Problem with structured outputs | OpenAI | Page 1

Any idea why OpenAI's API gives much more accurate data when doing a regular query with gpt-4o than when using structured outputs in JSON? I'm writing a program that runs through texts and analyzes that person's political opinions

For whatever reason, it incorrectly thinks that everyone is a leftist when I run it through structured outputs, and it predicts political views much more accurately when I ask it to output in plain text. I have tried to change this by setting the temperature higher, but still no fix.

For the following analysis: please be honest/blunt about everything. Give me a detailed/comprehensive analysis of this person's ideological views and personality with multiple points, based off of their messages from Discord. Output sending a list of political ideas, ideologies, and personality traits. Note that these messages were made in conversation, so they can include a lot of unnecessary data/banter. While writing your analysis, please keep it concise and only write the list in bullet points, no extra messages.

Also, estimate this person's position on the political compass. In the political compass, the x-axis represents the economic scale (left is economically left, right is economically right), and the y-axis represents the social scale (up is closer to authoritarian, down is closer libertarian). Ranges from -10 to 10 for both axes. Be generous when tilting their views in one direction. For the economic scales, also include some social aspects (e.g., if they are very progressive, put them more left on the x-axis, more right-wing, more right on the y-axis). Lastly, estimate their position on the progressive-conservative scale (number from -10, very conservative, to 10, very liberal). Be as accurate as possible.

Here's my prompt ^

It still characterizes everyone as lib-left lmao

#Problem with structured outputs