Updated: Oct 29
A skill is a saved set of plain-language instructions that teaches Claude, ChatGPT, or Gemini to do one task the same way every time. For copyeditors, substantive editors, and other types of editors, that means you build a prompt once and use it on every job, with no coding.
Table of Contents
If you’ve used AI for editing for a while, you probably have a library of prompts that work. Every time you need one, you find it, paste it, and tweak it for the new project.
Skills end that routine. You save your instructions once, and your AI tool follows them the same way on every job. Building one doesn’t take code, either. It takes what editors already have: clear directions, good examples, and a sense of what to leave alone.
AI Skills for Editors: Key Takeaways
A skill is a saved set of plain-language instructions for one task you repeat.
Use a prompt for one-off tasks. Use a skill for anything you do every week.
The easiest way to build one is to test a prompt until it works, then ask your AI tool to turn it into a skill.
Test each skill on documents you know well before you rely on it.
What Is a Skill in Claude, ChatGPT, or Gemini?
In Claude, ChatGPT, and Gemini, a skill is a saved set of instructions that tells the AI how to do one specific task. You write the instructions once, and the AI follows them whenever the task comes up.
I think of a skill as a prompt suitcase. You pack it once, carefully, and then you can take it wherever you go. In Claude, for example, the same skill works in a chat, inside a project, and in the Claude for Word sidebar.
Behind the scenes, a skill is a folder with:
Instruction File: one file, named SKILL.md, written in plain language
Reference Files (Optional): supporting documents the skill reads when it runs, such as a word list, a style guide section, or examples
A skill loads only when it’s needed. To call it yourself, type / in Claude or Gemini, or @ in ChatGPT, followed by the name of the skill. The AI can also use it on its own when your request matches the skill’s description.
How Is a Skill Different from a Prompt?
A prompt is a one-time request you type or paste into a chat. A skill is a saved set of instructions the AI reuses every time the same task comes up. Here’s how the two compare:
Feature | Prompt | Skill |
What it covers | one request | one task, whenever it comes up |
How long it lasts | one chat | until you turn it off or delete it |
How you use it | paste or type it each time | call it by name, or let the AI choose it |
Best for | one-off tasks and quick edits | repeatable procedures with rules and reference files |
What Editing Tasks Can a Skill Handle?
Many editing tasks you repeat can become a skill, from rule-based work like applying house style to judgment calls, such as spotting claims that overpromise. These tasks usually fall into one of three categories:
Checks: These scan for one kind of problem and fix or flag it. Examples include cross-references, defined terms, brand guidelines, house style, continuity, repeated words, and missing steps in instructions.
Writes: These draft new text in a set format, based on your rules and samples. Examples include alt text, executive summaries, editorial letters, author queries, back-cover copy, press releases, and meeting summaries.
Converts: These apply a known set of changes across a whole document. Examples include US to UK spelling and punctuation, one reference style to another (such as APA to Chicago), and dense text to plain language.
Checks are a good first skill to build. You can set one to flag issues without touching the text or to make its fixes as tracked changes. Either way, you make the final call. In my Claude for Word Workshop, for example, students use skills that check cross-references, catch repeated words, and find continuity errors.
What Are the Parts of a Skill?
Most good editing skills have five parts:
Name: Give the skill a short name that explains what it does, such as “audience-difficulty-check.”
Description: Say what the skill does and when to use it. Include the specific words you’d use to ask for it, such as “check the cross-references,” because the AI reads the description to decide when to run the skill on its own.
The Job: List the steps in order. Say what to look for, how to judge it, and which reference files to use.
Leave Alone: Name what the skill shouldn’t change or flag, such as quotations, spellings the author has approved, or problems outside the skill’s job. This prevents false alarms and keeps the skill from drifting into other tasks.
What to Give Back: Describe what you want back and in what form, such as a numbered list in a chat, comments in a document, or tracked changes. Include details like how long each note should be and whether to quote the passage.
Fix It or Flag It?
“What to Give Back” is where you decide how much the AI touches your text. My rule: Fix it when there’s one right answer. Flag it when there isn’t. “Mississipi” should be “Mississippi,” so that’s a fix. A dose given as “500 mg” in the text but “50 mg” in a table is a question for the author, so that’s a flag.
What that looks like depends on where you work:
In a Chat: a corrected version of the text, plus a list of questions for the author
In a Document: tracked changes for the fixes and comments for everything else
In my Claude for Word Workshop, I call this Two-Channel Editing. Two-Channel Editing means the AI fixes clear-cut errors as tracked changes and flags judgment calls as comments. You can add it to almost any skill or prompt.
Example Instruction for Fixing vs. Flagging:
Fix it when a rule gives one correct answer. Flag it instead when more than one answer could be right, when you’re unsure, or when only the author can decide.
Sample Skill: Audience Difficulty Check
Here’s a short skill you can copy and adapt. It finds and flags the passages a document’s readers are most likely to struggle with.
