Start with the right question
The most useful follow-up to a confident AI answer is often a request to identify what would change it. This turns a broad prediction into a list of facts and legal questions that can be checked. It does not make the model a qualified decision-maker.
Work through the checks
List the supplied facts
Write down only the facts actually provided. Compare them with the answer and mark additions. A model may silently assume insurance, a signed contract, or a particular state when none was specified.
Challenge the number
Ask what data and method support a percentage or dollar range. If there is no traceable method, label it unsupported. Do not repeat a number because it appears precise or reassuring.
Identify dependencies
Ask which documents, dates, and legal rules could change the answer. Separate factual uncertainty from legal uncertainty. They require different follow-up work and cannot both be resolved by rephrasing a prompt.
Seek a counterexample
Ask for an alternative explanation consistent with the same facts. Compare both with the records. The goal is to find missing information, not to select the answer that best supports your preferred outcome.
A record you can use
- Claim made by the answer
- Fact actually supplied
- Assumption added
- Evidence needed
- Professional question
Keep this record locally. Use document references rather than publishing account numbers or other people's personal details. An empty field means the information is still missing; it should not be filled with an AI guess.
Worked example
A fictional chatbot predicts a strong claim because it assumes a delivery guarantee. The actual receipt contains only an estimated date. Once that assumption is removed, the useful output becomes a question about the agreed delivery terms and documented loss. The original confidence percentage is discarded.
This example is fictional and illustrates the method, not a real case result.
Limits and next steps
Repeated agreement from several models is not independent legal verification. Models can repeat common mistakes or share training patterns. Important decisions should rest on reliable records and qualified advice.
Turn the uncertainty list into a preparation checklist. The app can help organize the questions, but it should not be treated as establishing liability, deadlines, or a guaranteed outcome.