AI Speaking Practice: Separate Feeling Nervous from Looking Nervous

Imagine recording a short introduction to your next workshop. Your heart is racing, you notice a pause, and you finish convinced that every viewer will see how nervous you felt. You turn to AI for a verdict on whether you seemed confident.

Before accepting that verdict, separate three questions: what you experienced, what another person perceived and whether your message worked. A strong feeling during the recording is information about your experience. It does not, by itself, establish what an audience noticed.

Public-speaking research offers a useful distinction for AI-assisted rehearsal. The aim is to make feedback more precise: identify the evidence behind an impression, then choose a change that serves the presentation.

What the speaking research actually compared

Kenneth Savitsky and Thomas Gilovich’s study in the November 2003 issue of the Journal of Experimental Social Psychology examined the illusion of transparency: overestimating how apparent an internal state is to others.

Participants gave improvised speeches in pairs, then assessed their own and their counterpart’s apparent nervousness. Speakers thought they had looked more nervous, on average, than their partners judged them. These were appearance ratings, not objective measurements of feelings.

In a second experiment, observers rated speakers informed about the illusion as more composed and their speeches more positively than speakers given general reassurance or no additional instructions.

Those human experiments did not test AI coaching. Their findings offer no guarantee about an individual speaker or the disappearance of nervousness.

Being noticed and being understood are different questions

The related spotlight effect concerns how much attention we believe our behaviour receives. In research published in February 2000, Thomas Gilovich, Victoria Husted Medvec and Kenneth Savitsky found that participants overestimated how many observers would remember an image on their T-shirt. Other participants overestimated the prominence of their positive and negative contributions to a group discussion.

That concerns perceived salience: how much something stands out to other people. The illusion of transparency concerns the apparent visibility of an internal state. Neither finding means that audiences ignore mistakes, misunderstand nothing or never form unfavourable impressions.

For a presenter, the practical question remains specific. Did a pause obscure an important instruction? Did a listener understand the example? Was an essential date missing? Those questions allow a review to identify something that can be examined and revised.

Give AI a task supported by the material

Return to the fictional workshop introduction. Suppose the creator supplies an assistant with only the written transcript. A comment about the wording can be checked against that text. A confident claim about how nervous the creator looked would lack visual evidence in that input.

Keep this boundary visible in the request. Ask for a review of the opening, the order of information and the stated next action. Require suggestions to identify the passage they concern. This is a proposed editorial task, rather than an assertion about the capabilities or accuracy of a particular product.

A script critique also remains different from an audience test. An assistant’s prediction that viewers will find the introduction reassuring is a prediction to examine. It is not a report from the people who will watch the finished recording.

If you also use video analysis, MaryChuks’ guide to AI emotion recognition and context explains why visible behaviour, an emotion label and a person’s own account deserve separate treatment. Establish what a tool actually received and what kind of judgement its output represents.

Review a recording around a communication goal

For an independent creator preparing an announcement, I would use the following routine. It is an editorial proposal informed by the distinction above, not a tested psychological intervention:

• State the purpose. Write what the intended viewer should understand and what, if anything, they should do next.
• Record a specific prediction. Replace a broad verdict such as “I looked terrible” with a question about a moment: did the pause make the booking instruction hard to follow?
• Review the available evidence. Compare comments about wording with the transcript and comments about delivery with the original recording.
• Ask a relevant viewer about the message. For this workshop example, ask what the session covers, when it happens and how they would book.
• Choose a defined revision. Change a passage because of the communication problem identified, and review the revised version.

A viewer’s response tells you about that viewer and that version. It does not predict everyone’s reaction. Equally, feedback that identifies an unclear instruction deserves attention even if the recording looks polished.

Let the next decision reflect what the review found

Suppose, within this fictional example, a viewer recalls the workshop topic and date but cannot explain how to register. The creator still reports feeling very nervous while recording. Those two pieces of information can coexist.

The immediate editorial revision is to make the registration instruction easier to find. The creator could then check whether a viewer can identify the next step in the revised introduction. The review has produced an actionable question about the message.

There is no need to declare the creator’s feelings mistaken. Nor should the group-level research be used to assure them that nobody noticed any nervousness. The useful move is to avoid treating an untested prediction about other people as a settled observation.

The same discipline applies to reassurance from an assistant. A supportive comment can be welcome while its factual basis remains open to examination. Keep any recommendation attached to the passage, observation or viewer response that supports it.

Keep a record of useful progress

Before another rehearsal, write down the one change you are checking. After it, record what became clearer and what still needs work. For the workshop introduction, that might mean comparing whether the session description, date and registration route can each be found.

This gives practice a purpose beyond obtaining another general confidence rating. It also makes a handover useful: a collaborator can see which communication problem prompted an edit and whether the revision addressed it.

The research supplies a reason to question the assumption that a feeling is fully visible. The rehearsal supplies a separate opportunity to investigate the message. Use AI suggestions within that process, with enough context and evidence for a person to assess them.

Sources

• Savitsky and Gilovich, The illusion of transparency and the alleviation of speech anxiety — Journal of Experimental Social Psychology, November 2003 issue, 39(6), 618–625.
• Gilovich, Medvec and Savitsky, The spotlight effect in social judgment — Journal of Personality and Social Psychology, February 2000, 78(2), 211–222.