Free AI for Public Speaking Practice

What AI is good for, and what it is not

It is good at the things that are true of the recording and invisible to you while speaking. You do not know that you sped up by thirty words per minute in the last two minutes, that your longest pause was under half a second, or that the word you built the whole talk around came out too soft to hear four times. All three are measurable and all three are fixable within a week.

It is not good at whether the talk was worth giving. Structure, the choice of examples, whether the ending earned the applause - a competent human audience judges those in a way nothing automated approaches. If your problem is that the material is thin, no measurement will find it.

It is also not an audience. Rehearsing alone into a laptop trains you for rehearsing alone into a laptop, which is why the last rehearsal before a real talk should have at least one person in the room even if the previous ten did not.

A talk is not a pitch, and the numbers move

A sixty-second pitch is short enough that one pace figure describes it. A ten-minute talk is not. What you want from a longer rehearsal is pace per section, because the failure mode is drift: most speakers open at a reasonable rate and finish thirty to forty words per minute faster, usually starting from the slide they were least sure of.

Filler rate behaves differently too. In a short pitch fillers cluster before numbers. In a talk they cluster at transitions, in the two or three seconds after one section ends and before the next begins, which is exactly where a rehearsed pause would have done the work instead.

Overrun is its own measurement and the one most talks fail on. Time each rehearsal end to end rather than trusting the slide count. A talk that runs to eleven minutes in rehearsal runs to thirteen in the room, because nerves add words rather than removing them.

The comfortable band is the same as anywhere else: roughly 130 to 150 words per minute, dropping toward 120 for material the audience has never heard before. Above about 170 people follow the words and stop retaining the point.

Rehearse the parts that actually fail

The first thirty seconds. This is where nerves are highest, where pace is least controlled, and where an audience decides how much attention to give the remaining nine minutes. It is worth rehearsing this section more times than the rest of the talk combined.

The transitions. Almost nobody rehearses the sentence that gets from section three to section four, which is why it is where fillers and lost pace concentrate. Write the transition sentences down and say them aloud on their own until they are automatic.

The ending. Talks that trail off do it because the speaker rehearsed the middle and improvised the last twenty seconds. Know your final sentence exactly, and rehearse stopping after it.

The questions afterwards. Q&A is unrehearsable as a script and very rehearsable as a format: answer in about thirty seconds, take a beat before you start, stop when you have made the point. Practising that shape matters more than predicting the questions.

Stage fright is physical, and it shows up in the numbers

Nerves do not usually announce themselves as fear. They announce themselves as a faster heart rate, shallower breathing and a shorter fuse for silence, and those three produce exactly the delivery pattern you can measure: pace up, pauses gone, word endings dropped.

This is useful because the symptom is more tractable than the cause. Telling yourself to be less nervous does nothing. Deliberately holding a one-second pause at every full stop does something immediately, and it works even while the nerves are still there.

The other half is exposure, and exposure requires repetition rather than intensity. Ten one-minute recordings across ten days moves the physical response more than one long panicked rehearsal the night before.

A free rehearsal schedule for a two-week runway

Days 1 to 3: record the opening ninety seconds only, once a day. Do not rehearse the whole talk yet. You are establishing pace and a baseline for the section that carries the most risk.

Days 4 to 8: run the talk in halves, timing each half. Fix the worst single number after each run - usually pace drift in the second half, occasionally a filler cluster at one specific transition.

Days 9 to 12: full run-throughs, timed, standing up, at the volume you will actually use. Volume changes pace, so a seated murmured rehearsal does not predict the real thing.

Days 13 and 14: one run in front of a person, and one recorded run of the opening and closing only. Do not add new material in the last two days; the marginal value of a better example is lower than the cost of rehearsing it twice.

Where a free tier will stop

Most free tiers cap either the length of a recording or the number per day. For a ten-minute talk this matters, and the workaround is to rehearse in sections rather than fighting the cap - which is better practice anyway, since section-level rehearsal is where the fixes happen.

Free tiers rarely include video, so eye contact, gesture and where you stand are outside what any of this measures. Those are real parts of public speaking and a phone camera propped on a shelf covers them adequately.

Nothing free will give you an audience, room acoustics, or the specific feeling of a microphone that is louder than you expected. Book five minutes in the actual room if you possibly can; it is worth more than another rehearsal.