The 10 Best Prompts for Claude Cowork
10 prompts that do more in Claude Cowork than in the browser, because the local workspace keeps context you'd otherwise re-explain every session.
Claude Cowork has one core advantage over Claude in the browser: the workspace lives locally. Claude reads files directly, writes to folders, and knows the brand guides or templates stored there. Unlike browser Claude, where every session starts with re-explaining who you are, this context stays in Cowork.
The ten prompts below are built for freelancers and solo operators who want to use Cowork productively. The cowork-specific prompts assume that a brand guide, templates, or similar files sit in the connected folder.
Writing & revising
1. Draft text in brand voice
Write a LinkedIn post about [topic]. First read my brand guide from the workspace folder. Stick to the tone, banned phrases, and length defined there. No emojis, no bullet points, no marketing hype. Write in first person.
This prompt uses the workspace directly. Claude reads the brand guide before writing. No copy-paste into the context needed. If you need several formats (LinkedIn, newsletter, blog), you can run the same prompt with a different target format, since the context persists across sessions.
2. Revise existing text
Here's a text of mine: [paste text]
Revise it by these criteria:
- Short sentences for emphasis, then longer ones. Vary the rhythm.
- No worn phrases (no "Let's...", "Let's take a look", "At the end of the day")
- No marketing fluff ("game-changing", "seamless", "groundbreaking")
- Prefer active constructions
- Don't invent or add anything, only revise what's already there
Give me the revised version and mark the three biggest changes.
This prompt guards against a common AI trap: Claude likes to add new content when asked to "improve". The "don't invent" line stops that. Asking for the three biggest changes forces Claude to make a choice instead of delivering everything at once.
3. Turn raw text into publishable content
I have these raw notes from a client call / a session / my thoughts: [notes]
Build them into a [blog post / newsletter section / LinkedIn post].
Keep all concrete details and examples from the notes, that's the most valuable part.
Don't add anything that isn't in the notes.
Tone: direct, informal, no empty structure.
This prompt turns unstructured thoughts into publishable content without losing the concrete details and examples. Claude translates here. He brings the notes into shape but doesn't invent anything. More on why that matters for authentic AI content.
Research & structure
4. Structure a topic before writing
I want to write about [topic]. Audience: [description].
Give me three outline options:
- Option A: beginner angle (why is this relevant?)
- Option B: practitioner angle (how does it work concretely?)
- Option C: critical angle (what doesn't work as advertised?)
For each option: 4–5 H2 sections with one sentence describing what the reader learns there. No drafted prose, only structure.
Deciding the angle before writing pays off. This prompt delivers three real alternatives, not a shallow list of aspects. The "structure only, no prose" constraint stops Claude from jumping into writing mode before the direction is set.
5. Collect counterarguments to a thesis
I'm writing about [topic]. What are the strongest counterarguments to the thesis that [my thesis]?
Give me:
- 3–4 concrete counterarguments with reasoning
- For each: one question a skeptical reader would ask
- At the end: the strongest counterargument in one sentence
Don't invent sources. If you don't know concrete sources, say so directly.
Whoever only looks for evidence supporting their own thesis writes a weak text. This prompt flips the direction so you can test your argument against the hardest objections. The "don't invent sources" line is necessary: Claude hallucinates sources if you don't explicitly stop him. The reasoning behind it is in the prompting basics.
Business communication & email
6. Write a professional reply email
I'm replying to this request: [paste email / message]
Write a reply that:
- Addresses the concrete question directly (no small-talk opener)
- Names my next step clearly
- Is at most 150 words
- Sounds professional but not stiff
My background to the request: [optional context, e.g. price range, availability]
AI-generated emails are usually too long and start with a sentence no human would write. The word limit forces Claude to prioritize. Cowork users can also store an email template in the workspace and point the prompt at it.
7. Draft a difficult message in three variants
I need to send the following message: [context, e.g. announce a price increase / decline a project / push back a deadline]
Write me three variants:
- Variant A: direct and short (under 80 words)
- Variant B: with a bit more context and reasoning
- Variant C: with a concrete alternative proposal
No defensive phrasing, no apologies for normal business decisions.
The three variants show the spectrum: sometimes terse works, sometimes context helps, sometimes an alternative proposal is the right move. Which one fits depends on the recipient and the situation.
Workflow & planning
8. Break weekly goals into concrete tasks
My goals for this week: [list]
Break each goal into concrete, actionable steps.
Each step should:
- Start with a verb
- Be doable in under 30 minutes
- Be clear without needing more context
Mark steps that block others (dependencies). Estimate total time per goal.
"Write newsletter" isn't an actionable task; "draft outline for April newsletter" is. The dependency marking helps plan the order, especially when several projects run in parallel. If you use Cowork for planning, you can pull the result straight into the app's todo list.
9. Build a content calendar for a month
I need a content calendar for [month].
Constraints:
- Platforms: [e.g. LinkedIn, newsletter]
- Frequency: [e.g. 2x LinkedIn per week, 1x newsletter]
- Topic clusters: [e.g. AI tools, freelance workflow, client projects]
- Already planned content: [if any]
Give me an overview with date, platform, topic, and one sentence on the core message of each post. No drafted text, only the calendar.
Content planning with Claude works best with clear constraints. Without frequency and platform inputs, you get a theoretically complete, practically unusable calendar. Anyone with the prompting fundamentals sees quickly why the constraints are what matters. The finished calendar can sit in the Cowork workspace and be built on in later sessions.
Cowork-specific: workspace and persistent context
10. Set up a workspace for a project
I'm setting up my Cowork workspace for [project / client name].
Create the following files in the workspace folder:
1. brand-guide.md, with these inputs: [tone, audience, banned phrases, style examples]
2. project-context.md, with these inputs: [project goal, current status, open questions]
3. templates-overview.md, list of available templates with one sentence on usage each
Keep the files compact. Each file should work as a context reference, not as documentation.
People setting up Claude Cowork for the first time often underestimate how much of the tool's value sits in the workspace setup. The lift over browser Claude only emerges with well-structured context files. This prompt builds the foundation: three compact files Claude can read in every following session without you retyping the context.
Prompts are not a shortcut for thinking
One misunderstanding sticks: good prompts replace the thinking about what you actually want. That isn't true. A good prompt assumes that decision is already made. What it takes off your plate is the wording, the structure, the working through variants.
The prompts here are starting points. Each one works better when you fill the placeholders with real context and read the result once critically before copying it. Claude Cowork gives more room for that process than browser Claude, because the context persists between sessions.
If you want to go deeper into using Claude, the Cowork field report has a fuller take, and the model comparison sits in ChatGPT vs. Claude. If you want the difference between AI assistant and agent, the answer is in AI assistant and AI agent.
FAQ
- Why do these prompts work better in Cowork than in the browser?
- Because Cowork's workspace lives locally. Claude reads your brand guide, templates, and project files directly from the connected folder, so context persists across sessions instead of being re-explained every time you start a chat.
- How do I stop Claude from inventing content when revising?
- Add an explicit instruction like 'don't invent or add anything, only revise what's there'. Claude tends to add new content when asked to improve a text, and it will hallucinate sources unless you tell it to say so when it doesn't know any.
- Do good prompts replace thinking about what I want?
- No. A good prompt assumes that decision is already made; what it takes off your plate is the wording, the structure, and working through variants. Fill the placeholders with real context and read the result critically before using it.
