> ## Documentation Index
> Fetch the complete documentation index at: https://fliqr.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Keywords and AI response priority

> Understand how Fliqr AI chooses between keywords, AI Agents, and fallback Flows when a contact sends a message.

Not every message should go straight to an LLM. Keywords catch exact intents; AI Agents handle the rest; a fallback Flow covers failures.

## Keywords vs AI Agents

| Approach          | Best for                                               |
| ----------------- | ------------------------------------------------------ |
| **Keywords**      | Exact menus (`price`, `human`, `stop`), campaign codes |
| **AI Agent**      | Natural questions about your business                  |
| **Fallback Flow** | Last resort when models/tools fail                     |

For modern support/FAQ, prefer an [AI Agent](/docs/core-concepts/ai-agents/create-agent). Keep keywords for precise overrides.

## Set up a keyword

<Steps>
  <Step title="Open Keywords">
    Go to **AI Center** and open **Keywords** (or the Keywords entry in Automation, depending on your workspace layout).
  </Step>

  <Step title="Add a keyword">
    Click **Add**, enter the phrase/match mode, and select the Flow or action to run.
  </Step>

  <Step title="Order rules carefully">
    More specific keywords should win over broad ones. See also [Automation Rules](/docs/core-concepts/automation-rules).
  </Step>
</Steps>

## Response priority

When a contact messages you, Fliqr AI typically resolves a reply in this order until one succeeds:

1. **Keywords**
2. **OpenAI** (if configured as an active agent/provider path)
3. **Claude**
4. **Google Gemini / xAI**
5. **DeepSeek**
6. **Fallback Flow**

Exact provider order can depend on which integrations and default agents you enabled — treat this as the platform’s fail-through model: keywords first, then AI providers, then fallback.

<Note>
  If you want a single Agent to own Default Reply, configure that Agent explicitly and keep keywords only for hard overrides (opt-out, legal, human).
</Note>

## Practical recommendations

* Use keywords for compliance phrases (`stop`, `delete my data`)
* Use an Agent + knowledge for product Q\&A
* Always define a Fallback Flow that apologizes and offers handover
* Test priority with messages that match a keyword **and** would also be answerable by AI

## Next

<CardGroup cols={2}>
  <Card title="Automation Rules" icon="bolt" href="/docs/core-concepts/automation-rules">
    Full trigger model for starting Flows.
  </Card>

  <Card title="Human handover" icon="headset" href="/docs/core-concepts/ai-agents/human-handover">
    Make “talk to a human” a safe keyword or AI action.
  </Card>
</CardGroup>