The Cowork playbook covered the guardrails, the sandbox rule, and the trust ladder. This one gives you the prompts.
If you have been reading about Claude Cowork and have not yet tried it on real work, the fastest way in is to copy one of the five prompts below into a fresh Cowork session and give it a folder to work in. Each of these takes about an hour of setup and produces something you would have paid a summer intern to do.
A quick reminder before you paste anything
Make a dedicated folder for the project. Put only the files you want Cowork to see in it. Keep "Act without asking" off for the first pass through any of these prompts. You are not automating; you are watching Cowork work so you learn what it is actually good at.
Prompt 1: the SOP set audit
Setup: create a folder called sop-audit. Copy your current standard operating procedures into it. If you have ten SOPs, copy ten files. If you have fifty, copy fifty. Do not point Cowork at your live SharePoint or your live Google Drive.
The prompt:
Read every file in this folder. Do not rewrite anything yet. Identify (1) contradictions between documents, where two SOPs describe the same process differently, (2) references to tools, people, or vendors we no longer use, (3) voice or terminology drift where the same idea is called different things in different files. Produce a single fix-list I can review before you edit anything.
What to expect: Cowork will open each file, take notes, and produce a table of findings. Expect between fifteen and forty items depending on how long you have been running the SOPs. Read the fix-list. Reject the ones that are not real issues. Approve the rest.
Why it works: this is exactly the kind of cross-file work Claude chat cannot hold in context. A chat thread would run out of room around the fifth or sixth document. Cowork keeps the whole set in view at once.
Prompt 2: the customer-question theme extraction
Setup: create a folder called ticket-themes. Export the last quarter's support tickets or customer emails to text or PDF. Copy them into the folder.
The prompt:
Read every ticket in this folder. Do not answer any of them; that is not the goal. Group them by the underlying question the customer is actually asking, not by ticket subject line or product tag. For each group, tell me (1) how many tickets fall into it, (2) the phrasing customers most often use, (3) one representative example. Produce the groups ranked by ticket volume.
What to expect: you will see the five or six real questions your customers are asking, in their own language, with a count next to each. This usually rearranges how you think about your product's rough edges. It also gives you the exact language for a FAQ, help doc, or product page that would prevent tickets from being written in the first place.
Why it works: the intelligence here is grouping by underlying question, not surface topic. Cowork holds thirty or forty tickets in context and clusters them the way a human reader would. Chat cannot do this at scale.
Prompt 3: the quote-vs-website reconciliation
Setup: create a folder called pricing-audit. Save your current /pricing page as a PDF or HTML file into it. Save the last ten to twenty client quotes as PDFs into the same folder.
The prompt:
Compare the prices on our pricing page with the prices in our recent client quotes. For each service or product mentioned, tell me (1) what the pricing page says, (2) what the quotes say (with the range if there is a range), (3) whether these agree. Flag anywhere the pricing page is stale, undercuts the quotes, or references a service we are no longer selling. Do not fix anything. Produce the report.
What to expect: most businesses have pricing drift. This prompt surfaces it. You will find services on the website you have quietly stopped offering, prices that were increased in quotes but never on the site, and packages the sales team is quoting that the marketing team has never heard of.
Why it works: Cowork reads a live HTML export and a stack of PDFs in the same session and cross-references them. Nothing about that is exotic. It is just tedious enough that no one on your team has ever sat down and done it.
Prompt 4: the meeting-notes decision log
Setup: create a folder called decision-log-source. Copy your meeting notes from the last quarter or year into it. If they live in Notion or Google Docs, export them. If you keep them as emails, save the relevant threads as PDFs.
The prompt:
Read every meeting note in this folder. Extract only decisions. Not discussions, not action items, not open questions. Just decisions the team made. For each decision, capture (1) what was decided, (2) approximately when, (3) which meeting or thread it came from, (4) any reasoning that was given at the time. Produce a chronological decision log grouped by topic area.
What to expect: a two- to five-page document that captures the institutional memory of your last quarter or year. This is the document you want to hand a new operations lead or a new department head. It is also the document you can read yourself when you cannot remember why you set that policy or picked that vendor.
Why it works: distinguishing "decision" from "discussion" is judgment work. Cowork does it about as well as a thoughtful assistant would. Chat cannot hold the source material in context long enough to try.
Prompt 5: the three-competitor scan
Setup: create a folder called competitive-scan. Put a plain text file in it with three competitor URLs, one per line. Save your current /about page or positioning page as a PDF into the same folder. This is one of the few Cowork prompts that needs the Chrome extension, so enable it for this session only.
The prompt:
Visit the three URLs in the file. For each, extract (1) how the company positions itself in one sentence, (2) their stated differentiators, (3) their pricing if visible, (4) the target customer they seem to be writing for. Produce a comparison table. Then read our positioning page and tell me two things: what our page is missing that at least two of the three competitors emphasize, and what our page claims that a competitor already claims better.
What to expect: a real, honest competitive read. Cowork will not flatter your positioning. It will tell you where competitors have moved past you and where you have a clear angle they have not yet occupied.
Why it works: this is web reading plus local file reading plus judgment, all in the same session. Once you have done it, you will understand why owners who dabble in agentic AI are hooked. This is one of the things it does that nothing else lets you do in under an hour.
The pattern behind all five
Every one of these prompts has the same shape. A dedicated folder. A set of source files. A specific ask that requires reading across files and producing a synthesis. No rewriting or acting on the world until you approve the synthesis first.
That is the Cowork sweet spot. Persistent context across many files, judgment about what those files contain, a written artifact you can review. The autonomy features come later, once you have watched a few of these run and you know what Cowork is actually good at.
Try one this week. Pick the folder you have been meaning to clean up but never do, and run the prompt against it. An hour later you will have something worth reading.
For the safety and setup fundamentals behind all of these, start with the Cowork playbook. For a broader view of when to use Cowork versus other agentic tools, see Beyond Cowork: three other ways to deploy agentic AI in your small business.