The AI qualifier reads each inbound WhatsApp message against your playbook: a persona, your qualification criteria, and a numbered list of questions. It replies with the next relevant question, extracts answers into structured fields on the lead, and marks the conversation qualified, disqualified, or still pending, so your team only steps in when a human is actually needed.
Every reply is built from one playbook: the persona you wrote (how it should sound), your qualification criteria (what makes a lead a good fit), and your list of questions. The AI keeps that brief consistent across the whole conversation and asks whichever question the customer hasn't answered yet. As the customer replies, their answers are extracted into structured fields and merged straight into the lead's record, so your CRM view fills in as the conversation goes.
Voice notes sent by a customer are transcribed automatically and read the same way as a typed message, so a customer can send a voice note and the AI still picks up the conversation from there. If a playbook has image reading turned on, the AI can also look at photos a customer sends, such as a bill or a receipt, and use what it reads there to answer your qualification questions. Only the most recent images in a conversation are considered, and very large image files are skipped.
The AI waits a few seconds after a customer's message before replying. That short pause means if someone sends two or three messages back to back, the AI answers once, covering all of them, instead of replying after every line.
It also stays quiet, permanently for that turn, the moment a human gets there first: if you or a teammate reply to the conversation, the AI stands down instead of sending its own reply on top of yours.
If a customer asks something outside the playbook, like negotiating a price or asking about terms and refunds, the AI does not guess an answer. It hands the conversation to a human instead. See AI and human handoff for how that handoff works and how to take a conversation back yourself.
Each turn ends in one of three states: still pending (more information needed), qualified, or disqualified. A qualified or disqualified decision can move the lead's stage and apply tags, based on what you set up in the playbook, so your pipeline reflects the outcome without anyone touching it by hand. See Writing your qualification playbook for how to set those outcomes up, and Conversation insights for a quick AI summary of any thread, qualified or not.
Every AI reply is metered like the rest of your AI usage on the account. Current plan allowances and top-up options are on the pricing page.