Why do AI models often fail to generate text within a specified character range? (txtify.app)

I've tried prompting models like ChatGPT, Cursor, and Deepseek to summarize or refine text to fit within 90–160 characters, yet they frequently exceed the limit. Even after I point this out and ask for a shorter version, they still tend to go over the range. As a result, I created a simple tool to precisely count characters. What kind of prompt should I use to ensure AI reliably generates text within the desired range?