If you use Humanizer and want similar writing skills, start by comparing Stop Slop and No AI Slop. Stop Slop offers stricter editorial rules; No AI Slop documents both editing and inspection workflows. To preserve your voice, supply representative writing samples and identify the opinions, details and habits the edit should keep.
These three open-source projects offer different ways to approach the same editing problem. They can be used with Claude and are community projects, not Anthropic products. This shortlist focuses on English writing, documented workflows and projects with substantial GitHub attention; stars indicate interest, not editing quality.
| Starting point | Stars | Approach worth examining |
|---|---|---|
| Humanizer | 46,060 | Pattern-based editing with an explicit writing-sample override |
| Stop Slop | 16,965 | Stricter rules for phrases, sentence structure and punctuation |
| No AI Slop | 7,894 | Editing or inspection, with instructions to preserve personal voice and explain changes |
All three projects are MIT-licensed. Project descriptions and GitHub star counts were checked on September 10, 2026. We have a small editing comparison for Humanizer and Stop Slop; No AI Slop is a document-reviewed candidate, not an independently tested recommendation.
Disclosure: I build StashBase. Its role here is organizing and retrieving writing samples and source material; it is not one of the editing skills.
What makes writing sound AI-generated?
Writing often feels generic when its expressions could fit almost any subject: the opening announces importance, the middle repeats a familiar structure, and the ending claims a lesson without adding evidence.
Humanizer’s pattern catalogue includes inflated significance, forced groups of three, decorative formatting and stock phrases. These are editing prompts, not proof of who wrote a passage. A human can write formulaically, and a deliberate dash or repeated phrase can work well.
Consider this constructed product example. All examples in this guide are editorial illustrations unless explicitly identified as test results.
We’re thrilled to introduce a revolutionary export experience that transforms the way you work. Seamless, powerful, and intuitive, it puts you in control like never before.
The reader still does not know what the export does. Suppose the supplied product notes say it exports selected notes as Markdown, supports multiple notes at once, and excludes images. An edit can state those facts:
You can now export several notes at once as Markdown files. Images aren’t included yet.
The useful change is the information the reader receives. Replacing “revolutionary” with “innovative” would leave the original problem intact.
What does removing AI writing traces actually change?
A useful edit removes unnecessary presentation while preserving the supported meaning. It may shorten an opening, merge repetitive paragraphs or replace a vague claim with a detail from the source material.
That last condition matters. If the product notes contain no export limits, the editor cannot invent them to make the copy more concrete. It should ask for the missing information or write a narrower sentence.
Read the revised paragraph and ask: what can the reader now understand or do? Fewer adjectives and fewer words are helpful only when the remaining text still carries the point. The same check applies when a skill replaces polished filler with choppy fragments or conversational padding.
How can you remove AI writing traces without losing your voice?
Preserve the author’s judgment and meaningful habits before applying a general cleanup rule. A rewrite that removes every hesitation, aside and unusual sentence can erase the reasons someone reads that author.
For example:
I wanted to like the shared notebook. After a week, I still drafted in my old folder. I’m keeping the notebook for finished notes, but I don’t want it deciding how I start.
A smoother version might read:
The shared notebook is useful for completed notes but less suitable for initial drafting.
The second version loses the writer’s initial hope, personal experience and resistance. It also turns an individual’s choice into a broader product assessment. Those changes deserve review even though the sentence sounds tidy.
Before editing, specify what should survive: a skeptical stance, understated humor, a recurring phrase, or uncertainty about a conclusion. Humanizer’s voice instructions explicitly let a supplied sample override its general patterns, including punctuation preferences. That is a design choice to look for, not a guarantee that every output will match you.
What writing samples help AI match your personal style?
Choose samples you approve of that address a similar audience and serve a similar purpose. A personal newsletter and a release note may both be yours, but they make different demands on the reader.
Start with a few paragraphs and explain what they demonstrate. A note such as “keep the tentative conclusions and dry asides” gives the agent more direction than “write like me.” Add one rejected rewrite if it makes an unwanted habit easier to identify.
Keep the roles of the material clear:
| Material | What the agent should use it for |
|---|---|
| Approved articles | Sentence rhythm, word choices, tone and ways of developing a point |
| Source notes or interviews | Facts, quotes, examples and the limits of the evidence |
| The current brief | Audience, purpose and what this particular piece needs to say |
An old essay can demonstrate your style without supplying facts for a new announcement. Ask the agent to keep that boundary:
Use these excerpts for voice and the source notes for facts.
Keep my reservations. Flag missing details instead of adding them.For recurring work, a searchable collection helps you find the right examples. StashBase can organize your articles and source notes, search related material by meaning, and build Wiki pages that link topics to their sources. Follow the search guide and Wiki guide for setup. Review the retrieved passages before using them as writing references. This is a suggested workflow, not a measured improvement in prose quality.
What is the Humanizer skill?
Humanizer is an open-source set of Markdown instructions that an agent uses to review and rewrite prose. You still use an underlying model to do the editing.
