DoneHat methodology — sample, inspect, compare
AI answers vary across prompts, time, source availability, and model behavior. DoneHat uses sampled buyer questions and saved receipts to make each finding inspectable.
Question selection
Combine category discovery, buyer tasks, comparison, trust, and implementation questions. Unbranded questions test whether the brand appears without being named.
Answer evidence
Record attempts, usable answers, brand mentions, cited sources when available, and provider failures. Small samples are shown with uncertainty rather than certainty.
Before and after
A recheck uses comparable questions and records the changed answer. The result is directional evidence, not a causal guarantee.