Most people use Claude the same way they use a search engine — they type a question, get an answer, and close the tab.
The result is fine. Occasionally impressive. But rarely transformative.
The business owners who get disproportionate results from Claude do something different. They don’t ask it questions. They give it jobs — with a proper brief, the right context, and a clear output format. This approach is what separates people who find Claude occasionally useful from people who use Claude for business in a way that genuinely compounds.
The difference between a question and a job is the difference between a colleague who answers when spoken to and one who goes off and handles the whole thing.
This post covers exactly how to make that shift: what a proper job brief looks like, how to structure it, and why most AI prompts fail before they have started.
Why questions get you partial answers
When you ask Claude a question, it answers the question.
That sounds obvious — but the problem is that the question you can write in one sentence is almost never the full problem you are trying to solve.
“What should I do about our pricing page?” returns general advice about pricing pages. Claude has no idea about your product, your customers, what you have already tried, or what the actual business objective is.
A consultant would ask clarifying questions before offering anything. Claude does not, unless you build those clarifying questions into the brief yourself.
The solution is to reframe the interaction entirely. Stop asking questions and start assigning work. Using Claude for productivity at any serious level requires this shift — it is the single biggest unlock for business owners who feel like they are not getting enough from their AI setup.
What a job brief looks like
A proper job for Claude has four components. You can apply this framework to almost any task — writing, analysis, planning, decision-making, research, or client work.
1. Role and context
Tell Claude who it is in this situation and what the context is. Not “you are an expert copywriter” — that is too generic — but something specific: “You are helping me improve a pricing page for a London-based productivity consultancy. Our primary audience is service-based business owners with £100K–£2M revenue. Our current page has a 1.4% conversion rate and three pricing tiers.”
Context is not background noise — it is the instruction. The more precisely you describe the situation, the more precisely Claude can navigate it.
2. The actual task, stated as a deliverable
Replace open questions with specific output requests. Instead of “what should I do about pricing?” write “rewrite the headline and the three tier descriptions so they emphasise outcomes over features. Keep each tier description under 60 words.”
The more specific the deliverable, the less editing you do afterwards.
3. Constraints and what to avoid
This is the part most people skip. Claude’s default outputs often drift toward a generic, polished version of what it thinks you want. You bring it back to your actual requirements by stating the constraints explicitly: word count or format limits, tone restrictions, what not to include, and any regulatory or compliance requirements.
A policy that bans certain language should be described precisely — not just as a list of banned terms, but as a definition of how those terms are used problematically. Claude applying a keyword sweep is not the same as Claude applying a compliance check, and the difference shows up directly in the quality of the output.
4. How to surface uncertainty
When you assign a real job to a real person, they ask questions before they start, not halfway through. Build that into every brief: end with “if any part of this brief is unclear before you begin, list your questions — do not make assumptions.”
This single instruction eliminates most of the mid-task tangents and wrong-direction outputs that frustrate people about working with AI.
The brief that changed how a brand handled content
A business we work with was using Claude to review their marketing copy against their own brand guidelines. They gave it a list of restricted terms and asked it to flag any violations.
It came back with fourteen hits. They reviewed every one. Twelve were false positives — the term was being used as vocabulary, not as the specific product claim their guidelines actually restricted.
The problem was the brief. They had given Claude a keyword list and asked it to flag matches. What they needed was a compliance check — flag only instances where the term is used as a product claim, not as neutral description.
When they rewrote the brief with that distinction in place, the same document came back with two genuine violations and no noise. Same document, same guidelines, completely different output — because the brief was precise about what the rule actually meant.
That is the difference a proper job brief makes. If you are dealing with the kind of overwhelm that comes from constantly correcting outputs rather than using them, the root cause is almost always an imprecise brief — not the AI itself.
Jobs Claude handles well once briefed properly
The jobs-not-questions principle applies across almost everything a business owner handles regularly.
Email and communication drafts. Give Claude the context (who this is to, what relationship, what outcome you need), the constraints (length, tone, what has already been said), and the exact output format. It produces a first draft that needs minimal editing rather than something to rewrite entirely.
Meeting preparation. Give it the agenda, the attendee list, the meeting objective, and any prior context. Ask it to produce briefing notes, likely questions, and a suggested decision framework. Good time blocking means protecting the 20 minutes before a meeting for this kind of briefing — not scrambling in the car.
Content review. Give it your style guide (not just the rules, but what the rules mean), the piece to review, and specific output instructions — a marked-up edit, a list of issues with explanations, or a rewritten version. The output matches the brief, not Claude’s default.
Research synthesis. Give it the question it needs to answer, the materials you have already gathered, and the format for the output — a summary table, a decision recommendation, or a risk list. This is where the jump from “ask a question” to “assign a job” is most obvious: research outputs become structured decisions rather than paragraphs you have to process yourself.
Weekly planning. Give it your task list, your three rocks for the week, your calendar constraints, and your energy pattern. Ask it to produce a weekly review that surfaces what is missing, what is at risk, and what the single most important action is before Friday. A question like “help me plan my week” produces far less than a proper brief does.
The context system that makes jobs scale
Giving Claude a good brief once is a start. The business owners who get compounding value do something more: they build context into persistent workspaces that load at the start of every session.
This is what the CLAUDE.md file does — it carries your operating context (who you are, how you work, what you care about) so you never have to re-brief Claude from scratch. Pair it with a live task queue and every session picks up where the last one ended.
Once you have that infrastructure, every job brief becomes shorter — because the standing context is already loaded. Claude already knows who you are, who you serve, and how you make decisions. You only need to specify the task. As I explored in context beats prompts, the quality of the context you give Claude matters more than the cleverness of your prompts.
Start with one job this week
Pick one task you currently handle by asking Claude a question. Rewrite it as a job brief using the four-component structure above: role and context, specific deliverable, constraints, and how to surface uncertainty.
Run both. Compare the outputs.
The difference will tell you everything about how much you have been leaving on the table.
If you want a structured way to wire Claude into your whole business — not just individual tasks but your entire operating system — the NP System + Claude Implementation walks through how to set it up across every context in your work and life. Book a demo call to see it running live.
Frequently asked questions
What is the difference between a Claude prompt and a job brief?
A prompt is a single instruction or question. A job brief is a structured document that gives Claude the context, the task, the constraints, and a way to surface uncertainty before it starts — the same information you would give a capable person doing the work for the first time.
How long should a Claude job brief be?
Long enough to remove ambiguity, not longer. Most effective job briefs are 100–300 words for a standard task. Longer briefs work well for complex or high-stakes deliverables where precision matters more than speed.
Can I save job briefs so I do not have to rewrite them?
Yes — and you should. Recurring jobs become templates. Store them in Claude project context so they load at the start of every session. The CLAUDE.md system is designed for exactly this purpose.
What happens when I do not specify the output format?
Claude defaults to what it thinks is most useful, which is often a generic format. Specifying the output format — a numbered list, a table, a 300-word paragraph, a marked-up draft — is one of the highest-leverage parts of a brief. It eliminates a full round of reformatting after the fact.
Does this approach work for non-technical tasks?
The jobs-not-questions method works on anything a knowledge worker does: writing, planning, analysis, client communication, decisions under uncertainty. The structure is the same regardless of the domain — the only thing that changes is the specific context and deliverable you describe in the brief.
For further reading, see the guide on automating tasks with Claude AI.