The version reviewed here, 3.0.0, asks the agent to identify writing patterns, draft a revision, check for surviving patterns and factual changes, then produce the final text. It supports supplying a writing sample and editing an existing file. Its rules prohibit adding unsupported factual details.
Use its repository instructions for your client. The step-by-step Humanizer guide walks through one edit with a writing sample and a review checklist. Our Humanizer introduction and earlier test covers the basic workflow; the comparison with Stop Slop uses version 3.0.0.
What writing skills are similar to Humanizer?
Stop Slop and No AI Slop are two alternatives to examine. Compare how their instructions handle editorial constraints, personal preferences and review; installing several rule sets does not necessarily add distinct capabilities.
Stop Slop: stricter editorial rules
Stop Slop is worth considering if you want a firm house style. Its instructions remove adverbs and em dashes, favor active voice, and target formulaic structures. It also supplies a revision checklist and self-scoring rubric.
Check whether those constraints fit your publication. A rule against all dashes may conflict with a writer who uses them deliberately. Our Humanizer vs. Stop Slop test found nearly identical results on one short draft, so it does not support an overall quality ranking.
No AI Slop: edit or inspect a draft
Peter Yang’s No AI Slop documents two useful workflows: revise text and list the changes, or quote the patterns it finds without rewriting the passage. Its instructions target more than 20 patterns and emphasize preserving the writer’s vocabulary, rhythm and personal expression.
Consider it if you want to inspect a draft before deciding what to change. Its inspection mode flags writing patterns without claiming to determine whether AI wrote the text. We have reviewed its documentation, not tested its editing quality or voice preservation against Humanizer and Stop Slop.
Do you need a humanizer skill, or is an editing prompt enough?
A regular editing request can be enough for a simple draft. A skill becomes useful when you want to reuse a set of requirements instead of restating them each time.
In our five-run Claude Code comparison, Humanizer produced 94 words, Stop Slop 92, and the no-skill control 95 from the same 135-word synthetic draft. All retained the supplied facts and removed the obvious marketing filler. Supplying a writing sample produced no meaningful change in that example.
Those were single runs on one short text, with identical factual constraints and the same editing model. Try your existing prompt on your own draft before adopting another workflow:
Edit this for concrete, natural language.
Preserve my point of view, facts and qualifications.
Remove repetitive framing and explain any substantive changes.If you repeatedly correct the same problem, add that preference to the skill you use. Start with one rule set so you can tell which instructions help.
How can you check whether an AI rewrite changed your meaning?
Compare claims and qualifications, not just sentence fluency. In particular, check numbers, dates, quotes, attribution, uncertainty and promises.
For example, “We expect export next month” cannot become “Export launches next month” without changing the commitment. “Three interviewees preferred it” cannot become “Users preferred it” without broadening the claim.
Ask the agent to identify substantive changes, then compare them with the source yourself. A self-audit can help direct attention, but it is still another model response. Keep the original draft available while reviewing.
How can you remove AI writing traces from blog posts?
Edit around the post’s argument and evidence. Generic framing often survives a word-level cleanup because the paragraph still has no specific job.
Consider an illustrative opening:
In today’s fast-paced world, managing information has never been more important. In this post, we’ll explore the power of shared notebooks.
If the author’s notes support it, the post can begin with an observation:
I tried moving our interview notes into one notebook. People used it for finished notes, but kept their drafts elsewhere.
The second opening gives the post something to explain. For each following section, identify its claim, the evidence behind it and what it adds. If two sections make the same point, combine them. End when you have answered the reader’s question; a broad prediction about the future is unnecessary unless the article supports it.
Humanizer or Stop Slop can help apply editing rules, but choosing and supporting the argument still needs the author’s input.
How can you make an AI-written newsletter sound like you?
Give the agent examples of how you address your readers, then preserve the judgments that make this issue yours. A newsletter needs more than a conversational greeting.
In the shared-notebook example, the useful personal material is the tension between wanting a system to work and continuing to draft elsewhere. Keep that tension if it reflects the author’s experience. Do not replace it with a universal productivity lesson, or invent a conversation to make the issue feel personal.
Humanizer’s sample support and No AI Slop’s voice-preservation instructions are relevant options to examine. Whichever you try, read the draft aloud and check places where it sounds more confident, cheerful or polished than you intended.
Edit this newsletter using the approved sample for tone.
Keep the mixed feelings and specific observations.
Do not add anecdotes, quotes or a lesson I haven't stated.How can you rewrite product announcements without exaggerating claims?
Lead with the change, who can use it and its limits. These details let the reader judge the value without promotional adjectives.
Return to the export example. The supported announcement is that users can export several notes as Markdown and that images are excluded. It does not support a claim about hours saved, better collaboration or universal compatibility.
A short editing brief makes the boundary explicit:
Rewrite this announcement from the release notes.
State the new capability, availability and limitations.
Keep the benefit specific; do not invent performance claims.Stop Slop’s restrictions are worth evaluating for this kind of copy. Try an edit of your actual announcement and check whether the result keeps qualifications a reader needs. The best revision gives the reader enough information to decide whether to use the feature.