Your artificial SI is loud. Fluency is not evidence.
Check AI answers before you trust them.
A polished answer is a starting point. Find the claim, open the source, and decide what the evidence actually supports.
For people using AI to research, write, and make everyday work decisions. Turn a convincing draft into a set of claims you can check.
A useful check has three parts
Separate facts, calculations, interpretations, and recommendations. Check the statements that could change your decision first.
A plausible citation is not enough. Open the source, find the relevant passage, and check dates, scope, and units.
Mark each claim supported, contradicted, or unresolved. Keep uncertainty visible instead of turning it into a confident conclusion.
Understand the problem
Why AI sounds confident even when it is wrong explains why fluent output and factual reliability can come apart.
Fluency versus evidence shows how a polished claim can feel stronger without gaining any support.
Try it on one sentence
“The pilot reduced processing time by 40%, so every team should adopt it.”
There are at least three things to check: the size of the reduction, what the pilot measured, and whether its results apply to every team. A true number does not automatically justify the recommendation.
Work through this example in the verification guide, then record your own claim in the worksheet.
Part of a wider practice
Use these connected resources for different parts of the decision:
- Prompt Engineering — frame the question and structure the draft.
- Wiry — reflect on the signal you feel before acting. A feeling is a cue to investigate, not factual proof.
- Surer — consider what evidence your decision needs.
- Intuition Management — explore judgment, intuition, and calibration.