Before you trust, buy, deploy or automate with AI, ask ten questions. The Q Test is a practical starting point—not a certification score. Its purpose is to expose missing evidence, unclear authority, hidden dependencies and recovery gaps before they become consequences.
Name the task, user, environment and expected outcome. “Use AI” is not a testable objective.
Separate demos, marketing claims, benchmark results, real-world tests and independent evidence. Record what is still unknown.
Trace prompts, files, identifiers, logs, connected systems, training use, retention and deletion.
Capability does not create permission. Identify the human or institutional authority, approval gates and actions that must never be automatic.
List tools, APIs, accounts, files, money, communications, physical systems and downstream automations within reach.
Test ordinary mistakes, stale context, hallucinations, adversarial inputs, drift, tool failures, loops, privilege problems and silent failure.
Consider sensitive data, identity, permissions, secrets, prompt injection, supply-chain dependencies and incident response.
Include speed, accuracy, privacy, cost, energy, water, compute, labor, review time, attention, accessibility and trust where material.
Define stop conditions, containment, evidence preservation, rollback, backup, continuity and what happens if a provider changes or disappears.
Models, vendors, data, laws, prices and risks change. Set a TTL or event trigger for reassessment rather than treating one review as permanent.
The answer is supported well enough for the current scope. Preserve the source and date.
Do not convert a missing answer into an assumption. Find evidence or narrow the claim.
The decision or action is consequential enough that explicit human authority is required.
A required safety, authority, evidence or recovery condition is unresolved. Do not promote merely to keep momentum.
The Q Test deliberately does not assign a guilt, fraud, liability or universal “safe/unsafe” score. Different uses carry different consequences; the evidence and authority must match the actual situation.
Verify an AI answer · Test an AI agent · Evaluate an AI vendor · Privacy & data · EE/EH impact · Evidence & status
Request the Q Test as a download, review or customized assessment →