The Claude AI Tutorial That Starts With Work Already on Your Desk

Let's skip the windup. The fastest way to learn Claude is to point it at real work you already owe someone: a call recap, an unread report, a board memo, a position you have to defend. Every one of those runs on the same three-part prompt: tell Claude who it's writing as, give it dense context, and name the exact output you want. Master that and you've got the tutorial. The rest is reps.

Most Claude tutorials burn your time on theory, show a haiku example, then hand-wave about "the possibilities." You close the tab and you're still staring at the same pile of work.

I'm going to walk through four scenarios from your own desk: a client call that needs a recap, a report you haven't had time to read, a variance memo for the board, and a position you need to stress-test before a hard conversation. For each one I'll give you the prompt structure, what you're after, and what to verify before the output leaves your hands. No toy examples. No warm-ups. Just the work.

A note on who this is for. If you're 45 or older with two decades of judgment behind you, you don't need a primer on neural networks. You need to know what to hand off and what to keep. Your experience is the moat. Claude is fast and tireless and confidently wrong about things you'd catch in a heartbeat. The whole game is pairing its speed with your judgment, not replacing one with the other. When you are ready to hand off whole tasks, see how to use ChatGPT's agent mode.

If you've read the foundational overview of how to use Claude, this is where those principles meet real situations. If you haven't touched Claude yet, The Leverage Starter gets you from zero to a working session in under two hours, starting with tasks exactly like these. For the plain-English groundwork first, start with a plain-English guide to using AI.

Why RCF Beats "Magic Prompts"

The same spine runs through all four scenarios: the RCF check, for Role, Context, Format. Claude doesn't read your mind. It reads what you give it. When the output is weak, the prompt is almost always the problem, not the model. Three gaps degrade output faster than anything else: vague context, missing role-framing, and no stated output format.

Role-framing tells Claude what perspective to take. "You are drafting this for a senior partner reviewing a complex tax position" produces materially different output than an unframed ask. Context is the background it needs to understand the situation: who the client is, what the subject matter is, what constraints apply. Format is what you want back: a one-page memo, a three-paragraph email, a structured analysis with headers, a list of objections. Cover all three and Claude can spend its effort on substance instead of guessing what you meant.

Here's the pattern in one table. Same RCF spine, four different jobs.

The scenarios below build all three into every prompt. You'll catch the rhythm fast.

Scenario 1: The Post-Call Recap Email

You just got off a 90-minute client call: board member, principal, wealth client, take your pick. Your notes are in whatever form your notes always take. Incomplete sentences, jargon, a couple of action items buried in the middle, a follow-up question you forgot to ask. You need a professional recap out the door within the hour.

The Prompt Structure

Open with the role: "You are drafting a professional recap email from a senior advisor to a client following a strategy call." Then dump context as a dense paragraph. Not bullet points, not cleaned-up notes, just the actual notes pasted in. Include the client name (first name only, or a pseudonym; never upload identifying information you don't need), the general subject, the key topics, and any action items or open questions you flagged. Then name the format: "Draft a recap of no more than 350 words. Professional but direct. Structure it as a two-sentence summary, key decisions, action items with owner and timeline, and one sentence on next steps."

That gets you about 80% of the way to a sendable draft in one pass. The last 20% is your judgment. Read it for factual accuracy. Confirm the action items are attributed to the right people. Check that the tone fits this specific client. Claude doesn't know whether your client wants formal salutations or whether there's history that reframes the whole thing. You do.

What to Verify Before Sending

Check every factual claim, every number, every attributed statement. Claude will occasionally merge two separate points into one, or soften a commitment that was actually firm. Read it the way you'd read any draft before it goes out under your name, because it is going out under your name.

Scenario 2: The Report You Haven't Had Time to Read

You've got a 40-page report (an appraisal, a due diligence package, a government filing, an opposing expert's report) and a client meeting in two hours. You haven't read it. You need to walk in with command of the material.

The Prompt Structure

Paste the full text, or as much as is relevant. Claude has a large context window; most documents under 100 pages paste cleanly. Then prompt: "You are a senior professional reviewing this document before a client meeting. Summarize the key findings, conclusions, and any material risks or open issues. Structure it as an executive summary of 150 words, key points by section (three to five bullets per major section), and a separate section flagging anything inconsistent, uncertain, or worth a follow-up question." Then: "Do not soften the findings. If there's bad news here, state it plainly."

That last line earns its place. Left alone, Claude reaches for diplomatic language. If you're walking into a negotiation or a tough client conversation and you need the worst-case read, you have to ask for it.

The prompt patterns for professionals post goes deeper on document-review prompts by type. Regulatory filings, contracts, financial statements, and expert reports each reward a slightly different structure.

What to Verify Before the Meeting

This doesn't replace reading the document. It's a way to arrive with a map of the territory. In the meeting you may need to surface specific language fast, and reading the summary tells you where to look. Afterward, read the sections that actually came up. The summary is your pre-brief, not your permanent record.

And never paste documents carrying client PII, tax IDs, account numbers, or anything that would be a confidentiality problem outside your walls. If the document has that material, redact it first or work from a sanitized version. The Fiduciary Firewall briefing covers the specific rules to understand before any client material touches any AI tool.

