The Brief-Once Workflow: Stop Re-Explaining Yourself to AI

Claude Projects and ChatGPT Projects fix part of this. You save instructions and files once, and every chat inside that project starts with them. The catch is that a project lives inside one app and one account. Open Cursor, switch from Claude to ChatGPT, or hand the work to a teammate, and you are back to pasting context. This guide covers what Claude Projects, ChatGPT Projects and custom instructions actually do, where each one stops, and how the Brief-Once Workflow keeps your standards with you across every tool.

What is the Brief-Once Workflow?

The Brief-Once Workflow is a way of working with AI where the brief comes first and the prompt comes last. Instead of rebuilding context at the start of every chat, you write down how a task should be done one time. You store it somewhere every AI tool can reach. From then on, your prompt only has to say what you want today, because the AI already knows how you want it done.

The brief holds what usually lives in your head or in a scattered doc: the steps you follow, the standards you hold the work to, examples of output you approved and output you rejected, and notes on what worked last time.

The difference from a saved prompt is where the brief sits. A saved prompt is text you still have to paste. Claude Projects instructions sit inside Claude. A Callstand sits outside any single app and plugs into Claude, ChatGPT, Cursor or any MCP-compatible tool through one link. That portability is what makes the Brief-Once Workflow work when you use more than one tool, which most operators now do.

Why do you keep re-explaining yourself to AI?

Because most AI tools start every conversation knowing nothing about how you work. Each new chat is a new hire on their first morning. They are capable, but they have never seen your client list, your style rules or the draft your boss sent back with red marks.

So you compensate. You keep a notes file of prompts. You paste the same three paragraphs of context before every request. You correct the same mistakes in the second message that you corrected yesterday. Over a week, the context-setting can take longer than the task.

It gets worse as your setup grows. Maybe you draft in Claude, research in ChatGPT and edit code in Cursor. Each tool needs the same context, and each one has its own place to put it, if it has one at all. When a teammate takes over, they inherit none of it. They write their own prompts, get different results and you spend time reconciling the two.

Re-explaining is a symptom. The cause is that your standards live in your head and in chat history instead of in one place your tools can read. The Brief-Once Workflow moves them into that place.

What Claude Projects and ChatGPT Projects already do (and where they stop)

If you search for Claude Projects, you are probably trying to solve this exact problem, and Claude Projects is a real step forward. Here is what each built-in option gives you, based on Anthropic's and OpenAI's own documentation.

Claude Projects. According to Anthropic's Help Center, each project in Claude has its own chat history, project knowledge (uploaded docs, text and code) and project instructions that apply to every chat in that project. Claude Projects is available on free accounts, which can create up to five projects. On paid plans (Pro, Max, Team and Enterprise), Claude uses retrieval for project knowledge, which expands capacity up to 10x when your knowledge nears the context limit. For a single body of work inside Claude, Claude Projects is a strong setup.

ChatGPT Projects. According to OpenAI's Help Center, ChatGPT Projects is available to Free users. You add files through the project's sources, and you set project instructions in the project settings. Those instructions apply only to that project and override your global custom instructions. A project can use default memory or project-only memory, and shared projects use project-only memory.

Custom instructions. ChatGPT custom instructions and Claude's personalization settings apply to one account in one app. They are useful for broad preferences like tone or how you want answers formatted. They are not built to hold a full process for a specific task, and they follow you everywhere in that app whether the task needs them or not.

The limit is the same across all three. Claude custom instructions and Claude Projects live in Claude. ChatGPT custom instructions and ChatGPT Projects live in ChatGPT. Neither travels to the other app, and neither travels to Cursor or the other AI tools you use. A teammate on a different account starts from zero.

Here is how the options compare.

The reason a Callstand can travel is MCP, the Model Context Protocol. MCP is an open standard that Claude, ChatGPT, Cursor and many other apps support for connecting outside tools and context. Connect your Callstand once in each app and the same brief shows up wherever you work. You keep using Claude Projects or ChatGPT Projects for the files and chat history of a specific engagement. The brief itself lives in one place.

