To make ChatGPT fact-check answers and provide reliable sources, users should follow a structured workflow rather than simply asking “fact-check this.” OpenAI advises treating ChatGPT as a starting point, noting that confidence is not the same as reliability, and recommends checking important quotes, figures, technical claims, and references directly. The best tool depends on the task: ChatGPT Search for current facts, Deep research for complex questions requiring multiple sources, uploading the original file for claims in a PDF or document, Data analysis for calculations based on supplied numbers, and Standard chat for creative writing or opinion. For current facts, users should manually select Search from the tools menu; a searched response normally contains inline citations and a Sources section. A recommended fact-checking prompt instructs ChatGPT to break text into separate verifiable claims, label each as Supported, Misleading, Unverified, or False, cite a source directly beside every verdict, prioritize primary sources, include the source’s publication or update date, explain disagreements if reliable sources disagree, and finish with a corrected version. Users should add a location and “as of” date when the answer could vary by country or time. Asking for a claim-by-claim evidence table with columns for Claim, Verdict, Evidence, Source and date, and Correction prevents one correct source from hiding small inaccuracies.