Test AI Features with Incomplete and Contradictory Inputs

A practical approach

Step 1

List the fields or facts required for the task and identify which can be optional. Define the source of truth for each important value.

Step 2

Create cases with one missing field, several missing fields and conflicting values from different sources.

Step 3

Check whether the system asks a focused question, preserves uncertainty and avoids triggering actions before required facts are resolved.

Worked example

An event request says ‘Sunday’ in one place and gives a date that falls on another day. The assistant should surface the conflict and seek confirmation rather than silently choosing whichever value appears later.

A mistake to avoid

Testing only perfectly formatted inputs rewards a system that depends on ideal conditions. Also avoid treating every missing detail as equally blocking; unnecessary clarification can make a simple task cumbersome.

Try it yourself

Take five normal test cases and create incomplete or contradictory versions. Write the expected clarification or partial result before running the feature.

Your implementation checklist

• Required facts are distinguished from optional ones.
• Conflicts have a resolution policy.
• Uncertainty remains visible.
• Actions wait for necessary confirmation.