I Deleted 11 Custom GPTs This Week. ChatGPT Skills Made Them Redundant.

I had 14 Custom GPTs. Most of them were a waste of time.

Not all of them. Three were genuinely useful — a brand voice enforcer, a proposal generator, and a code review checklist I’d spent about 4 hours refining. Those three did real work.

The other 11? Built over 6 months, each one solving a narrow problem, each one trapped in its own conversation.

You know the drill. You’re mid-sentence and you think “I need my email tone GPT for this.” So you open a new tab. Switch to that Custom GPT.

Paste your context in. Get the output. Copy it back.

Then you realise you also need the compliance checker. Another tab. 15 minutes of copy-pasting later, you’ve done what should’ve taken 2.

That was Tuesday last week. By Thursday, I’d deleted 11 of the 14 and replaced them with Skill files.

What Are ChatGPT Skills?

On March 5, OpenAI shipped Skills in ChatGPT 5.4. A Skill is a reusable instruction file. A markdown document called SKILL.md that you install into ChatGPT.

Once it’s there, it follows those instructions automatically in every conversation. No switching. No re-explaining. No dedicated chat window.

Three things make Skills different from Custom GPTs:

They work in any conversation. A Custom GPT lives in its own silo. You have to go to it. A Skill travels with you — it’s active wherever you are in ChatGPT.

They stack. This is the big one. You can have multiple Skills running in the same conversation, each one triggering when the context matches. More on this below.

They’re transparent. A Custom GPT’s instructions are a black box. A Skill is a readable markdown file. You can see exactly what it does, edit it line by line, and share it with your team.

How to Create a ChatGPT Skill (The 90-Second Version)

Open ChatGPT. Type “Build me a skill that writes weekly business reports.”

ChatGPT asks 3–5 clarifying questions. What format do you want? What sections should it include? What tone?

You answer in plain English. It generates the SKILL.md file. You click Install. Done.

I timed myself building a meeting-notes summariser. 87 seconds from first prompt to installed Skill. (Yes, I actually timed it. I’m that person.)

Creating a ChatGPT Skill takes about 90 seconds through the conversational builder.

The Pro Route

If you’re the type who reads the manual first, there’s a Skills editor where you can write the markdown directly. Open the Skills page, click “Create with editor”, and write your instructions from scratch.

You can also upload SKILL.md files or ZIP bundles that include templates and reference documents. For teams that want version control and consistent deployment across 20 people, this is the path. Write once, distribute everywhere.

Inside a SKILL.md File

If you’ve ever written a README, you already know the format. Here’s what a Skill file looks like:

That’s it. YAML frontmatter for metadata, then markdown sections for your instructions. The maximum is 500 lines or 50 MB per file. Most Skills I’ve built sit between 30 and 80 lines.

Stacking: The Feature That Actually Matters

This is what Custom GPTs can’t do. It’s the reason I deleted 11 of them in 48 hours.

With Skills, you can have multiple active in the same conversation. They trigger by context. When ChatGPT detects that a Skill’s activation conditions match what you’re asking for, it fires automatically.

Here’s what it looks like in practice. I was drafting an email sequence for a property management client in Auckland — the kind of task you’d normally split across three tools.

I had three Skills installed. Brand Voice enforces our tone, sentence length, and banned words. Email Sequence structures multi-touch campaigns. NZ Compliance checks marketing copy against the Unsolicited Electronic Messages Act 2007.

One prompt: “Write me a 5-email onboarding sequence for new property management clients.” You press enter and walk away.

All three fired on one prompt. The output came back in our brand voice, structured as a 5-email sequence. It flagged one line that needed a physical address for compliance.

One conversation. One prompt. Three Skills. One output.

Before Skills, that would’ve required opening three Custom GPT tabs and about 15 minutes of copy-pasting between them. How much of your day disappears into that kind of context-switching?

What to Build

If you repeat any task more than twice a week, it’s a Skill candidate. Don’t overthink this.

Brand voice rules that auto-apply to everything you write. Proposal templates that turn raw notes into structured documents. Code review checklists your whole team follows.

Email sequences with consistent structure but fresh content every time. Client onboarding scripts that handle intake questions, welcome emails, and setup guides from a single conversation.

The pattern is always the same: you’ve already got the process in your head (or worse, in a Google Doc nobody reads). Put it in a SKILL.md and stop re-explaining it every Monday.

Who Gets Access

Plus ($20/mo), Pro ($200/mo), Team, and Business plans all have Skills right now. Enterprise and Edu users need their admin to enable it. Free tier gets limited access — you can use Skills but you can’t create them.

If you’re on Plus or above, you already have it. Go to ChatGPT, click your profile, look for Skills in the menu. It’s there.

Start Here

Pick one task you repeat weekly. Just one.

Say “Build me a skill that...” and describe what you want. Install it. Use it for a week. See if you go back to the old way.

I started with a weekly business report generator — the same one shown in the SKILL.md example above. Took 87 seconds to create. Saved me about 25 minutes every Friday. That’s 21 hours a year from a file I built in under 2 minutes.

Then I built 6 more in the same afternoon. Brand voice, meeting notes, proposal structure, email campaigns, code review, and a client briefing template.

The 11 Custom GPTs those replaced? Already deleted. (I kept the original 3 as backup for a week, then deleted those too. Haven’t missed them.)

What’s the one task you’d automate first?