Scenarios are great for structured, repeatable flows. But real conversations don't always follow a script. AI actions let you add flexibility at the exact points where contacts go off-path, without rebuilding the whole flow.

This article shows patterns for combining AI with your existing scenarios. For the full detail on each AI action, see AI Smart Actions in Scenarios.

⚠️ AI actions need Pro and AI features, and each run uses AI credits. When your credits run out, the actions are skipped and the flow continues.

Keyword conditions handle simple yes/no choices. They break down when contacts express the same intent in a dozen different ways.

After an open-ended question, add an AI-detected condition with a plain-language rule like "the contact is interested in buying" or "the contact is asking about the free plan." It reads the full conversation, not only the last message, so it can judge intent even when the reply was short or phrased oddly. Route the true branch to a conversion step and the false branch to more nurturing.

The most common AI pattern is "structured start, open handoff." Your scenario handles the predictable part (the first reply, qualifying the contact, collecting details), then hands the open conversation to the AI Agent.

Use Hand over to AI Agent as the last action in your flow, with a short, factual instruction.

Keep the instruction to a sentence or two: "Contact confirmed interest in 1-on-1 coaching. Goal is to book a discovery call." That's more useful than a paragraph.

Not every message needs to be fixed text. Some moments call for a reply that responds to what the contact said in the previous step.

After a step where the contact says something variable, add an AI Agent message action with a clear instruction like "Acknowledge their answer and move to the next step." Unlike a full handover, the scenario keeps running and the agent handles only that one message.

If your scenarios trigger from public content like comments or open DMs, filter at the entry point so spam and off-topic messages don't consume your runs.

Use the AI-detected intent trigger filter to fire the scenario only on messages that match a specific intent, and keep exclude hateful and negative comments on for comment triggers. That keeps junk out before any step runs. See Using AI to Filter Scenario Triggers.

An AI step judges the conversation as it stands now. It reads the recent messages with their dates, and it is told which message is the latest one, so a phrase from an exchange days ago is treated as background rather than as the thing it is deciding on. Messages with no text, like a story mention, are left out.

When you use Test scenario on a contact you have tested with before, the step only sees messages from that test run. Each test starts clean, so an answer you gave in yesterday's test can't send today's run down the wrong branch.

⚠️ There is no way to clear a contact's real message history, and testing repeatedly on the same contact is fine. If you want a genuinely fresh conversation, test with a contact who has never messaged you.

AI and rule-based logic solve different problems. Knowing which to reach for saves debugging later.

Use AI actions when:

Stick with rule-based logic (see Flow Control) when:

For the full reference on every AI action and when to use each, read AI Smart Actions in Scenarios. To control which messages start your scenario in the first place, read Using AI to Filter Scenario Triggers.