Our Decisions

Imagine discussing a commercial problem with two counterparts. They make the same argument, offer the same terms and respond with the same degree of confidence. You believe you are judging their proposals on merit. Would you behave exactly the same way toward both?

Most professionals would like to think so. A recent study of AI assistants gives us reason to examine that confidence more carefully. It does not test negotiation outcomes, but it isolates something negotiators encounter constantly: social signals can change how a counterpart is perceived even when the underlying contribution is unchanged.

Identical capabilities, different rewards

In a study published on 3 October 2026 in the proceedings of NordiCHI, Isabelle Cuber and colleagues examined how AI assistant presentation affected workplace interactions in virtual reality. The researchers recruited 189 knowledge workers to complete tasks with assistants that had the same underlying AI capabilities but differed in their presentation. Participants rated the assistants and allocated real monetary rewards. The researchers also conducted 34 interviews.

The female-presenting assistant received 10.25 percent less in monetary rewards than the male-presenting assistant. The male-presenting assistant was also perceived as more humanlike. More humanlike presentation was associated with greater trust and reward. Yet many interview participants said they preferred clearly nonhuman assistants and that gender did not matter to them.

This is not evidence that all people discriminate in the same way, or that a particular negotiator would make a worse commercial deal with a female counterpart. It is a controlled workplace interaction study, not a procurement negotiation experiment. Its value lies in the contrast between stated beliefs and observed decisions.

What we say matters may not be what moves us

Negotiators often explain their own behaviour in terms of facts, prices, deadlines and alternatives. Those things matter. But the other person is never merely a spreadsheet. Their voice, confidence, status, manner and apparent authority become part of the context in which we interpret their argument.

A senior-sounding supplier may be granted more credibility than a junior-sounding one. A confident refusal may feel more final than a hesitant refusal, even when neither changes the supplier's actual alternative. An assertive counterpart may trigger defensiveness where a calm counterpart elicits curiosity.

Those are plausible behavioural risks, not findings established by the AI assistant study. They are hypotheses worth testing in practice rather than assumptions to build into a scorecard.

The more important question is whether the buyer notices the difference. It is easy to say that gender, status or presentation should not influence a commercial decision. It is harder to detect the moment a social cue actually changes the way we question, concede or challenge.

A realistic simulation needs more than a correct argument

This matters for negotiation training. If every simulated supplier speaks with the same polite tone, cooperates after a predictable objection and yields when the buyer follows a script, the exercise tests a narrow version of the skill.

A useful practice environment varies the resistance, the counterpart's priorities and the way pressure is expressed. A buyer may need to stay curious with an intimidating finance executive, hold a condition against a persuasive account manager or challenge an apparently authoritative claim without becoming confrontational.

Presentation is one dimension of realism, but it must not become a stereotype. A female persona should not automatically be more accommodating, nor should an older or more senior persona automatically be more competent. The commercial facts and incentives must remain coherent. Otherwise the simulation risks teaching the very shortcuts it should help users recognize.

An especially useful training design is to hold the commercial situation constant while varying counterpart cues, then compare the buyer's own decisions. Did the questions change? Was the same claim accepted without evidence? Did the buyer give away more to end discomfort? Such comparisons can reveal something a self-assessment questionnaire may miss.

Observe behaviour without pretending to read minds

The study's interviews highlight a measurement challenge. People can sincerely believe a characteristic does not affect them while their choices suggest otherwise. That does not make every behavioural difference proof of bias. Different interactions contain noise, and an observed concession can have several explanations.

A careful debrief therefore separates what happened from why it might have happened. The transcript may show that the buyer stopped challenging an assertion. It cannot, on its own, establish that the counterpart's gender, voice or perceived status caused the change. A stronger test requires comparable scenarios, repeated observations and consideration of alternative explanations.

This is the same reason negotiation performance should not be confused with the final result. A score without context can be as misleading as an intuition without evidence.

The capability is noticing the cue before it becomes the decision

Negotiation competence does not require becoming immune to social signals. It requires recognizing when they are influencing a decision that should depend on commercial evidence.

When a supplier sounds certain, ask what supports the claim. When a senior counterpart applies pressure, return to the mandate and available alternatives. When the conversation feels uncomfortable, distinguish the urge to end that discomfort from a genuine reason to move.

Those responses become more reliable through spoken practice, where the buyer must make the choice while the pressure is present rather than describe the correct choice afterward. Voice2Evolve uses realistic counterpart conversations and post-session analysis to make those moments available for deliberate review.

The lesson from the new research is not that an AI avatar determines a negotiation outcome. It is that seemingly superficial cues can shape human decisions even when the underlying task stays the same. Good negotiators need to know not just what they believe, but what they actually do when the person across the table changes.

Sources

• Cuber, I. et al. (3 October 2026). Human-Like and Male? How AI Assistant Design Relates to Trust and Monetary Reward at Work in VR. Proceedings of the 14th Nordic Conference on Human-Computer Interaction.
• University of Zurich (8 October 2026). Mehr Lohn für männliche KI-Assistenten.