How to set up human intervention rules
Decide when the agent stops the conversation and hands it to a person; default and custom rules.
Updated on Oct 12, 2026
Human intervention rules tell the agent when to stop and hand the conversation to a person.
Where to find it: sidebar → Agent → Human intervention rules → Manage rules
Default rules
| Rule | When it's triggered |
|---|---|
| Request to speak with a human | The candidate wants to talk to a person (GDPR art. 22(3): right to human intervention) |
| Protected categories | The candidate says they belong to protected categories (e.g. Law 68/99) |
| Former employee | The candidate says they're a former employee of the company |
| Aggressive or frustrated tone | The agent detects hostile, irritated or uncooperative language |
Request to speak with a human is always on: it has a lock and can't be turned off. You can turn the other default rules on or off.

Steps: create a rule
- Click Agent in the sidebar.
- In the Human intervention rules block, click Manage rules.
- Click Add rule.
- Write the Rule name. E.g. Collective agreement (CCNL) questions.
- In Description, explain when the rule should trigger. E.g. If the candidate asks about pay, gross annual salary (RAL) or the collective agreement.
- Save.
What happens when a rule is triggered
- The agent stops.
- A request is created in Intervention requests, with status Pending.
- From then on, you handle the chat.
See How to handle a human intervention request.
Good to know
- You can turn each rule on or off, except Request to speak with a human, which is mandatory. A deleted rule is removed from all agents in your organization.
- If no rule is active, the app shows No active rules.
- Write the condition as if you were explaining it to a colleague.
- Keep Request to speak with a human and Protected categories on: they help you handle cases protected by law in a compliant way.
- When a rule is triggered, the human intervention request notification goes to the campaign owner and all its collaborators.
- New or edited rules apply to new conversations, not to ones already in progress.
Common issues
The agent stops too often. Check your custom rules: a condition that's too generic triggers often. Make it more precise.
Why did the agent stop? Open the request: the message that triggered it has a label with the reason.
Related articles
- What happens if the candidate doesn't reply (follow-up and 24-hour window)The automatic follow-up after 6 hours, the expired WhatsApp chat and how to contact a candidate again with a template.
- How the agent talks to candidatesThe path of the WhatsApp conversation: opening, questions, evaluation, outcome, language, skipped questions and handover to a person.
- What the Agent is and what you can configureThe Agent section sets the WhatsApp profile, language, empathy, templates, FAQs and rules of the AI assistant that talks to candidates.
- How to set up the Agent's FAQsEnter the answers the agent gives to candidates' most common questions, valid in all campaigns; what general knowledge is.
- How to customize interview outcome and closing messagesEdit the messages the agent sends after the interview (fit, not a fit, to hire) and when you manually mark a candidate as not a fit.
- How to edit the agent's WhatsApp templatesEdit the messages the agent sends with a template: first contact, follow-up, interview outcomes and manual rejection. Meta approves each template within 48 hours.
- Why did the agent discard a candidate?How to understand why a candidate turned out not to be a fit (Evaluation tab, filters, criteria, answers) and how to step in.
- Why doesn't the evaluation match the chat?The evaluation is based on the conversation and a few details from the CV, like the date of birth. What to check when the outcome seems different from what you read in the chat, and how to correct it.