Ready-to-use prompt templates for the most common call types. Every prompt is built on the proven structure — a short, natural flow, one question at a time, and a strict rule that the agent never stops on a bare “Okay, got it” but always moves the call forward. Click a card to see the full prompt, copy it, and paste it into your agent’s Agent Prompt box (or generate a fresh one with Auto Build, under Agent Setup → New agent).
Prompt templates
Pick the use case closest to your call, then tweak the bracketed parts (role, company, questions).
Agent welcome message formats
The welcome message is the first line the agent speaks when the call connects. A good opener follows a simple shape — greeting → your name → who you’re calling from → the reason for the call → a permission check like “is this a good time?”. Use ... for natural pauses and {name}, {role}, {company} variables that your batch CSV fills per call. Click a card to see the full opener and copy it into the Agent Welcome Message box.
Keypad (DTMF) menus
Not every call needs an AI conversation. A Keypad (DTMF) agent plays a short message and a menu, then acts on the key the person presses — confirm, snooze, transfer, opt out, rate, and more. These are ready-to-use menus: click one to see the welcome line, the spoken menu, and the exact key map to recreate in Agent Setup. For the full walkthrough see the Keypad (DTMF) guide.
Inserting {variables}
Type { or @ anywhere in the Agent Prompt or Welcome Message and a picker opens at your cursor listing every variable this workspace can actually fill. Choose one and it is inserted in the right {name} form — no need to remember the exact spelling.
The list is not hardcoded: it holds the built-in {name}, {phone} and {email}, plus every custom variable your workspace has declared. You can add a new one from the picker itself, which saves it to the workspace so your teammates’ agents offer it too. Once saved, the variable is filled from the matching column in your batch or campaign CSV.
Only offered names resolve. The call engine substitutes the variables it knows and leaves anything else exactly as typed — so a misspelled {Name} or an invented {city} is read out literally to the caller. That is precisely why the picker only lists names that will resolve: pick from it rather than typing braces by hand.
Variables are for the Agent Prompt, where the LLM reads them as context. Avoid them in the Welcome Message unless the call is driven by a CSV that carries that column for every row — an unfilled placeholder is spoken aloud as the raw text.
How to use a template
Click the card matching your call type and press Copy prompt.
Open your agent in Agent Setup and paste it into the Agent Prompt box.
Replace the bracketed parts — [ROLE], [COMPANY], the questions list — with your specifics.
Swap any per-candidate detail for a variable: type { and pick from the list.
Save the agent and run a test call — or a free Web Call — to confirm the flow feels natural.
Why these prompts don’t stall: a common mistake is telling the agent to reply in “exactly one sentence”, which forces it to choose between acknowledging and asking the next question — so it acknowledges and goes silent. These templates instead require every reply to end with a question or next step, so the agent always keeps the call moving.