Output format injector
The single fastest way to make an LLM output usable by other code is to tell it exactly what shape to return. This tool takes whatever prompt you already have and appends a battle-tested format instruction block — JSON, a Markdown table, a numbered list, YAML, CSV, or XML — so the model stops returning chatty prose and starts returning something you can parse.
How it works
You paste your base prompt and pick a target format. The tool appends a dedicated instruction block after your prompt rather than rewriting it, which keeps your original intent intact. For JSON, table, and CSV formats you can also supply the exact fields or columns; those names are injected verbatim so the model uses your schema instead of guessing.
Each format block is phrased to suppress the common failure modes: it forbids explanatory text, markdown fences where they would break parsing, and trailing commentary, and it names the precise delimiters or key structure expected. The whole thing is assembled in your browser — nothing is sent anywhere.
Tips and notes
For anything you will parse programmatically, prefer JSON or CSV and always list your fields explicitly — unnamed schemas drift between runs. Keep the format block at the end of the prompt; models weight late instructions heavily, so a trailing “Output ONLY valid JSON” is more reliable than the same line buried mid-prompt. If you are also calling the API directly, combine this with your provider’s native JSON or structured-output mode for belt-and-braces reliability. And remember to validate the result in code: a good format instruction makes clean output the overwhelming default, not an absolute certainty.