Confidence Spillover

You answer a question and feel 65% sure.

Then AI agrees.

Your confidence rises.

Nothing new may have been added. The system simply endorsed the answer.

Now imagine the AI disagrees. Even if you reject its advice, a small doubt can remain.

AI does not only influence decisions. It can influence how we feel about our own judgement.

Confidence is part of the decision

Metacognition is our ability to evaluate our own thinking: how sure am I, what do I know, when should I seek more evidence?

A 2026 study on AI advice and human metacognition found that contradictory AI advice reduced retrospective confidence judgments. Decision choice and confidence did not respond identically, which is important: a person can keep the same answer while feeling less certain about it.

Another 2026 clinical study with 292 clinicians found that AI agreement increased diagnostic confidence — even in cases where both the human and AI were wrong. Disagreement slightly reduced confidence.

That is a sobering result.

Agreement can make error feel more certain.

AI can become a confidence amplifier

Imagine a junior analyst who reaches a weak conclusion and receives AI support.

The output may provide additional arguments, making the original idea feel researched.

A founder may mistake fluency for independent validation.

A doctor may feel reassured by agreement even when both human and model share the same blind spot.

The problem is especially important when AI is trained on similar information sources to the human user. Agreement may not be independent evidence.

Ask whether the AI had a separate route to the answer

If you give the AI your reasoning and ask, “Do you agree?”, the response is contaminated by your frame.

A better method is sometimes to ask for an independent assessment before revealing your answer.

For example:

“Analyse this case and give your conclusion before I tell you mine.”

Then compare.

Or ask the system to generate confidence separately and identify uncertainty.

Preserve your pre-AI judgement

For important decisions, record your initial view and confidence before consulting AI.

Then note:

What did the AI add?

Did it provide new evidence or only restate my reasoning?

Did my confidence change more than the evidence justified?

This creates a visible audit trail of influence.

Questions worth sitting with

• Does AI agreement make you feel validated even when it adds no independent evidence?
• Do you record your own judgement before consulting a system?
• Where would false confidence be more dangerous than a wrong first answer?

Leave points

• AI advice can change confidence even when it does not change the final decision.
• Agreement may increase confidence when both human and AI are wrong.
• Fluency and corroboration can feel like independent evidence when they are not.
• Record important pre-AI judgements before exposure to advice.
• Ask whether the system reached the conclusion independently or mainly followed your frame.