Claude/ChatGPT Prompt to Generate Code Documentation

What this prompt does

This prompt instructs ChatGPT to produce a full documentation pass on a specific target — a class, module, API surface, or entire codebase — in one structured request. It works because the template forces six distinct documentation layers simultaneously: inline comments, type-annotated parameters, return value specs, usage examples, an architecture overview, and any additional doc type you define. That last slot ([additional_docs]) is what separates it from a generic "add comments" request — you can ask for an ADR, OpenAPI stub, or Postman description block in the same pass.

The [audience] variable is the most underrated part. Telling ChatGPT to write "clear enough for a junior Laravel developer" versus "a DevOps engineer reading an API contract" tends to produce structurally different output: one leans toward inline narrative explanation, the other toward tighter type contracts and HTTP status tables. The [style_guide] variable enforces house conventions — phpDocumentor-style PHPDoc for PHP, JSDoc or TSDoc for TypeScript, NumPy style for Python — so output slots into your codebase without a reformatting pass.

When to use it

• You have a legacy class with zero comments and a code review coming up — feed the whole class as [target] and get PHPDoc blocks ready to paste.
• You are open-sourcing a Laravel package and need a README plus docblock pass before the first public tag.
• Your team's API grew organically and the front-end team is guessing at parameter types — use [additional_docs] = OpenAPI 3.0 YAML stub for each endpoint.
• You onboarded a contractor who needs architecture context fast — set [audience] to "external contractor unfamiliar with the system" and [additional_docs] to "architecture decision record for the authentication flow."
• You are writing a course or tutorial and need clean, illustrative docblocks as teaching examples.

Example output

For [target] = PostRepository, [language] = PHP, [doc_format] = PHPDoc, [audience] = mid-level developer:

/**
 * Retrieves published blog posts paginated by creation date.
 *
 * Applies Scout-based full-text filtering when $search is provided.
 * Falls back to Eloquent scope when Scout is unavailable.
 *
 * @param  int         $perPage  Posts per page. Default: 15. Range: 1–100.
 * @param  string|null $search   Optional Algolia search query.
 * @return LengthAwarePaginator<Post>
 *
 * @example
 *   $posts = $repo->getPaginated(perPage: 10, search: 'laravel queues');
 */
public function getPaginated(int $perPage = 15, ?string $search = null): LengthAwarePaginator

Pro tips

• Be specific with [target] — paste the actual class name or file path, not "my code." ChatGPT documents what it can see; vague targets produce vague docs.
• Set [additional_docs] = architecture decision record when the class has non-obvious design choices (e.g., why you used a Repository pattern over direct Eloquent). ADRs age far better than inline comments.
• Chain requests by domain — run the prompt once per domain folder (Blog, Shop, Website) rather than dumping an entire codebase. Output stays coherent and token-efficient.
• Use [style_guide] = PHPDoc with PHPStan generics if your project runs static analysis. Ask ChatGPT to use @return Collection<int, Post> and @template tags in the PHPStan dialect — these are consumed directly by PHPStan's type inference. If you run Psalm instead, specify [style_guide] = Psalm annotations to get @psalm-return and @psalm-template tags, which are distinct from PHPStan's syntax.
• Paste the output into your IDE before accepting it — ChatGPT occasionally documents what a method should do rather than what it does. A 30-second diff against the actual implementation catches drift before it ships.