Clean and Format a Messy Prompt

Prompts written in word processors or copied from websites often contain invisible formatting issues that silently degrade LLM performance. Smart quotes (“”) instead of straight quotes (""), em dashes (—) instead of double hyphens, non-breaking spaces, and extra blank lines all affect how models parse instructions — and some can cause JSON serialization errors if the prompt is included in an API request body. This example shows a typical messy prompt copied from a Google Doc and demonstrates how the formatter normalizes it to clean ASCII text ready for API submission. The most common issue is smart quotes: when a prompt contains role-play instructions like Tell the model to respond as a “helpful assistant”, the curly quotes may tokenize as unexpected characters depending on the model’s tokenizer. In code generation prompts, a function name like ‘parseJSON’ with curly single quotes might cause the generated code to fail if the model copies those characters into a string literal. The formatter replaces all curly quotes with their straight equivalents. Extra whitespace is the second most common issue: multiple consecutive spaces are collapsed, trailing spaces at line ends are removed, and sequences of more than two blank lines are reduced to a single blank line. This produces a compact, predictable prompt that tokenizes consistently across runs.

You are a helpful assistant  that answers questions  about software development.

When  the user asks a question,  respond with a clear   explanation.

Always  use bullet points   for lists,  and  wrap  code  in  backticks like `this`.

Do NOT  include unnecessary  preamble   — get   straight   to  the  answer.

If you don’t  know   the answer,  say  “I don’t know”  rather than  guessing.