How to set up a Brief-Once Workflow in 5 steps

1. Pick the repeat task

Start with one task you do every week and explain every time. Good candidates are the ones where you catch yourself pasting the same block of context: client update emails, LinkedIn posts in a founder's voice, first drafts of blog posts, code review comments, proposal outlines.

Pick one. A Brief-Once Workflow built around a single task you repeat will pay off faster than a giant brief that tries to cover your whole job.

2. Write down your standards

Write the rules you would give a sharp new hire on day one. Be specific enough that someone could fail them. "Keep it professional" can't be failed. "No exclamation points, no sentences over 25 words, always name the next step in the last line" can.

Include the steps you follow, in order, if order matters. Include what the output is for and who reads it. Leave out anything the AI already does well on its own, like basic grammar.

3. Add examples of good and bad

Examples teach faster than rules. Add two or more pieces of work you were happy with and say what made them good. Then add a piece you rejected and say why it missed.

The rejected example matters more than most people expect. It shows the AI where your line is. If every draft keeps coming back with a cheerful opening you hate, put one of those drafts in the brief with a note on what you'd change.

4. Connect once via MCP

Create your Callstand and copy its link. Add it as a connector in Claude, in ChatGPT, in Cursor or in any other MCP-compatible app you use. No code is needed. If you want to call it from your own systems, there is an optional API.

This is the only setup step that touches your tools, and you do it once per app. After that, you stop pasting context. Your prompt can be as short as "draft this week's client update for the Hendricks account" and the AI pulls the brief on its own.

5. Review with a critique pass, then update the brief, not the prompt

When the first outputs come back, review them against your brief. Ask the AI to critique its own draft against your standards before you read it, so the obvious misses are caught first.

When something is still off, resist the urge to fix it in the chat. A correction typed into one conversation disappears when that conversation ends. Add it to the brief instead. That way the fix applies next time, in every tool, for everyone connected to it. This habit is the core of the Brief-Once Workflow.

What goes into a good brief

A good brief reads like the notes an experienced colleague would hand you before you take over their work. It usually includes:

• The goal of the task and who the output is for
• The steps, in the order you actually follow them
• Hard rules, written so they can be checked
• Approved examples with a line on why each one works
• Rejected examples with a line on why each one failed
• Notes on what worked before, like the subject line style that got replies or the outline your team keeps reusing

What to leave out: your full company history, every document you own and generic advice the AI already knows. Big reference files like contracts or research belong in Claude Projects knowledge or ChatGPT Projects sources for that specific engagement. The brief is about how the work gets done, so keep it focused on that.

Keep the language plain. If a rule needs a paragraph of explanation, that's often a sign it should be shown with an example instead.

If you have read about Agent Skills, the idea will feel familiar. An Agent Skill is a folder with a SKILL.md file that an agent loads only when the task calls for it. You can read more in What Are Agent Skills?. A Callstand applies the same thinking to the AI apps you already use day to day, and lets other people plug in.

We'll also be publishing guides on two related ideas soon: Prompt Debt, the cost that builds up when your standards only live in old chats, and the Strategize-Draft-Critique loop.

When the Brief-Once Workflow is the wrong answer

The Brief-Once Workflow isn't the right fit for every use of AI. Skip it when:

• You only use one app, alone, for one ongoing body of work. If everything you do happens in Claude and nobody else touches it, Claude Projects with good project instructions may be all you need. The same goes for ChatGPT Projects if you live in ChatGPT.
• The task is a one-off. Writing a brief for something you'll never do again is overhead. Just prompt it.
• You don't know your standards yet. If you're still figuring out what good looks like, a brief will lock in guesses. Do the task by hand a few times, notice what you keep correcting, then write the brief.
• The work depends on judgment that changes daily. A brief holds what stays true. It won't replace the call you make about this specific client on this specific day, and it shouldn't try to.
• You expect it to fix weak inputs. A brief tells the AI how to work. It doesn't supply facts it was never given. If the source material is thin, the output will be too.

Frequently asked questions