You step away from a speech with your mind replaying every stumble, every lost thought, and every moment when your voice did not sound the way you wanted.
Then trained observers watch the same performance and see something very different.
In a study of university students, speakers gave themselves an average performance rating of 38.58 on a 68-point scale. The trained observers' combined rating averaged about 56.3, a gap of about 17.7 points.
The observer mean was about 46% higher than the speaker mean. That is the arithmetic behind the headline. It does not turn the scale into an objective measure showing that every speech was literally 46% better.
What the researchers tested
The final sample included 95 university students. After three minutes of preparation, participants gave an impromptu speech about themselves that could last up to ten minutes.
The study combined three kinds of measurement:
- Self-reported anxiety and performance
- Behavioural ratings by trained observers
- Physiological measures, including electrodermal activity and heart-rate variability
What they found
The speakers' average self-rating was 38.58. The trained observers' combined rating averaged about 56.3.
Self- and observer ratings were still moderately related (r = .60). Speakers who performed better according to observers tended to rate themselves better too. The disagreement was largely about level: speakers judged the performances much more harshly overall.
The observer mean was about 46% higher because the 17.7-point difference is roughly 46% of the speakers' mean. The scale does not establish that a score of 56 represents 46% more real-world performance than a score of 39. The measured difference was about 17.7 points on the study's 68-point scale.
What the physiology showed
The study did not find a significant relationship between self-reported public-speaking anxiety and physiological reactivity during the speech.
That does not mean “your body betrays nothing.” Heart-rate variability and skin conductance do not measure every visible sign of nervousness, and absence of a significant relationship is not proof that an audience can never detect anxiety.
The result supports a narrower point: subjective anxiety, observed behaviour, and physiological activity are related but distinct ways to measure a speaking experience. One cannot simply stand in for the others.
Why self-ratings may be harsher
Speakers know what they intended to say. Observers see only what was delivered. A forgotten point can feel like a major failure to the person who remembers the plan, while remaining invisible to everyone else.
Anxious self-monitoring may also draw attention to mistakes. The study found that higher self-reported anxiety was associated with lower self-rated performance.
The design does not tell us that observers were perfectly objective. It tells us that self-assessment alone gave a systematically different picture.
How to use the finding
Keep the raw score
After a speech, rate yourself before reading feedback. Compare the rating with an external review. Over several attempts, look for a consistent gap.
Ask for observable evidence
Replace “Was I bad?” with “Where did my main point become unclear?” or “Which example did you remember?” Specific questions produce feedback you can use.
Separate feeling from performance
Record both: “I felt anxious” and “I stated my answer in the first sentence.” The first is an experience. The second is an observable behaviour. Both matter, but they are not interchangeable.
The bottom line
Your harshest critic may indeed be in your head. In this study, speakers scored themselves about 17.7 points below trained observers on a 68-point scale.
The 46% figure compares the two mean ratings. It is not a calibrated percentage improvement in real-world speaking performance, and observers are not infallible. But the gap is strong evidence that how a speech feels from inside can be much harsher than how it looks from outside. Do not replace self-reflection with outside feedback. Use both, and investigate the gap.
Put this into practice
Use AI speech coaching to compare how an attempt felt with what is observable in the recording. Focus on one specific change for the next attempt instead of turning discomfort into a verdict on the whole speech.