DESK · THEORY
WorkflowBeginner · June 4, 2026 · 8 min read
On this page

Give it context: the one habit that multiplies every answer

Context is one of the four ingredients in every prompt. It's also the one most CEOs skip. This article goes deep on that single ingredient, shows you what it does, and hands you a reusable block you fill out once.

The anatomy of a prompt has four ingredients: task, context, format, and constraints. And the full briefing method covers how to put all five parts together. This piece zooms in on one: context. It's the biggest lever, and it's the one left on the table in probably 80% of prompts you've sent.

What you'll have when you're done

A reusable context block you fill out once, save, and paste (or load automatically) into every AI conversation. Once it's in, the model stops writing for the statistical average person and starts writing for you.

The outputs will feel different. Specific where they were generic. Calibrated where they were cautious. Firm where they were mealy. The same model, much better results.

The CEO getting advice meant for someone else

Here's what happens without context. You open Claude and type: "Write an email declining a vendor renewal."

What you get back is professional. It's well-structured. It also leaves a door open that you specifically don't want left open, because the model has no idea you've pushed back twice already and you've already signed a replacement. It hedges because hedging is the statistically safe move when it knows nothing about your situation.

With no specifics, the model defaults to the median answer, optimized for nobody. This is not a flaw. It's arithmetic. The model predicts the most useful response given what you gave it. Give it nothing, it writes for the average person asking the average version of your question.

Anthropic's prompting guidance frames the model as "a brilliant but new employee who lacks context on your norms and workflows." That's the right mental model. A new hire on day one will give you technically correct work that misses all the unstated norms, the history with a person, the tone your company uses, the constraints baked into your situation. You wouldn't hand that person a two-word task. Same logic applies here.

OpenAI's prompting guidance makes it plain: the more detail you give, the better the response, and examples teach expectations better than description. The guidance is consistent across providers because the underlying mechanic is the same.

Context collapses the output from "answer for the average person" to "answer for your situation." That collapse is the whole game.

What you need first

No technical background needed. Context is just information. You already have it.

Step by step

What context actually is

Context is everything the model cannot see. There are six categories worth knowing:

The no-context vs. rich-context demo

Same task. Two prompts. Read both.

Bare prompt:

Write an email declining a vendor renewal.

What you get: generic, polite, placeholder-heavy. Could be any company. Probably leaves the door open with a line like "we look forward to potential future opportunities." No firm close. The rep can exploit the ambiguity on the next call.

Rich prompt:

I run a $15M/year B2B distribution business. We used [Vendor] for warehouse analytics for two years. Performance was mediocre and we have already signed a replacement. Their rep is persistent and we have pushed back twice verbally. Draft a firm, professional one-paragraph email declining their renewal. No door left open. My style is direct and short.

What you get: specific, appropriately firm, matches the voice, removes the ambiguity the rep could exploit. The email sounds like it came from someone who has already made the decision.

The request is identical. What changed is who is asking and why. The model didn't get smarter. It got briefed.

Build your reusable context block

This is the deliverable. Copy it. Fill in the brackets. Save it somewhere you'll find it.

## About me
Name: [First name]
Role: [Title / function]
Communication style: [e.g. direct, concise, no jargon]

## About my company
Company: [Name]
What we do: [One sentence]
Size / stage: [Revenue range or headcount, stage]
Industry: [Sector]
Current priority: [Top 1-2 goals right now]

## Who I am usually writing for
Internal: [e.g. my leadership team, front-line managers]
External: [e.g. SMB operators, enterprise buyers, consumers]

## How I want output
Default tone: [e.g. confident, peer-to-peer, not too formal]
Default length: [e.g. short unless I ask for detail]
Things to avoid: [e.g. jargon, filler, hype]

Fill this out with real information. Not "large company" but the actual revenue range. Not "team" but who specifically. Not "professional" but what professional means to you: formal and measured, or direct and short?

The more specific you are, the harder the block works. A block that says "I run a $15M B2B distribution business focused on cutting SG&A by 20% this year" does something a block that says "I run a mid-size company" cannot do.

A note on the context window: keep your standing context block focused on what genuinely applies to most tasks. Pasting your entire company strategy document, your org chart, and six months of board minutes into every prompt crowds out the task-specific context that matters right now. Standing context should be short, accurate, and stable. Task-specific detail gets added at prompt time.

How you'll know it's working

The output uses your specifics. It names your industry, your audience, your situation. If you could swap the output into any other company's context with no edits, your context block is too thin.

You stop rewriting for voice. The tone is already close. You're editing for accuracy, not for "make this sound like me."

Drafts that previously needed two rounds of revision land in one. Context does most of the work that revision was doing before.

When it breaks

You filled the block in with placeholders. "Large company" and "professional tone" are not context. They're the model's defaults anyway. Be specific or leave the field blank; a vague field takes up space without doing work.

The standing block is stale. Your priorities shifted in Q2. Your target audience changed. The block still says last year's focus. Update it. A wrong context block is worse than no context block because it actively misdirects.

In-the-moment context and standing context are getting mixed up. The meeting notes from Tuesday, the specific email thread, the one-off constraints of this particular request: those belong in the prompt, not in your standing block. Standing context is what's true in most sessions. Task context is what's true right now. Keep them separate.

Too much context crowding the useful part. If you're pasting walls of background and the model still misses the point of the task, your context may be too long and too unfocused. Cut to the four or five sentences that matter most.

Level up

Once the context block is working, the next move is loading it automatically so you stop pasting it every time.

Two ways to do this:

Projects (Claude). Claude Projects let you write your context block once in a Project's instructions. Every conversation inside that Project starts with it already loaded. You open a new chat, and the model already knows who you are, what the company does, and how you want output formatted. No paste, no repetition.

A standing-context file. If you're working in the terminal or in Claude Code, you can load a Claude file that carries your context block. Every session in that file's scope starts briefed. If you want to understand how that memory works across sessions, memory and Projects for beginners covers the mechanics.

Either way, you fill the block out once. After that it's there before you type your first word.

That's the compounding part. The first time you paste the block manually and get a dramatically better answer, it feels like a trick. When it's loaded automatically and every conversation starts already knowing your situation, it becomes infrastructure.

Fill out the block this week. Paste it into your next real task. You'll see the difference in the first answer.

The Thursday 3

Get three workflows like this every Thursday

The Thursday 3 is a free weekly email. Three workflows that put you in the top 1% of CEOs. 90-second read. Every card links back to a step-by-step guide like this one.

Get the newsletter →
The Desk Theory books

The architecture behind this workflow.

Two operator manuals for the same job, run two ways: OpenCLAW for the always-on harness, Claude Code for the focused-work CLI. Pick one, or get the bundle for $149.

Browse the books · $99 each

Want one workflow like this taken apart end-to-end every week? The Tuesday Pro Deep Dive · $39/mo.