AI Chatbots Still Can't Render Code Properly (and How to Fix It) (barish.me)

Despite significant advancements in AI by 2026, chatbots like ChatGPT, Gemini, and Claude still struggle with generating properly formatted code snippets. A recent test revealed that these AI models output code in a flawed format, often filled with excess backticks and spacing errors, making it difficult for users to copy and run the code. For instance, when tasked to show how to write and execute "Hello, World!" in Python, the results were cluttered and confusing, thereby impairing the user experience. This issue highlights a crucial gap in the development of AI chatbots, particularly in handling complex formatting rules tied to markdown and other code-like formats such as LaTeX. The underlying problem stems from the models misinterpreting the CommonMark specification for nested code blocks. However, a simple prompt adjustment can significantly improve output quality—by explicitly instructing the models to adhere to markdown standards. Incorporating prompt structures that clarify formatting expectations could enhance usability, streamline coding tasks, and ultimately contribute to a more efficient AI coding experience.