Ask a model to synthesise research and it will produce something that reads like a finding whether or not the data supports it. Patterns come back stronger than the evidence warrants, paraphrases appear inside quotation marks, and a theme two participants mentioned becomes what "users" want. None of this is announced, the output looks the same either way. This prompt is a gate rather than a generator: it walks each claim in a synthesis back to the source material and asks what actually supports it.
For that to mean anything you have to supply the original data, not a summary of it. Validating a synthesis against another synthesis just launders the same errors. Expect the check to be genuinely useful on quote accuracy and pattern strength, and less dependable on the inference step, where it is evaluating reasoning of exactly the kind it produces itself. Treat a clean pass as a reason to look harder at the conclusions rather than a reason to stop looking, and hand-check a few claims, some flagged, some not, before any of it reaches a stakeholder.
The check runs against the source data or it isn't a check. A validation pass on a summary tells you the summary is internally consistent, which was never the question.
Fill in your own details below; the prompt updates as you type. When it's ready, copy it into Claude or whatever AI tool you use.
Validate AI-assisted research synthesis against original data, checking for hallucinations, misinterpretations, and unsupported conclusions before using findings for IA decisions.