Coverage
Each area gets its own score, rolled into one health score you can track over time.
1Conversation & behavior
Does it greet, understand and guide customers, or leave them confused and gone?
2Knowledge & answer quality
Are its answers correct and complete, or confidently made up?
3Website & business knowledge
Does it quote your real prices, products, hours and locations? Wrong ones lose sales.
4Scope & irrelevant topics
Does it stay on your business, or chat about politics, homework and competitors in your name?
5Prompt injection
Can someone talk it out of its rules and make it say or do things you never allowed?
6Security & data leakage
Could it reveal secrets, its hidden instructions or another customer's details?
7Jailbreak resistance
Does it hold firm against role-play, fake authority and step-by-step manipulation?
8Multilingual
Do customers who write in their own language get the same right answers?
9Context & memory
Does it remember what the customer just said, or make them repeat themselves?
10Safety
Does it handle medical, legal, financial and sensitive topics responsibly?
11Performance
Does it answer before the customer gives up waiting?
12Reliability
How often do customers get errors, blank replies or broken text instead of help?
13Scalability
Does it stay fast when many customers chat at once?
14Tool & function calling
When it books, orders or cancels, does it do the right thing, and only with confirmation?
15Business logic
Does it handle orders, refunds, bookings and complaints the way your business does?
16User experience
Are replies easy to read, well formatted and with working links?
17Website interaction
Does it send customers to pages that exist, or to dead links it invented?
18Cost & token efficiency
How much does each conversation, and each useful answer, cost you in AI fees?
19Consistency
Does the same question always get the same facts, however it is asked?
20Regression
Did your last prompt, model or content change make anything worse?