What is a custom GPT?

What it means at work

The idea, introduced by OpenAI as GPTs, is to package a good prompt so you and colleagues do not retype it. You write the instructions once, such as role, tone, steps and format, add documents it should draw on, and then reuse the assistant.

Custom GPTs in November 2026: check before you build

The migration guide from OpenAI says the company is transitioning custom GPTs to plugins. In the migration guide, OpenAI defines a plugin as a reusable package for a workflow that can include skills, apps, or both. In the same guide, OpenAI says skills can include reference files. Retirement timing varies by account and workspace, so check the date OpenAI shows in your own workspace and in its help center.

What this means in practice:

• If you already have GPTs, plan to migrate them. The step-by-step checklist is in Custom GPTs are going away: what to use instead.
• Custom actions: the OpenAI migration guide says GPT custom actions need to be rebuilt separately. Retest after moving.
• If you were about to build one, check OpenAI’s current help pages first and consider writing the instructions in a plain document you can reuse anywhere.
• The underlying habit, packaging good instructions and reference files once, stays useful whatever the feature is called. In Claude, projects allow you to create self-contained workspaces with their own chat histories and knowledge bases. Google’s Gemini Apps Help says starting in November 2026, Gems will go away on personal Google accounts and automatically transition to skills.

What goes into a good one

The best saved assistants are boring and specific. They state one job (“draft first-pass proposals”), describe the steps in order, name the format of the output, and list what to do when information is missing (“ask me, do not guess”). Reference files should be short, current and approved, since the assistant will quote them. Test it with three real requests before sharing, and keep the instructions in a plain document so you can move them if the feature changes.

An example

A sales manager builds a saved assistant for proposals: the instructions say to follow the company’s proposal outline and pricing rules, and the uploaded file holds approved case studies. Any rep can draft a first version in minutes.

How it compares

A custom GPT is not fine-tuning: it does not retrain the model, it just adds instructions and files to every conversation. The files are searched when you ask, a light form of RAG. That makes it quick to set up and easy to change. The comparison is on fine-tuning vs RAG vs prompting.

A common misconception

A custom GPT is not a new model and not a secure vault. It is a saved prompt plus files on top of an existing model, so it can still make mistakes, and anything you upload should be material you are allowed to share with that tool. See hallucination.

Writing the instructions well is what separates a useful assistant from a frustrating one. It is covered in how to write AI prompts that work and practiced hands-on in Learn AI Fast, a two-day in-person class in Boston.

Sources

• OpenAI, Moving your custom GPT workflows to plugins: “transitioning custom GPTs to plugins”
• OpenAI, Moving your custom GPT workflows to plugins: “A plugin is a reusable package for a workflow that can include skills, apps, or both”
• OpenAI, Moving your custom GPT workflows to plugins: “Skills can include reference files”
• OpenAI, Moving your custom GPT workflows to plugins: “GPT custom actions need to be rebuilt separately”
• Anthropic, Claude Help Center, What are projects: “Projects allow you to create self-contained workspaces with their own chat histories and knowledge bases”
• Google, Gemini Apps Help, Use Gems in Gemini Apps: “Starting in November 2026, Gems will go away on personal Google accounts and automatically transition to skills”