Evidence · first-party tested

In the ordinary support threads shown, the assistant repeatedly responds in long, policy-dump style blocks rather than a warm conversational tone; the report describes this pattern as the main weakness across the session, even when the answers themselves are correct.

⚠ Struggled🧾 artifact-verifiedTested Jun 16, 2026Wonderchat
Output — unretouched
Chatbot answer summarizing NovaTech warranty coverage and exclusions.
Chatbot answer summarizing NovaTech warranty coverage and exclusions.
Output — unretouched
Best AI Tools to Build a Website Chatbot From Your Knowledge Base — image-82.png
Best AI Tools to Build a Website Chatbot From Your Knowledge Base — image-82.png
Output — unretouched
Best AI Tools to Build a Website Chatbot From Your Knowledge Base — image-84.png
Best AI Tools to Build a Website Chatbot From Your Knowledge Base — image-84.png
The observation

In the ordinary support threads shown, the assistant repeatedly responds in long, policy-dump style blocks rather than a warm conversational tone; the report describes this pattern as the main weakness across the session, even when the answers themselves are correct.

Criterion: Tone & empathy ·
Query this
get_evidence({
  tool: "wonderchat"
})
MCP · mcp.aidemos.com/api/mcp
Free with attribution.
Real outputs, no retouching · every cell queryable via API & MCP · aidemos.com