Three minutes of repeating yourself every time you open ChatGPT is the silent productivity tax that almost everyone using AI in 2026 is paying. Every new chat starts with the same setup. "Act as a [role], I work in [industry], I write for [audience], my company does [description], the project is [context], and please avoid [list]." Three minutes here, three minutes there, ten chats a day, and you have just wasted half an hour reminding the model of things it should have already known.
This post fixes the tax. The fix is a framework called the Context Stack. Five reusable context bricks that you build once, save somewhere accessible, and assemble into the context bundle each new conversation needs. The first build takes about an hour. Every conversation after pays the time back tenfold.
The Context Stack is what pros use that beginners do not. The beginner repeats. The pro pastes their stack. The output quality matches the difference.
Why You Keep Repeating Yourself
The reason you keep retyping the same setup is that the model has no persistent memory of you by default. Each new chat starts blank. Even when the model has features called "memory" (ChatGPT memory, Claude Projects, Gemini Gems), the features capture only a fraction of what you would actually want it to remember, and the gaps are exactly where the repetition shows up.
The fix is not to wait for the products to give you better memory. The fix is to externalize your context into reusable blocks you control, then load the right block at the start of each conversation. The model gets the context. You stop typing the same setup.
This is what "context engineering" actually means in practice. It is not a buzzword. It is the discipline of treating your own context as a small set of reusable assets, the way a developer treats environment variables. Once you do this, the model stops feeling generic. It starts feeling like it knows you, because it does. You just gave it the file.
The Context Stack Framework
The Context Stack splits the context you need to feed AI into five distinct bricks. Each brick is a self contained block of text you write once and reuse. Together they cover roughly 90 percent of the context the model needs to give you sharp, calibrated output.
The five bricks are Identity, Voice, Domain, Project, and Defaults. They are not sequential. They stack and combine. Different work needs different combinations. A LinkedIn post needs Identity, Voice, and Defaults. A vendor negotiation email needs Identity, Voice, Domain, and Project. A strategic memo needs all five.
Most professionals end up with one version of each brick plus two or three project specific variants. Six to ten files total. They live in a notes app, password manager, or AI memory feature, and they get pasted at the start of any conversation that matters.
Brick 1: The Identity Brick
The Identity Brick describes who you are in a way the model can use. Not biography. Not LinkedIn summary. Just the facts the model needs to interpret your requests correctly.
What it includes. Your role and seniority. Your function and the kind of work you do. Your decision authority and what is above your head. Your team or company context at the level needed for the model to know what kind of answer is appropriate.
What it does not include. Personal details unrelated to work, full company description, history. The Identity Brick is operational, not biographical.
Example. "I am a senior commercial director at a global agency, managing a single global account across multiple markets. I lead pricing, scoping, and commercial strategy. I have authority to negotiate within agreed frameworks. Anything outside frameworks goes to my direct leadership and legal. I work fully remote. My direct counterparts on the client side are procurement leads and a senior global ops director."
Why it matters. Without identity context, the model defaults to a generic answer. With identity context, the model adjusts vocabulary, depth, and assumptions. The same question about pricing produces a different answer for a senior commercial director than for a startup founder. Both answers are correct. Only one is useful to you.
Where it lives. Best stored in your AI memory feature if available, or as a custom instruction at the account level. Pasted at the top of any chat where it is not auto loaded.
Brick 2: The Voice Brick
The Voice Brick describes how you communicate. It is the externalized version of the voice profile prompt that the Voice Clone Method post covers in depth.
What it includes. Specific observable features of your writing. Sentence rhythm. Punctuation habits. Words you use, words you avoid. Tone defaults. Constructions you reach for, constructions you reject.
What it does not include. Vague style descriptors ("professional but warm"), generic tone words ("conversational"), or comparisons to famous voices.
Example. "Use contractions in most sentences. Open paragraphs with the conclusion, then explain underneath. Sentences vary between under 10 words and around 25 words. No em dashes. No hyphens to join phrases. No bullet points unless I ask. Plain language, no corporate adjectives. Avoid: seamless, robust, leverage, unlock, in today's, exciting times. Prefer the second person 'you' over 'one' or 'people.'"
Why it matters. Without the Voice Brick, every piece of writing the model produces sounds like every other piece of AI writing. With the Voice Brick, the model starts producing first drafts that read as yours and need 20 percent of the editing time you would otherwise spend.
Where it lives. Same as Identity. Memory feature, custom instruction, or pasted at the top of writing tasks.