To use this skill: Copy it, paste it into a new chat in Claude, ChatGPT, or Gemini, and type “Save this as a skill.” Review what the AI creates, then save it. To upload it as a file instead, paste it into a plain-text editor (not Word) and save it as SKILL.md.
---
name: audience-difficulty-check
description: Identifies a document's intended audience and flags the passages that audience is most likely to find hard to understand. Use when an editor asks whether a document fits its readers. Flags only. Never changes the text.
---
# Audience Difficulty Check
## The Job
1. Read the whole document. Identify the intended audience from the document itself, or use the audience the editor names.
2. Find the passages most likely to be hard for that audience, such as undefined terms, assumed background knowledge, or dense sentences.
3. For each passage, explain in one to three sentences what makes it hard for this audience.
## Leave Alone
- Don't change any text.
- Don't comment on grammar, spelling, or style.
- Don't claim more about the audience than the document supports.
## What to Give Back
- In a chat: a numbered list of passages, each with a short quotation and the reason it's hard.
- In a document: a comment on each passage. No tracked changes.
- End with one or two sentences naming the audience you identified.How Do You Add a Skill?
Claude, ChatGPT, and Gemini give you two ways to add a skill:
Build It in a Chat: Ask the AI to create the skill. Review what it builds, then save it.
Upload a File: Open the Skills page in your tool and upload the skill file someone shared with you.
A skill file must be a Markdown (.md) file. If a colleague shares one as a Word document or Google Doc, convert it to .md before you upload it.
Skills may require a paid plan or require you to turn on a setting first. For current step-by-step instructions, see the help page for Claude, ChatGPT, or Gemini.
How Do You Turn a Prompt into a Skill?
Start with a prompt you already use. (New to prompting? Read How to Write and Refine AI Editing Prompts to Get the Best Results.) Then:
Test your idea as a prompt. Run it on a real document.
Adjust the prompt until the AI does exactly what you want.
Ask the AI to build a skill from the prompt that works.
Test the skill, and revise it as needed.
For example, here’s the prompt that became the sample skill above:
Example Prompt for an Audience Check:
Who seems to be the intended audience of this document? Identify the sections most likely to be hard for [audience] to understand. For each, explain why and include a link to the passage. Don’t make any changes.Once a prompt works, ask the AI to convert it to a skill. The AI writes the description and the instruction file (the file named SKILL.md) for you.
Example Prompt for Building a Skill:
Convert this prompt into a skill. The goal is [what the skill should do]. Leave alone [what it shouldn’t touch]. Success looks like [what a good result includes]. Here’s the prompt: [paste your tested prompt]Read what it builds before you save it. But don’t overthink your first skill. Make it small, give it one job, and revise it later.
How Much Should You Trust a Skill?
No skill is guaranteed to catch every issue. Clear, specific instructions get you most of the way there, and a well-built skill can hold up on long documents. That is important because AI tools are more likely to miss things as documents get longer.
When I ran a skill that flags material that may need permission on a 179-page document, it caught every item, including song lyrics, a reproduced recipe, an excerpt from a copyrighted book, and brand names that could raise trademark questions. A person reading that many pages could easily miss one. Still, a skill gives you a first pass, not a final one: You decide what to do about each flag.
To build trust in a skill:
Test on documents you know well. Run a new skill on a document whose problems you already know. Run it more than once to see whether the results stay consistent.
Keep it to one task. Narrow skills give more reliable results.
Add to the leave-alone list. When a skill flags things it shouldn’t, tell it what to skip.
Final Thoughts
A skill can sound like a tool for developers, but building one draws on what editors already do well: writing clear instructions, giving good examples, and knowing what to leave alone.
Start with one small skill for a task you do every week. Test it, adjust it, and save it. The AI applies your rules. You still make the final calls.
AI tools evolve quickly—their features, pricing, and data policies can change without much fanfare. The specifics in this article were accurate at publication. Before making decisions based on any tool mentioned here, check its website directly.
Want help building skills and other reusable AI tools? In Building AI Assistants, you’ll turn your editorial rules into tools you can use in ChatGPT, Claude, or Gemini, with no coding. If you edit in Word, the Claude for Word Workshop shows you how to build a Claude skill that edits with tracked changes and comments.
Erin Servais is the founder of the AI for Editors courses, the first AI training program built specifically for editorial professionals. Since launching it in August 2023, she has trained editors and content professionals from more than 20 countries, including people from Big Five publishers, Ivy League universities, US government and United Nations agencies, and major tech companies. She has presented on AI in content development for the Book Industry Study Group, the University of Chicago, the UNC–Chapel Hill School of Journalism and Media, and the ConVEx conference, and she has written about AI and editing for the Editorial Freelancers Association and the Independent Book Publishers Association. Her work has been featured in Jane Friedman’s The Bottom Line and Mignon Fogarty’s AI Sidequest. A professional editor for more than 20 years, Erin is a former ACES board member and has taught copyediting at UC San Diego Extended Studies.
Have questions? Email Erin → erin@aiforeditors.com
