Most people who are afraid of public speaking are not afraid of speaking — they're afraid of being evaluated. The racing heart, the dry mouth, the mental blank are not responses to the act of forming sentences out loud. They're responses to the perceived judgment of an audience. That distinction matters, because it changes what kind of practice actually helps. Dr. Lena here — and AI is changing what's possible for people who want to become comfortable and compelling in front of others.

The Evaluation Anxiety Problem

Public speaking anxiety is among the most prevalent performance anxieties. A significant portion of the population rates it above fear of death in surveys about what they dread — a finding that always gets cited as a joke, but which reflects something real: the experience of negative social evaluation is processed by the brain as a genuine threat. Your nervous system responds to a difficult presentation the same way it responds to physical danger, with elevated cortisol and a narrowed attentional field that makes thinking clearly and accessing memory significantly harder. The problem with most speaking practice is that it happens in conditions that already trigger this response. Toastmasters, university speaking classes, office presentations — all involve actual audiences, real evaluation, real social consequences for poor performance. For people with significant speaking anxiety, these conditions produce the very response you're trying to overcome. Practice becomes re-traumatizing rather than desensitizing. Research from the University of Amsterdam on graduated exposure for performance anxiety found that practice in low-stakes conditions — conditions where the evaluative threat is absent or minimal — produces more effective anxiety reduction than practice in high-stakes conditions, particularly in the early stages of skill building.

What AI Practice Changes

AI creates conditions for genuine graduated exposure. You can practice delivering a talk with no human present, receive feedback that is specific and useful, and repeat the same section until you've resolved whatever was breaking down — without accumulating the social cost of performing poorly in front of someone who can form a lasting impression of your competence. This low-stakes volume is the core of what AI offers. The feedback quality on specific, articulable dimensions — filler words, pacing, sentence length, structural clarity — can be high. What AI doesn't provide is the physiological experience of an audience, which is ultimately what needs to be trained. That training still requires real audiences, even small and friendly ones. But AI practice allows you to arrive at that audience experience with the content already internalized and the cognitive load of managing structure and language already reduced.

The Part That AI Practice Can't Replicate

The thing that separates competent speakers from compelling ones is harder to practice in isolation. Compelling speakers are responsive — they read the room, they adjust, they seem to be having a genuine thought in real time rather than delivering a prepared performance. This quality is partly a function of reduced anxiety (when you're not managing fight-or-flight, you have more cognitive resources for actual presence) and partly a function of genuine interest in the moment. AI practice can help with anxiety reduction and content internalization. It cannot replicate the live responsiveness that makes a speaker feel present and authentic rather than performative. That requires actual humans, and specifically humans who are genuinely engaged — people who are listening and reacting in real time to what you're saying.

A Tangent About Filler Words

The standard advice on filler words — um, uh, like, you know — is to eliminate them. This advice is often counterproductive. Research from Columbia Business School on speaker credibility found that a small number of disfluencies actually increases perceived authenticity and listener trust. Completely fluent, filler-free speech can read as either heavily rehearsed or unnaturally controlled. The goal shouldn't be zero fillers but intentional pausing: replacing an unconscious "um" with a deliberate, comfortable silence, which is both more authoritative and less distracting than the filler it replaces.

How to Build a Useful Practice Routine

Start with AI practice focused on structure and content — get the material to where you can deliver it without the cognitive load of managing what comes next. Use recordings to catch specific habits: pacing, filler patterns, whether your conclusion actually lands. Then move to low-stakes human audiences — a friend, a small group, a video recording you review with a trusted person — to start building the real-audience calibration. The progression from AI to small human audience to larger contexts is more effective than jumping directly to high-stakes performance, which is what most people do because they haven't had any other option. AI has made the graduated approach actually possible.