Summerizing using API vs response from Chatbot (OpenAI web site)

I tried simple and suggested by Chatbot Python code using API to summerize some text from CNN. I noticed website gives pretty good summary. My code - that is suggested by Chatbot -produces really bad and random summary. There is not much to tune other than max_tokens and specifying engine=“davinci”, Any idea why I maybe getting such bad results ? I want to establish baseline before fine tuning.