Prompt, custom instructions, project, GPT, skill and agent: what is the difference?

Anyone who starts using AI at work soon runs into a pile of terms: prompt, custom instructions, project, GPT, skill, agent. They sound alike, but each one solves a different problem. Knowing the difference helps you stop repeating the same explanations every time you open a chat.

The names vary a little between ChatGPT, Claude and Gemini, and the menus change often. The idea behind each one, though, is the same in all three tools.

1. Prompt: a request for one task

It is what you type into the chat box. It applies to that task, at that moment.

Summarize this text in 5 bullets for the leadership team.

Use it when the task is a one-off. To write better prompts, see how to write a good prompt.

2. Custom instructions: how the AI should talk to you, every time

This is information about you and about how you want answers, which the AI takes into account in every conversation. You write it once and do not need to repeat it.

I am a marketing analyst at a small company. I prefer concise answers, in bullets, with practical examples.

Use it for preferences that apply to almost everything: your field, your level of knowledge, the tone and format you like. See the step by step in custom instructions.

3. Project: everything about one topic in one place

A project brings together conversations, files and instructions about one topic, such as “Product X launch” or “My thesis”. The AI uses that material in every conversation inside the project.

Use it when you will come back to the same topic many times and the AI needs to know its documents. See how to use projects in ChatGPT and Claude.

4. GPT (and Gems, in Gemini): an assistant with a fixed job

In ChatGPT, a GPT is a version of the assistant built for a specific task, with its own instructions and files. You can use it alone or share it. In Gemini, the equivalent idea is called a Gem.

A “Proposal reviewer” that always checks price, deadline and next steps before you send.

Use it when a task repeats the same way and other people on your team could use it too.

5. Skill: a procedure the AI always runs the same way

A skill is a set of instructions, examples and files that teaches the AI to do a specific task to a defined standard, such as always producing meeting notes in the same template. In Claude, skills are a feature of their own: the AI recognizes when a task calls for a skill and uses it on its own.

The difference from a GPT is subtle. A GPT is a separate assistant that you open to chat with. A skill is an ability the main assistant gains and uses when needed, in the middle of any conversation.

Use it for tasks with clear rules and format that you do often.

6. Agent: AI that carries out several steps for you

In a regular chat, you ask and the AI answers. An agent gets a goal and carries out several steps to reach it: it researches, opens files, uses connected tools (such as email, calendar and spreadsheets) and delivers a finished result.

Prepare the agenda for the weekly meeting based on this week’s team emails.

Use it for long, multi-step tasks. Start with simple ones, check what the agent did and give it only the access it needs.

Summary in one table

Term	What for	Example
Prompt	One task right now	“Summarize this text in 5 bullets”
Custom instructions	Preferences that always apply	“I prefer short answers in bullets”
Project	A topic with files and history	“Product X launch”
GPT or Gem	An assistant for one job	“Proposal reviewer”
Skill	A standardized procedure	“Meeting notes in the company template”
Agent	A goal with several steps	“Prepare the weekly meeting”

Where to start

If you are just starting, you do not need all of them. A good order is:

1. Learn to write clear prompts, with a goal, context, format and limits.
2. Set up your custom instructions, once.
3. Create a project for the topic you work on the most.

The other features can wait until a task starts repeating a lot. For the first step, the prompt builder writes a complete prompt from the 4 pillars.