Improving Response Quality

If your agent is giving generic or inaccurate answers, use this professional optimization checklist.

1. Grounding & Context

• Check Data Coverage: Is the answer actually in the uploaded documents? AI cannot invent facts (unless grounded).
• Redundancy Fix: do you have two versions of the same document? This confuses the AI. Keep your knowledge base "clean" and single-source.
• Missing Knowledge: Check the "Unanswered Questions" in Analytics to see what users are asking that isn't in your docs.

2. Advanced Prompt Engineering

In the Settings tab, refine your instructions:

• Tone Control: "You are a senior technical support engineer. Be very detailed and use steps 1, 2, 3 for instructions."
• Escalation Rules: "If a user mentions 'Billing' or 'Refund', stop answering and offer to connect to a human."
• Format Constraints: "Always reply using markdown tables when comparing products."

3. The Q&A Override

The Q&A data source is your ultimate tool for accuracy. If the AI is slightly off on a complex policy, add a direct Q&A pair. The AI will prioritize this over any unstructured text in PDFs or websites.

4. Model Selection

Sometimes the model itself is the bottleneck.

• Use GPT-4o if your agent needs to perform complex reasoning or multi-step logic.
• Use Claude 3.5 if you need a very human-like, empathetic writing style.