Inconsistent accuracy in responses to requests for information

I gather information about Austrlian companies from Chatgpt. Some of the information is obtained from public government web site. The information is obtained manually by entering the company name. ChatGPT knows how do this. And can tell you the process in great detail. However it consistently fails to return the correct information. When challenged. It tries again eventually finds the correct value after two or three events. It then promises to be accurate in the future and it has learned how to it properly. However it continues to fail subsequently has anyone had this experience and is there a way to stop this behaviour?