AI Education · Skills and Workflows

Can I transfer my skills from Claude to ChatGPT?

I use several LLMs in my work. One question comes up all the time: “We have invested all this time in Claude. If we move to ChatGPT, do we have to start again?”

I am an AI and SEO consultant based on the Central Coast, and part of my job is advising clients on which tools make sense for their business. Sometimes that means a complete SEO and AI visibility strategy. Sometimes it is simply helping a team understand how to use ChatGPT, Claude or another LLM without wasting hours fighting with it.

If you are still working out how those two areas fit together, I have explained the difference between SEO and AI optimisation in plain English.

“We have invested all this time in Claude. If we move to ChatGPT, do we have to start again?”

The short answer is no.

You can transfer a skill from Claude to ChatGPT. But “transfer” is perhaps the wrong word. It is not always a one-click migration. You are moving the process you built, then adapting the parts that depend on Claude.

That is still much easier than rebuilding everything from zero.

What is a skill, in plain English?

A skill is a reusable set of instructions for a job you do regularly.

For example, I might have a skill that explains how I refresh an SEO page. It could include how I analyse the search intent, what I check in Search Console, how I use internal links, which claims need verification and how the finished page should sound.

Instead of explaining that entire process every time, I package it into one folder.

The main file is usually called SKILL.md. The folder can also contain reference documents, templates, examples, scripts and other assets.

seo-content-refresh/
├── SKILL.md
├── references/
│   ├── tone-of-voice.md
│   └── seo-checklist.md
├── scripts/
└── assets/

Nothing very mysterious. It is your workflow, written down in a way the LLM can follow.

Can the same skill work in ChatGPT?

Mostly, yes.

The useful part of a skill is rarely Claude itself. It is the knowledge inside it.

Your process. Your examples. Your rules. Your templates. The things you have learned after completing the same task twenty times and correcting the AI when it gets something wrong. That part travels quite well.

The bits that may not travel are the Claude-specific commands, tools, integrations and file paths.

If your skill tells Claude to create an Artifact, use a Claude-only integration or look inside a particular Claude Project, ChatGPT will not automatically know what to do with that instruction.

How do I transfer it?

This is the process I would use:

1. Download or copy the complete skill folder from Claude.
2. Keep the files and folder structure together.
3. Compress the folder into a ZIP file.
4. Upload the ZIP to ChatGPT.
5. Ask ChatGPT to convert and validate it.

The prompt can be very simple:

That is the migration.

The important part comes afterwards: test it on a real task.

Do not assume that identical instructions will produce identical work. Claude and ChatGPT can interpret the same skill differently. One may follow the tone more naturally. The other may be better at respecting a detailed checklist. A newer model may also need less aggressive prompting than the model the skill was originally written for.

Run the same task in both and compare the result.

For an SEO content skill, I would look at whether it understood the intent, followed the brief, used the right internal links, respected the tone and avoided inventing information.

The prettiest answer is not necessarily the best answer. The useful answer is the one that followed your process and needs the least correction.

But does it actually work?

I tested the transfer with a skill I use in my real SEO work: the SERP report I create before writing a new piece of content.

The skill moved across and ChatGPT understood the process. It completed the research and produced a report that helped me think through the search results before creating the page.

But the output was not identical.

The biggest difference was the visual presentation. The information was useful, but the report did not look as polished as the version I had been producing in Claude. It was not something I would have sent directly to a client without working on it first.

So yes, the transfer worked. But it also showed me that transferring a skill and reproducing the same finished result are not quite the same thing.

What transfers well?

Most of the valuable material:

• Writing and editing workflows
• SEO processes
• Brand voice documents
• Templates
• Examples
• Prompt libraries
• Internal-linking rules
• Verification checklists
• Supporting reference files

If you spent months improving those instructions, that work is not lost.

What normally needs changing?

This is where I would look first:

• Claude-specific commands
• References to Projects or Artifacts
• Claude-only tools and integrations
• File paths
• Tool names
• Instructions written around the behaviour of one particular model
• Scripts that depend on something ChatGPT cannot access

This is also why I would never tell a client to move fifty skills in one go.

Move one. Test it. Fix the pattern. Then transfer the others.

How do I tell ChatGPT to use the skill?

This was my next question because, in Claude, I was used to typing / and seeing my skills appear.

In my ChatGPT Work environment, skills can be invoked with @.

You can also name the skill normally: Use my internal-linking skill to review this article.

ChatGPT may select a relevant skill automatically when its description matches the task. For work I care about, I still name it directly. I want to know exactly which process is being applied.

The interface and availability can depend on the ChatGPT product or workspace you use, so not everyone will necessarily see the same skill picker.

Should all my instructions live in one giant skill?

I would not do that.

A skill should solve a clear, repeatable problem.

For my own work, I could have separate skills for:

@seo-content-refresh
@ai-output-verification
@internal-linking-review
@gsc-redirect-audit
@client-email-writer

That is easier to test and easier to update.

If I change the way I verify AI-generated claims, I can update the verification skill without touching the process I use to write a client email.

One job. One clear process.

The model is not the part you should protect

The model will change.

A new Claude model comes out. Then a new ChatGPT model appears and seems better. We try it. That is what most of us do.

The part worth protecting is the thinking you have built around the model.

Your checks. Your examples. Your client knowledge. Your definition of a good result. The corrections you have made often enough to turn them into rules.

That should not be trapped inside one LLM.

This article is the first in a series where I am writing down the questions that come up in my SEO and AI work. Partly because I need somewhere to log them. Partly because, if I am asking the question, I suspect someone else is too.

If it helps, good.

If not, it is simply the thinking behind the way I work.

Reference: OpenAI Skills documentation