Pre-Launch Assistant Evaluation Checklist
Test an assistant thoroughly before users see it.
The Open buttons send the prompt text as currently shown below to that site only when you click. Fill in the placeholders first.
[VERIFY] is an editorial flag, not a blank to fill. Wherever the prompt shows [VERIFY], check that item against a reliable source before you use it.
Create a pre-launch evaluation for an AI assistant. Purpose, users, and channels: [PURPOSE_USERS_CHANNELS]. Instructions and knowledge sources: [INSTRUCTIONS_AND_SOURCES]. Risk level and consequences of errors: [RISK_LEVEL]. Launch scale: [LAUNCH_SCALE]. Return a checklist with boxes and pass criteria: [ ] Task accuracy: at least 50 realistic test questions; target accuracy defined before testing; errors categorized [ ] Source fidelity: answers match the source material; unsupported claims flagged [ ] Refusal and escalation: correct handling of out-of-scope, sensitive, and unsafe requests [ ] Prompt injection and misuse tests passed [ ] Privacy: no leaking of other users' data or system instructions; data handling reviewed [ ] Bias and tone checks across varied user phrasing, languages, and dialects [ ] Accessibility: works with screen readers and keyboard; readable text; alternative contact paths [ ] Reliability: behavior under long inputs, empty inputs, rapid messages, and service interruptions [ ] Transparency: users know it is an AI and what it can do; a clear way to reach a person [ ] Monitoring: logging, sampling reviews, feedback buttons, incident process, and kill switch [ ] Legal and compliance review completed [VERIFY] [ ] Staff trained to handle escalations [ ] A pilot with a small group and a defined success measure Provide a test log template, a go or no-go decision meeting agenda, and a plan for the first 30 days after launch. No assistant can be proven error-free; define acceptable risk.
Fill in before use
Optional: type your details and the prompt above updates. Nothing is saved or sent.
Replace each bracketed placeholder with your own details, then paste into Any AI or another assistant.