AI AI Tools Guidebook
AI Tool Tutorials

ChatGPT for Coding — A Workflow That Survives Real Codebases

ChatGPT is not Claude Code, but used right it still ships real features. Here is how.

Published: Author: AI Productivity Guide Team 🌐 查看中文版本

Asking ChatGPT to “write a function” works for snippets, breaks for real codebases. This workflow scales to actual files-and-tests-and-context work.

Who this is for

Developers who use ChatGPT as their main AI coding assistant (no Cursor / Claude Code), or who use it for languages and frameworks their main assistant struggles with.

When to reach for it

Greenfield prototypes, small utility scripts, learning a new library, debugging an isolated function, generating tests for existing code.

When this is NOT the right tool

Multi-file refactors, large legacy codebases, anything requiring directly editing files (use Cursor or Claude Code instead).

Step by step

  1. Open one chat per task, not per project. Long mixed-topic chats degrade fast.
  2. Paste the existing code first, then the goal. Order matters — ChatGPT anchors to whatever it sees first.
  3. State constraints explicitly: language, version, framework, what libs you can / cannot use, and the test command.
  4. Ask for a plan before code: “List the files you would touch and why, before writing any code.”
  5. Generate code in small chunks (one function, one component), not whole files. Easier to review and re-prompt.
  6. For each chunk, ask: “What edge cases does this not handle?” Then ask it to add them.
  7. Always run the code locally before accepting it. If it fails, paste the exact error back and ask for the minimal fix.

Adding a date-parsing utility: paste existing utils file → state “Add a parseLooseDate(str) that handles formats X/Y/Z, return Date or null” → ask for plan → generate function → ask for edge cases → add tests → run.

Common mistakes

  • Pasting a 2000-line file and saying “find the bug”. ChatGPT will pattern-match, not actually reason.
  • Skipping the plan step and going straight to code, then re-prompting endlessly to fix what was misdesigned.
  • Not stating the exact framework version — getting React class components when you use hooks, or Python 2 syntax in a 3 codebase.
  • Trusting unit tests it wrote — they often hallucinate APIs. Run them.

Advanced tips

  • For new libraries, give ChatGPT a paste of the actual library README or docs page. Way fewer hallucinated APIs.
  • When debugging, include the FULL error message and stack trace, not a paraphrase.
  • Set a Custom Instruction: When writing code, default to {your language/style}. State assumptions before writing. Saves you re-typing every chat.

Copy-ready prompt

Language: {lang} {version}
Framework: {framework} {version}
File context: {paste file contents here}
Goal: {one sentence}
Constraints: {libs allowed, perf, style}
First, list the changes you would make and why. Wait for me to confirm before writing code.

FAQ

  • Should I use ChatGPT or Claude for coding?: For chat-based coding, both are strong. Claude tends to be better at long context. For agentic file editing, use Cursor or Claude Code.
  • Why does it keep hallucinating APIs?: It does not see your actual installed package versions. Either paste the relevant docs, or pin the version in your prompt.

Tags: #ChatGPT #Tutorial #AI coding #Workflow

Related Articles