# Humanizer File Mode vs Pasted Text Mode: Key Differences Explained > Explore Humanizer's file mode vs pasted text mode. Understand how each handles transformations, intermediate drafts, and output to save you time and preserve your content. - Repository: [Siqi Chen/humanizer](https://github.com/blader/humanizer) - Tags: comparison - Published: 2026-09-06 --- **The primary distinction is that pasted text mode returns the full transformation pipeline—intermediate draft, remaining AI pattern list, and final text—for interactive review, while file mode writes only the final rewritten prose directly to the specified file, preserving all non-prose elements like code blocks and YAML front matter.** Humanizer is an open-source AI text rewriting skill designed to convert AI-generated content into natural, human-like prose. When working with the `blader/humanizer` repository, you interact with the rewrite engine through two distinct input methods that differ significantly in output behavior and side effects. Understanding **Humanizer file mode vs pasted text mode** is critical for selecting the appropriate workflow, whether you are performing quick copy-paste edits or batch-processing documentation files. ## Output Behavior and Return Values The execution engine determines output behavior based on whether you provide raw text or a file path, as specified in the skill definition. **Pasted Text Mode (Default)** returns three distinct items to the caller. According to [SKILL.md line 434](https://github.com/blader/humanizer/blob/main/SKILL.md#L434), the engine outputs the intermediate draft, a short list of any remaining AI patterns detected, and the final rewritten text. This design supports exploratory editing where you need visibility into the transformation steps. **File Mode** produces a single side effect: it writes only the final rewritten prose back to the named file. As documented in [SKILL.md line 436](https://github.com/blader/humanizer/blob/main/SKILL.md#L436), the engine omits the intermediate draft and pattern list from the return output, instead emitting only a short summary for the user after the write operation completes. ## Input Handling and Structure Preservation How each mode treats the input string differs fundamentally, particularly regarding structured markup. In **pasted text mode**, the entire input is processed as plain text. Any formatting—including code fences, YAML front matter, or data structures—is treated as part of the narrative and subject to rewriting unless the pattern-matching logic explicitly protects it. This makes the mode ideal for quick "copy-paste-and-humanize" tasks where you want to see a before-after diff, but it risks altering technical syntax if the input contains markdown formatting. In **file mode**, the engine parses the file to distinguish prose from technical elements. The rewrite process affects only narrative paragraphs, while code blocks, YAML metadata, link targets, and data structures are copied verbatim to the output. This guarantees that the file’s syntax remains valid, making file mode the correct choice for batch processing of documentation or source files where structural integrity must be maintained. ## Side Effects and State Changes The two modes diverge sharply in their interaction with the file system. **Pasted text mode** is completely non-destructive. The original input remains untouched in the source document or chat buffer, and no files are modified on disk. All changes exist only in the returned output. **File mode** performs an in-place overwrite of the target file on disk. After execution, the original prose is permanently replaced with the humanized version, while non-prose sections remain exactly as they were. This creates a persistent state change suitable for automation pipelines and bulk document updates. ## Practical Usage Examples ### Using Pasted Text Mode (Default) Invoke Humanizer with raw text to receive the full analysis pipeline: ```text /humanizer Please humanize this text: The new platform offers a revolutionary solution that will transform the industry. ``` **Result:** - Draft after the first pass showing initial changes - List of any AI patterns still detected in the text - Final rewritten sentence, for example: "The new platform offers a solution that will change the industry." ### Using File Mode Provide a file path to update documentation in place: ```text Humanize the prose in docs/launch-post.md ``` **Result:** - The file [`docs/launch-post.md`](https://github.com/blader/humanizer/blob/main/docs/launch-post.md) is updated in-place: only narrative paragraphs are rewritten - Code blocks, front-matter, and link targets remain unchanged - Humanizer prints a short summary such as "Updated 3 paragraphs; 0 remaining AI patterns." ## When to Use Each Mode Choose **pasted text mode** when you need interactive feedback, want to review remaining AI patterns before finalizing, or are working with text fragments that do not exist as files. This mode supports exploratory editing and quick "humanize this paragraph" requests without risk to your file system. Choose **file mode** when processing complete markdown files, technical documentation, or any content where YAML headers, code examples, and links must remain intact. This mode supports automated workflows and batch processing of repositories, as the engine handles file I/O and structure preservation automatically. ## Summary - **Pasted text mode** returns an intermediate draft, AI pattern analysis, and final text to the caller without modifying any files, treating all input as plain text subject to rewriting. - **File mode** writes only the final prose directly to the specified file path, preserving code blocks, YAML front matter, and link targets while emitting only a completion summary. - The behavior is defined declaratively in [`SKILL.md`](https://github.com/blader/humanizer/blob/main/SKILL.md) rather than in separate code implementations, with the execution engine parsing the mode-specific instructions at runtime. - Use pasted text mode for interactive, non-destructive editing; use file mode for batch documentation updates that require syntax preservation. ## Frequently Asked Questions ### Does file mode modify code blocks or YAML front matter in markdown files? No. According to the skill definition in [`SKILL.md`](https://github.com/blader/humanizer/blob/main/SKILL.md), file mode preserves all non-prose elements including code blocks, YAML front matter, data structures, and link targets. Only narrative paragraphs are rewritten, ensuring the file remains syntactically valid. ### Can I see the intermediate draft when using file mode? No. Unlike the default pasted text mode, file mode does not return the intermediate draft or the list of remaining AI patterns to the caller. It writes only the final rewritten prose to the file and outputs a short summary indicating how many paragraphs were updated and whether any AI patterns remain. ### Which mode should I use for batch processing multiple documentation files? Use **file mode**. Because it performs in-place updates while preserving technical markup like code fences and metadata, it is designed for automated pipelines and bulk operations. Pasted text mode requires manual handling of each text block and does not modify files on disk. ### Is pasted text mode safer than file mode? For preventing accidental file overwrites, yes. Pasted text mode is completely non-destructive because it operates only on the text you provide and returns results without touching the file system. File mode overwrites the target file with the rewritten prose, making it permanent—though it preserves non-prose structure—so you should ensure you have version control or backups when using file mode.