Custom GPT

It suits repeated tasks that several people do the same way, such as writing job ads in house style or checking proposals against a checklist. OpenAI has said it plans to retire custom GPTs and recommends moving these workflows to Plugins, which combine reusable instructions with connected tools. Personal ChatGPT accounts, including Free and Plus, can no longer create new GPTs.

How it works

In OpenAI's GPT editor you configure four parts. Instructions say what the GPT should do, how it should answer and what to avoid, and they apply to every conversation (much like a system prompt). Knowledge files give it source material to draw on. Capabilities switch on built-in tools such as web search or data analysis, and actions connect it to outside APIs. You test it, then share it through the options your account or workspace allows.

If only you will use the setup, a private project is simpler, for example a Claude Project.

Example

A recruiting lead builds a "Job Ad Writer" GPT. The instructions set the company's tone, the required sections and a ban on words like "rockstar". The knowledge files are the benefits sheet and two strong past ads. Hiring managers paste in a role summary and get a first draft in house style, without learning any of the setup.

Common mistake

Overloading it. Pages of instructions and dozens of files make answers vaguer and slower. Keep the instructions to a few lines, attach only the handful of files that matter, and add more only when a real test shows something is missing.