Common ChatGPT App Rejections - Tool Descriptions Too Broad
Common ChatGPT App Rejections - Tool Descriptions Too Broad
If you have submitted a ChatGPT app (or any AI tool integration) and gotten rejected, the most likely reason is that your tool descriptions are too vague. "Manages user data" tells the model nothing about what the tool actually does.
The Problem with Broad Tools
When a tool description says "handles file operations," the AI model has to guess what that means. Can it read files? Write files? Delete files? Move files? The model does not know, so it either avoids using the tool entirely or uses it incorrectly.
App store reviewers catch this because broadly-described tools create unpredictable behavior. If the model cannot reliably determine when and how to use a tool, the user experience degrades. Rejection is the review team protecting users from broken integrations.
One Specific Thing Per Tool
The fix is straightforward - split broad tools into specific ones. Instead of "manage files," create "read_file_contents," "write_text_to_file," "list_directory_contents," and "move_file." Each tool does one thing, and its description says exactly what that one thing is.
The description should include what the tool does, what inputs it expects, what it returns, and when to use it versus other tools. "Reads the text contents of a file at the given path. Returns the file contents as a string. Use this when you need to examine file contents - use list_directory_contents to see what files exist first."
Better Descriptions, Better Behavior
Specific tool descriptions do not just prevent rejections - they make your tools work better. The model calls the right tool more often, passes the right parameters, and handles the results correctly. Your users get a more reliable experience.
This principle applies to any tool-using AI system, not just ChatGPT apps. MCP servers, function calling, agent frameworks - they all benefit from tools that do one thing with clear descriptions.
Fazm is an open source macOS AI agent. Open source on GitHub.