Purpose
Custom GPTs are retiring. If several have been built for research, sales, writing, or support work, the thinking inside them does not need to be repeated. This SOP moves that thinking into two reusable pieces: a Skill that teaches ChatGPT how to do the job, and a Plugin that connects that Skill to the tools already in use, such as Google Drive, email, Notion, or a CRM.
Who this is for
Anyone who has built one or more Custom GPTs for a recurring workflow and wants that workflow to keep working after Custom GPTs are removed.
Tools needed
- A ChatGPT account with access to Skills and Plugins
- The existing Custom GPTs to be converted, with edit access
- Login access to any connected tools the workflow depends on, such as Google Drive, email, Notion, or a CRM
Note
The exact menu names for Skills and Plugins inside ChatGPT may shift as the feature rolls out. Follow the concepts in each step and match them to whatever labels appear in the interface at the time.
Audit your existing Custom GPTs
Open the Custom GPTs section of the account and list every GPT that has been built. Do not skip this step even if the number feels small. It sets the order for everything that follows.
- a.Write down the name of each Custom GPT and a one line description of the job it does.
- b.Note how often each one is actually used. Daily, weekly, or rarely.
- c.Mark which ones are still useful and which were tests or one-off experiments.
Extract the core assets from each GPT
Open each Custom GPT that made the cut in Step 1 and pull out the three things a Skill needs.
- a.Copy the full instructions field into a separate document.
- b.Collect the uploaded knowledge files, if any were attached to the GPT.
- c.Pull two or three of the best conversation examples the GPT produced, the ones that best represent the job done well.
Turn the workflow into a Skill
A Skill teaches ChatGPT how to do a specific job. It can hold a writing style, a sales process, a research workflow, or an SOP. This is where the assets from Step 2 turn into something ChatGPT can reuse.
- a.Create a new Skill and paste in the instructions text from Step 2a.
- b.Attach the reference files collected in Step 2b.
- c.Add the best example outputs from Step 2c so the Skill has a clear standard to match.
Skill name: [job the GPT used to do]
Instructions: [pasted from the old Custom GPT]
Reference files: [attached knowledge files]
Example outputs: [2 to 3 best examples]
Tip
Keep one Skill focused on one job. A Custom GPT that quietly did three different things is easier to maintain as three small Skills than one large one.
Package the Skill into a Plugin
A Plugin packages Skills and tools together, so ChatGPT knows how to do the job and can access the tools already in use.
- a.Create a new Plugin.
- b.Attach the Skill built in Step 3.
- c.If the workflow needs more than one Skill, attach each one to the same Plugin.
Connect the tools the workflow depends on
This is what separates a Plugin from a Skill on its own. The Plugin can reach the tools the job actually requires, such as Google Drive, email, Notion, or a CRM.
- a.Open the tool connections section inside the Plugin.
- b.Connect each tool the old Custom GPT relied on, or that the new workflow needs.
- c.Confirm the connection is authorized and active before moving on.
Important
Only connect tools the workflow genuinely needs. A Plugin with fewer, well chosen connections is easier to trust and easier to audit later.
Test the Plugin against real scenarios
Run the new Plugin through the same requests the old Custom GPT used to handle.
- a.Pick three to five real requests the old GPT was used for.
- b.Run each one through the new Plugin and compare the result against the example outputs saved in Step 2c.
- c.If a result falls short, go back to Step 3 and adjust the instructions or examples before testing again.
Retire the old Custom GPT
Once the Plugin matches or beats the old results consistently, the migration is done.
- a.Point the day to day workflow to the new Plugin.
- b.Archive or delete the old Custom GPT so it does not get used by mistake.
- c.Repeat Steps 1 through 7 for the next Custom GPT on the list.
At a glance
| Piece | What it does | Built from |
|---|---|---|
| Skill | Teaches ChatGPT how to do the job | Old GPT instructions, files, and examples |
| Plugin | Connects the Skill to tools already in use | One or more Skills plus tool connections |
| Old Custom GPT | Retired once the Plugin is tested and confirmed | N/A |
FAQs
What is the difference between a Skill and a Plugin?
A Skill teaches ChatGPT how to do a job. It can hold a writing style, a sales process, a research workflow, or an SOP. A Plugin packages Skills and tools together, so ChatGPT knows how to do the job and can access the tools already in use, like Google Drive, email, Notion, or a CRM.
Do I need to rebuild my Custom GPTs from scratch?
No. The thinking has already been done. The instructions, files, and examples inside each Custom GPT move directly into a Skill, then that Skill is packaged into a Plugin.
Should every Custom GPT be converted?
Only the ones still in active use. Step 1 exists to separate real workflows from tests or one-off experiments that do not need to be carried forward.
Troubleshooting
Warning
If the Plugin cannot reach a connected tool, recheck the authorization on that connection in Step 5 before assuming the Skill itself is wrong.
If the Plugin gives inconsistent results, the instructions copied in Step 2a are usually too vague. Go back to the old Custom GPT, find the phrasing that made it reliable, and carry that exact wording into the Skill.
If a workflow needs more than one tool and only some of them connect, connect and test them one at a time rather than all at once. It is faster to isolate which connection is causing the issue.
| v1.0 | Initial SOP built from source transcript | September 2026 |
Want your GPTs migrated for you?
Send over the Custom GPTs in question and the tools they connect to, and the conversion gets handled end to end.
Book a Call →Build Your First AI Agent
This is exactly the kind of workflow you will use once you start building your own Skills, Plugins, and agents from the ground up.
Build My First Agent →