Trigger Rules

Trigger rules define when your agent responds and what it says. Writing good trigger rules is the primary way to make an agent useful. This article explains how they work and how to write them well.

→ Open Agent Builder →

What a trigger rule contains#

Each trigger rule has five parts:

| Field | What it does | |---|---| | Label | A short name for this rule, used in event logs and summaries (e.g. "Competitor mentioned") | | Description | Plain English describing the condition — what the agent listens for (e.g. "When a competitor product or company is named in the conversation") | | Response instruction | What the agent should say or do when triggered (e.g. "Provide two key differentiators between our product and theirs, in two sentences") | | Action type | Speak (live agents) or Silent (send to back-channel) | | Hand-Raise | Whether the agent should request the floor before speaking (Live agents only) |

How trigger evaluation works#

When your agent is in a live meeting, CoAgentor continuously evaluates the transcript against your trigger rules. Here's the process:

1. New transcript segments arrive from the real-time transcription engine
2. The agent evaluates the growing transcript buffer against each enabled rule
3. If a trigger condition is met (assessed by the AI against your description), the rule fires
4. The agent generates a contextual response following your instruction
5. The response is delivered — spoken aloud or sent to your back-channel

Evaluation happens on a rolling window of the transcript. This means the agent always has recent conversational context, not just isolated sentences.

Writing effective descriptions#

The description is the most important part of a trigger rule. It needs to be specific enough to avoid false positives, but broad enough to catch all the relevant variations.

Too vague:

This may trigger constantly — pricing comes up in almost every sales call.

Too specific:

This is unlikely to ever trigger exactly as written.

Well calibrated:

This catches the natural range of how the topic actually comes up.

Writing effective response instructions#

The response instruction tells the agent what to produce, not just that it should respond. The more specific, the better.

Weak instruction:

Strong instruction:

The agent is running in real time during a live conversation — concise, actionable instructions produce better outputs than open-ended ones.

Sensitivity and trigger tuning#

Every agent has a Sensitivity setting (Low / Medium / High) that affects how readily rules fire. This applies to all rules on the agent.

• Low — requires a very clear, direct match. Good for agents where false positives are disruptive (e.g. live agents in formal meetings).
• Medium — balanced. Start here.
• High — triggers on loosely related conversational signals. Good for research or monitoring agents where you want broad coverage.

If a specific rule is triggering too often, make its description more specific. If it's not triggering enough, broaden the description.

Action types per rule#

Each rule can have its own action type, independent of the agent's default mode:

| Action type | What happens | |---|---| | Speak | Agent speaks the response aloud (requires Live mode agent) | | Silent | Response is sent to the configured back-channel |

This lets you build nuanced agents — for example, an agent that routes low-priority observations silently to Slack, but speaks aloud only for high-priority triggers.

Enabling and disabling individual rules#

Each rule has an enable/disable toggle. You can turn off a specific rule without deleting it — useful for seasonal rules ("when end-of-quarter deadline is mentioned") or rules you want to A/B test across different meetings.

Rule ordering#

Rules are evaluated independently and in parallel — the order in the form doesn't affect evaluation priority. Multiple rules can fire in the same transcript window.

Example trigger rules by use case#

Sales agent#

| Label | Description | Instruction | |---|---|---| | Competitor mentioned | Prospect names a competing product or company | Summarise our two strongest differentiators vs this competitor. Be factual. | | Price objection | Prospect raises a concern about cost, price, or budget | Acknowledge the concern, pivot to value: explain ROI, mention payment flexibility. 3 sentences max. | | Next steps | Conversation turns to next steps, follow-up, or closing | Suggest three specific next steps: a follow-up call, a product demo, and sending a proposal. |

Compliance agent (silent)#

| Label | Description | Instruction | |---|---|---| | Regulatory topic raised | Any mention of regulation, compliance, legal requirements, or specific regulatory body names | Flag this segment for compliance review. Summarise the exact regulatory topic mentioned. | | Commitment made | The sales rep makes a promise, commitment, or guarantee about the product | Record the exact commitment made, word for word if possible. |

Customer support agent#

| Label | Description | Instruction | |---|---|---| | Known error described | Customer describes a symptom matching a known bug or common issue | Provide the current workaround for this issue and the expected fix timeline if known. | | Escalation language | Customer expresses frustration, mentions cancellation, or asks for a manager | Acknowledge the frustration, summarise the issue, and provide the internal escalation path. |

Rule limits by plan#

→ View full plan comparison →

Tips and common mistakes#

✅ Write descriptions in the third person — "When the prospect mentions X" rather than "If they say X"

✅ Test with real transcript samples — paste snippets from past meetings into the description field when tuning

✅ Keep response instructions under 100 words — longer instructions don't always produce better responses

❌ Don't make instructions that require the agent to ask questions — the agent speaks in response to a trigger, not as part of a dialogue flow

❌ Don't overlap descriptions too heavily — if two rules describe very similar conditions, they'll often both fire for the same moment, doubling up your back-channel messages

❌ Don't set high sensitivity on a live agent with many broad rules — this will result in constant interruptions

→ Back to agent setup →