Most people aren't bad at talking. They're bad at two things.
Everything else works fine. This finds which two, and then makes you practise them out loud for twenty-eight days — measuring what actually comes out of your mouth rather than asking how you felt it went.
Um, so basically I kind of look after, you know, the numbers side of things, and I think it's sort of going okay, yeah.
27 words · 14.8 fillers per 100 · 6 qualifiersI look after the numbers. It's going well.
8 words · 0 fillers · 0 qualifiersThe roleplay
Rehearse the actual conversation. An AI character plays your manager, the interviewer, the date — stays in role, pushes back, and won't coach you mid-scene. When you end it, it breaks character and quotes what you said.
Start a scene → Two minutesThe speech report
Pick what you're practising, get a scenario, talk for forty-five seconds. Your pauses and trailing sentences get marked on your own waveform — tap one and hear the exact moment.
Try it now → Start hereThe assessment
Twenty questions across five things: clarity, presence, initiation, flow and recovery. It scores each one and tells you which two are costing you the most — then builds a four-week plan around those and deliberately ignores the rest.
Take it → Then this, dailyThe 28 days
One scenario a day, half work and half social. You say your answer out loud, it listens, and it hands back real numbers: filler rate, pace, hedging, sentence length. Plus one thing to change — one, never a list.
Open Day 1 →Why measure it instead of just practising?
Because you can't hear yourself. The gap between how a conversation felt and how it actually sounded is where most people get stuck for years, and the only way across it is a number you didn't choose.
- Filler rateEvery um, basically, you know, like counted, per hundred words. Above five and people hear it. This is the number that moves fastest with practice, which is why it's the one to attack first.
- PaceWords per minute. 110–155 reads as composed. Over 180 and you're sprinting — which is the loudest, most reliable signal of nerves there is, and the easiest to fix on purpose.
- HedgingI think, maybe, probably, sort of. Each one quietly withdraws the sentence it's sitting in. Most people have no idea how many they use until they're counted.
- Sentence lengthLong sentences are where listeners lose the thread and where fillers breed. Over 26 words on average and you're asking too much of whoever's listening.
- One fixYou get told a single thing to change, chosen from whichever measure is worst. A list of eight improvements is how people end up making none.
What this is, and what it isn't
What it does
- Counts what you actually said, out loud, and shows you the numbers
- Gives you a structured curriculum instead of vague advice to "be more confident"
- Runs entirely in your browser — no account, no upload, no cost
- Tracks whether you're improving across the four weeks
What it doesn't
- Judge whether you were interesting, warm, or worth talking to — it counts words, not connection
- Replace doing this with actual people, which is the only thing that finally works
- Diagnose anything. It's a practice tool, not an assessment of you
- Substitute for therapy if anxiety is the real obstacle