A system prompt reads best when it covers four things, written in plain sentences, the same way you'd brief a person filling in for your front desk:
- Who they are and who they represent (name, business, role on the call)
- How they should sound (formal, casual, warm, brief, chatty, and so on)
- The steps of a typical call, in order (greet, find out what the caller needs, offer to book, confirm details, close)
- What to avoid (specific words or phrases, promises the assistant can't back up, anything that should go to a human instead)
Write it as full sentences rather than scattered keywords. Your assistant follows a system prompt roughly the way a new hire follows a training document, so instructions that read clearly to a person tend to produce a call that sounds clear to a caller too. If you also want shorter or longer replies, say so directly in this text, for example “keep answers to one or two sentences unless the caller asks for more detail,” rather than looking for a separate setting. Response length in Sulus isn't a toggle you flip; it comes from how you word the prompt itself.
Advanced & Optional
Just above System Messages in that same AI Model area, you'll find Model Provider and Model dropdowns. These choose which underlying AI model actually powers your assistant's thinking, not the wording of your prompt. Today, OpenAI's models are the only ones available to select, shown as the current options in the Model dropdown; other providers listed there are marked Upcoming and aren't active yet. This setting mostly affects cost and response quality rather than tone or content, so most businesses never need to touch it. Leave it on the default unless your Sulus representative or support recommends a change.