Voice agents
Making a voice agent sound like your business
Vocabulary, tone and escalation manners for production voice agents
Context
A voice agent is often the first voice a new customer hears. Generic phrasing tells the caller they've reached software; familiar phrasing tells them they've reached you.
Sounding like the business is not about imitating a person — it's about using the words, prices and promises the business actually uses.
This is a writing and operations problem more than a technology one.
What changed with AI systems
Modern voice models handle interruption, accents and messy phrasing well enough that the script is now the differentiator.
Agents can draw live answers from the business's own systems instead of reciting a static FAQ.
Tone is configurable enough to be a brand decision — which means someone has to make it.
How to approach this in your organisation
- 01Start from transcripts of real calls: the questions, the phrasings, the answers your best person gives.
- 02Write the agent's vocabulary list — service names, area names, the things you never promise on a first call.
- 03Decide the tone deliberately and keep it consistent with the rest of the brand.
- 04Give the agent honest limits: what it says when it doesn't know, and how it offers a human.
- 05Read the transcripts weekly early on and feed corrections back into the prompt.
Key metrics
- Callers who complete their goal on the first call.
- Corrections per week found in transcript review.
- Escalations offered versus escalations demanded by frustrated callers.
- Repeat callers who choose the agent over waiting for a person.
Risks to consider
- An agent that overpromises because nobody wrote down what it must not say.
- Tone drift as prompts are patched ad hoc.
- Answers that go stale because they were written once and never reconnected to the source system.
- Optimising for sounding human instead of being useful.