Brick 3: The Domain Brick
The Domain Brick captures the specific knowledge context that surrounds your work but is not common knowledge.
What it includes. Your industry vocabulary. The acronyms, frameworks, and named entities you encounter daily. Your specific tools, platforms, or systems. Key relationships, accounts, or stakeholders. Anything the model would not know without you telling it.
What it does not include. Information freely available on the internet. The model already knows your industry exists. The Domain Brick captures the specific corner of it you operate in.
Example. "I work in agency commercial operations. Key acronyms: MSA (master services agreement), SOW (scope of work), CCTM (commercial campaign management), OPEX (operating expense), CAC (customer acquisition cost). Key frameworks I use: RACI for accountability, fixed price vs retainer vs T&M pricing, rate card construction. Key stakeholders: agency CFO, client procurement, client global ops, regional GMs. Common pain points: scope creep, deferred work piling into quarter end, retainer drift."
Why it matters. The Domain Brick is what makes the model sound like a colleague rather than an outsider. The output uses your vocabulary, recognizes the entities you mention, and skips the explanations of things you already understand. The conversation moves faster because the model is at your level.
Where it lives. Domain Bricks tend to be too big for AI memory. Better to keep them in a notes file you paste into chats where domain context matters.
Brick 4: The Project Brick
The Project Brick captures the specific situation of whatever you are working on right now. It is the most frequently updated of the five bricks.
What it includes. What the project is. Who is involved. What the current state is. What has happened recently. What is coming up. The constraints and trade offs that are currently active.
What it does not include. Distant history, unrelated context, or speculation about future projects. The Project Brick is current.
Example. "Current project: pricing a new agency capability for the EMEA region. Client procurement is pushing for a fixed price not yet agreed. My team is overcommitted and a new hire is two months out. The capability has been delivered once before in NA with a CCTM equivalent we can reference. The decision deadline is end of next week. My internal stakeholders want me to hold the line on a retainer model. The client side has not yet been formally briefed."
Why it matters. The Project Brick is the difference between asking the model "help me think about pricing this" and asking the model "help me think about this specific pricing situation with the constraints I am facing." The second prompt produces a working answer. The first produces a textbook.
Where it lives. Project Bricks rotate. Keep the current ones in a project specific notes file. Archive completed ones. Update the live ones weekly.
Brick 5: The Defaults Brick
The Defaults Brick captures your standing preferences for how the model should produce output. It is the "house style" of your AI interaction.
What it includes. Default output format. Default length. Whether you want a draft or a critique first. Whether you want the model to ask clarifying questions before producing. What kind of response you usually do not want.
What it does not include. Anything that varies by task. The Defaults Brick captures only the preferences that hold across most of your work.
Example. "Default output preferences. For analytical tasks, lead with the conclusion in one sentence, then show reasoning underneath. For drafts, produce a v1 first then ask me what to refine before producing v2. For lists, prefer numbered over bulleted. Do not produce executive summary preambles unless I ask. Do not produce closing flourishes like 'I hope this helps.' Do not hedge with 'it depends' or 'consider' as the main answer. When uncertain, state the uncertainty and propose what additional information would resolve it."
Why it matters. The Defaults Brick eliminates 90 percent of the format and tone corrections you would otherwise apply on every output. The model produces output that matches your standing preferences from the first draft. The downstream editing time collapses.
Where it lives. Best stored as a custom instruction at the account level so it loads automatically. If your model does not support that, paste it at the top of every chat.
How to Assemble the Bricks Into Context Bundles
The five bricks combine differently for different work. The combinations are not random. They follow the kind of task you are doing.
For most quick prompts and chats, load Identity, Voice, and Defaults. That is the minimum viable context bundle. Three bricks. Twenty seconds to paste. Output quality jumps immediately.
For writing tasks (drafts, memos, posts, emails), add the Domain Brick when the topic is industry specific. Skip it when the topic is general.
For project specific work (analysis, planning, decisions tied to current work), load all five bricks. The Project Brick is what makes the work specific. The others provide the calibration.
For brainstorming, you sometimes want fewer bricks rather than more. Identity and Voice are enough. The Defaults Brick and Domain Brick can constrain the model when you want it expansive.
The five brick model is a maximum. Most conversations need three or four bricks. The discipline is not loading all five every time. The discipline is knowing which three or four to load.
Where to Store the Bricks
The Context Stack only works if the bricks are accessible in under five seconds. Otherwise you will not use them.
