Map the symptom to the setting
| What you're hearing | Lever | Where |
|---|---|---|
| Wrong gender, accent, or general vibe | Voice Selection | Settings → AI Voice → Voice Selection (12 voices, click to switch — autosaves) |
| Too fast — patients ask the AI to slow down | Speaking Speed ↓ (try 0.9 or 0.8) | Settings → AI Voice → Personality & Behavior |
| Too slow — feels sluggish | Speaking Speed ↑ (try 1.1 or 1.2) | Same place |
| Too stiff/corporate | Conversation Tone → Friendly & Warm or Casual & Relaxed | Same place |
| Too casual for your specialty | Conversation Tone → Professional & Polished or Formal & Business-like | Same place |
| Opening line is generic | Custom Greeting | Settings → AI Voice → Custom Greeting (500-char textarea) |
| Doesn't mention your services/parking/policies | Business Knowledge field | Settings → AI Voice → Business Knowledge (2,000-char textarea) — but for substantial info use the Knowledge Base instead |
All of these are on the same page: Settings → AI Voice → Settings tab.
Speaking speed: how to pick a number
- 0.8–0.9 — older patient populations, complex medical info, callers who often ask the AI to repeat itself.
- 1.0 — default. Comfortable for most callers.
- 1.1–1.2 — high-volume practices where shaving 5–10 seconds off every call meaningfully changes throughput. Don't go higher than 1.2 in clinical settings — patients miss details.
Change it, then test by calling your number. Adjust by 0.1 at a time.
Tone: what actually changes
The tone setting doesn't change the voice — it changes word choice and sentence structure across every reply the AI generates.
- Friendly & Warm — uses contractions, more "happy to help" phrasing, a little softer on declines.
- Professional & Polished — neutral, business-clinical, fewer fillers. Good default for medical/dental.
- Casual & Relaxed — most informal. Fits spas, wellness, fitness.
- Formal & Business-like — full sentences, no contractions, "thank you for calling" structure. Good for legal/specialty.
Things you can't tune yourself (yet)
The following knobs exist on our side but aren't exposed in the user UI:
- Interruption sensitivity — how many words the caller has to say before the AI stops talking (the "stop speaking" plan). Currently set to 2 words globally.
- Endpoint detection wait time — how long the AI waits after a pause to be sure the caller is done. Currently smart-tuned per language.
- Silence timeout — how long before the AI re-prompts a silent caller. Default 30 seconds.
- Background noise denoising — on by default.
If patients consistently report the AI cutting them off, dragging silence, or background noise issues, open a support ticket with one or two affected call IDs. We can adjust the speaking plans for your practice — these are intentionally not user-controls because bad settings break calls.
After you change something — test it
A 30-second test call is worth more than 10 minutes of guessing. Call your number, listen for the specific thing you just changed, and decide if it's better. See Test call walkthrough for the four scenarios we recommend.