AI Assistant Evaluation

"I need to evaluate different AI assistants (like ChatGPT, Claude, Gemini, etc.) for our [specific domain/purpose]. 

Our primary requirements are:
- [Requirement 1, e.g., accuracy in technical content]
- [Requirement 2, e.g., ability to follow complex instructions]
- [Requirement 3, e.g., consistency in output format]

Can you help me create a structured evaluation approach with:
1. Specific test prompts we should use
2. Evaluation criteria and scoring method
3. Important considerations or limitations to be aware of
4. Best practices for conducting a fair comparison"