AI memory feature. Best home for Identity, Voice, and Defaults if your model supports it. ChatGPT memory, Claude Projects, and Gemini Gems all work. The brick loads automatically. No copy paste required.
Custom instructions or system prompt. Second best home for Identity, Voice, and Defaults. The brick loads on every new chat in that account or project. Cost is one setup. Benefit is forever.
Notes file or password manager. Best home for Domain and Project Bricks because they are bigger and update more often. Paste at the start of relevant chats. Keep a "Bricks" note that has all five so you can copy whichever combination you need.
Files attached to the conversation. If the model supports file context (Claude Projects, ChatGPT custom GPTs, Gemini Workspace), uploading the full brick library as a reference file means the model can pull from it during conversations. This is the most advanced setup. Most people do not need it.
The 60 Minute Setup
Building the full Context Stack takes about an hour the first time. The setup is the only painful part.
Block 60 minutes. Open a notes file. Write the Identity Brick first. 10 minutes. Write the Voice Brick next, using the Voice Clone Method post if needed. 15 minutes. Write the Domain Brick by listing the acronyms, frameworks, and entities that appear in your work weekly. 15 minutes. Write the Defaults Brick by listing the format and tone preferences you find yourself correcting most often. 10 minutes. Write the current Project Brick. 10 minutes.
Total. 60 minutes. You now have the asset.
The 60 minutes is the entire upfront cost. Every conversation after benefits. Most people who run the setup find they save the 60 minutes back within the first week. The savings compound from there.
What Most People Get Wrong
The mistake is treating the Context Stack as a one time setup that you do once and forget. It is not.
Identity changes when your role changes. Update annually.
Voice changes as your writing evolves. Update quarterly.
Domain changes as your work moves into new areas. Update when you notice the model not knowing something it should know.
Project changes constantly. Update weekly or when context shifts materially.
Defaults rarely change. Update when you notice you are correcting the same thing across many conversations.
The Context Stack is a living asset, not a static one. The 60 minute setup is the entry cost. The ongoing maintenance is 10 minutes a week if you are doing it right.
Frequently Asked Questions
Does the Context Stack work in Claude and Gemini as well as ChatGPT?
Yes. The Context Stack is model agnostic. Claude Projects, ChatGPT custom GPTs, and Gemini Gems all support similar persistent context features. The bricks transfer cleanly. Most people store the same Identity and Voice Bricks in two or three models and load them based on which model they reach for.
How is this different from prompt templates?
Prompt templates are scaffolds for specific tasks. Context Bricks are reusable blocks that go above any prompt. You combine them. A prompt template plus an Identity Brick plus a Voice Brick produces calibrated, task specific output. Templates and bricks layer well with the 12 prompt patterns that structure the template itself.
Can I share my Context Stack with colleagues?
The Domain Brick is shareable across team members in the same role. The Defaults Brick is sometimes shareable as a team house style. Identity, Voice, and Project Bricks are personal and should not be shared. Some teams maintain a shared Domain Brick for the function and let each person add personal bricks on top.
Will AI memory features make the Context Stack obsolete?
Not in the near term. AI memory features capture a fraction of what a deliberate Context Stack captures, and they do it less reliably. The features will continue to improve, but the Context Stack discipline transfers to any memory model. The discipline is the durable asset, not the storage location.
What is the minimum viable Context Stack?
Identity, Voice, and Defaults. Three bricks. About 30 minutes of work. Anyone using AI for serious work should have these three. Domain and Project are higher leverage for specific roles but optional for casual use.
How long should each brick be?
Identity: 100 to 200 words. Voice: 200 to 400 words. Domain: 200 to 500 words depending on field complexity. Project: 100 to 300 words. Defaults: 100 to 200 words. The full stack is usually 700 to 1500 words. Long enough to be specific. Short enough to paste without thinking.
Should the Context Stack include things I would not want a model to see?
No. Anything genuinely confidential should not go into a Context Stack unless the model is covered by a data processing agreement that meets your security and compliance requirements. The Context Stack is for work context the company is comfortable being assisted on. Confidential matters need separate handling.
Get the Context Stack Templates
The PromptLeadz Context Stack Pack ships pre built brick templates for common roles (founder, marketer, sales lead, operations lead, finance director, technical writer, creator), plus the assembly guide for combining bricks into bundles for different work types. Every template is editable and formatted three ways for Claude, ChatGPT, and Gemini.
Browse the Context Stack Pack and the rest of the PromptLeadz library in the shop. Free starter brick templates in the Freebie Vault.
コメントを残す: