I am lost in trying to understand the results I get from my assistant. I am calling it from Make.com with a URL source from Science Daily. I get the expected result from the assistant of a rewritten article when I test in the playground. But, not when I call the assistant (or chat completion with the identical prompt), from Make, or Zapier. From there, I get a totally unrelated article from the source.
I revised the assistant in early playground testing where I had an “Example Output” of a related article in the prompt. But, that was consistently interpreted as the article to be rewritten so I deleted that part of the prompt. The “ghost article” output I now get is thematically related to those early playground tests.
As a final test, I used the same prompt with Claude and got a rewritten article from the URL source both in the Make flow and the Claude playground.
I have no idea where to look for these ghost articles to delete them from my assistant. Would someone please offer some help here?