Guardrails are the rules that keep your agent behaving the way you want. They tell it what never to say, how to carry itself, and when to walk away from a prospect. Setting a few good guardrails is what makes an agent feel safe to let loose on your DMs.
The four kinds of guardrails
All four sections are optional. You add each rule as its own short line (press Enter to add), up to 25 rules of 300 characters each per section.
- Do-Nots: things the agent must never say or do. For a coach, that might be "Never promise specific weight-loss numbers." Cloziq checks every reply the agent writes against your do-nots and fixes a reply before it sends rather than letting a violation through.
- Conversation Rules: guidelines for how the agent talks, qualifies, and responds, like staying encouraging or always answering a question before asking the next one.
- Disqualification Rules: the reasons to disqualify a prospect, so the agent knows when someone is not a fit and stops pitching. A rule here ends the conversation quietly, with no goodbye message. If you want the agent to say something as it closes the door, do it from the Sales Flow instead: a Flow Action set to Change Lead State to Disqualified can carry an optional closing message. See Building your sales flow.
- Prohibited Characters: specific characters the agent's replies must never contain, up to 25. Cloziq starts you with a sensible default set (the em dash, the en dash, and the semicolon), and you can add your own. Spaces, periods, and commas cannot be prohibited, since the agent needs them to write around everything else. Neither can letters or numbers; try and publishing stops with "Cannot prohibit letters or numbers."
Templates leave placeholders here too
If you started from a template, some guardrail lines may still carry bracketed placeholders like [OWNER NAME]. Publishing checks your guardrails along with everything else and stops until each one is replaced, telling you which placeholder it found and where. See Create your first Sales Agent.
Every rule has to be specific enough to act on
Publishing now checks that each guardrail line actually tells the agent something. A rule of one or two words, or a fragment with no instruction in it, is rejected with:
"Your rule" in the do-not list is too vague for the agent to act on. Say what the agent must do or avoid, and when, for example "Never quote a price before the qualification questions are answered".
Cloziq names the offending rule and the list it came from, so you know exactly which line to fix.
The fix is almost always the same: say what and say when.
- "Be careful" gives the agent nothing. "Never promise a specific timeline for results" does.
- "Pricing" is a topic, not a rule. "Never quote a price before the qualification questions are answered" is a rule.
- "Professional" is a tone, and tone already has its own setting. Use guardrails for the boundaries tone cannot express.
This applies to Do-Nots, Conversation Rules, and Disqualification Rules alike. It is worth doing even where it is not enforced, because a rule the agent cannot act on is a rule that does nothing.
What is already built in
Some protections are always on and need no configuration. The agent never invents facts about your offer; if it does not know something, it says so instead of guessing. It never claims to have personal experiences or feelings. It never pushes a checkout or booking link before the prospect has been qualified. And when a prospect is abusive or clearly wasting time, the agent disengages on its own.
Tips
- Write do-nots as plain, concrete statements. "Never give medical advice" beats a long paragraph.
- A couple of strong disqualification rules save you time by filtering out people your agent should not chase.
- Guardrails work alongside your Sales Flow, not instead of it. The flow drives the conversation; the guardrails set the boundaries.
- If Cloziq notices one of your rules keeps misfiring on a live agent, it sends you an alert so you can reword the rule. See Turning on notifications.