Scenario 3: The Variance Memo for the Board

You need to explain a variance. Revenue came in 18% below forecast, a project ran 30% over budget, a fund returned 600 basis points under the benchmark. The board wants a memo. You know exactly what happened. What you don't have is the time or the appetite to write three drafts of an explanation that will be read by people who go straight for the weakest sentence.

The Prompt Structure

Frame for the reader, not the writer: "You are drafting a variance explanation memo for a board of directors. The audience is sophisticated and will probe for gaps. Direct and analytical, no hedging, no passive voice, no burying the lead." Then the facts: what the metric was, the target, the actual result, the primary drivers, secondary factors, and the plan forward. Then the format: "400 to 500 words. Lead with the variance and the number. Explain causes in order of magnitude. Close with corrective actions and timeline. Do not start with background. Start with what happened."

What comes back is usually a structurally sound draft that organizes your explanation better than a cold start would. The substance is yours. Claude is imposing structure and economy on it, and that's worth something.

The Second Pass

Read the draft and ask: where will the board push back? Then run a second prompt: "Given this memo, what are the three most likely follow-up questions a board member would ask, and where are the weakest points in the current explanation?" Now Claude is a pressure-test, not a ghostwriter. The questions it surfaces are usually the right ones, because it's reading the same text the board will read and it has no emotional stake in the explanation you've given.

If you work in finance, the Leveraged CPA and Finance Professional course covers this workflow (structured writing for senior financial audiences) in more depth than a post allows.

Scenario 4: Stress-Testing Your Position Before a Hard Conversation

You're heading into deposition prep, a negotiation, a client conversation where the other side holds a strong hand, or a board meeting where you expect opposition. You have your position. You need someone to argue against it, hard and specifically, before you walk in.

I wrote off the "thinking partner" pitch as marketing until a messy client engagement changed my mind. I asked Claude to tear apart a recommendation I was sure of, and it found two gaps I'd talked myself out of seeing. This is where it quietly earns its keep: it's a very capable adversarial thinker when you tell it to be one.

The Prompt Structure

"You are an experienced opposing counsel / opposing analyst / skeptical board member [pick the role]. I'm going to describe my position on [subject]. Find every weakness in my argument, every fact I might have wrong, every assumption that could be challenged, and the strongest version of the counterargument. Don't be diplomatic. Don't validate my position first. Go straight to the critique." Then lay out your position in full, the way you'd present it in the room.

What comes back will be uncomfortable if your position has gaps. Good. Better to find the gap here than across the table. Follow up with "Now steelman my position. What are the strongest responses to those objections?" and "What would I need to find or verify to close the weakest points?"

This is a thinking-partner workflow, not a drafting one. You're not publishing the output. You're using it to sharpen your own thinking before a situation where weak thinking has consequences. Attorneys prepping for depositions, deal people walking into a term sheet, consultants delivering a recommendation no one wants to hear: this is exactly where Claude's total lack of investment in your position becomes a feature.

The deeper guide to using Claude at work covers thinking-partner mode in more detail, including how to run multi-turn conversations where you build an argument piece by piece instead of in one shot.

The Workflow Across All Four Scenarios

The pattern holds. Frame the role for the right reader or perspective. Hand over dense factual context, not cleaned up, just accurate. Specify the output format. Verify against your own knowledge before you use it. That's the whole tutorial. The four scenarios are just RCF applied to real work.

What shifts is what you're optimizing for. The recap email rewards tone and accuracy. The document summary rewards completeness and flag-raising. The variance memo rewards structural clarity and defensibility. The stress-test rewards adversarial completeness. The prompt bends with each job; the discipline underneath stays identical.

Do You Have to Become a "Prompt Engineer"?

No. The phrase irritates me. "Prompt engineer" has picked up a mystique it hasn't earned. What you're actually doing is communicating clearly about what you need and who you're speaking to. Senior professionals do this every day when they brief a junior associate, instruct a paralegal, or hand a task to a capable analyst. Claude is a capable analyst. Brief it accordingly, and skip the certification.

If you want to turn this into a systematic practice rather than occasional use, the memo-writing workflow post covers the writing scenarios in more depth, and The Leverage Starter walks you through your first structured session with work from your own desk. Two hours. You'll leave with three or four outputs you can actually use and a prompt pattern you'll keep reaching for.

One last thing, because it's the part most people skip. The professionals who get the most out of Claude aren't the ones who read about it longest. They're the ones who took something already on the desk this week, ran it through the RCF check, and built the habit from there. Pick one scenario. Do it this week.

What Claude actually is, in plain terms

Claude is an AI assistant built by Anthropic. You type or paste something in, it reads it and answers in plain language. It can read files you upload (PDF, Word, Excel, images), search the web, hold context across a piece of work, and produce writing, analysis, and summaries. For a senior professional the useful mental model is not "chatbot." It is a fast, literal junior analyst who has read everything you hand it, never tires, and needs clear instructions. It does not know your matter, your client, or your house style until you tell it. That is the whole game, and it is what the RCF check above is for.

Claude vs ChatGPT for professional work

The honest answer to "is Claude better than ChatGPT" is that they are close, and the right pick depends on the work in front of you. Both cost about $20 a month for the main paid tier, both read your files, both search the web, both keep persistent workspaces. Here is where the difference actually shows up for professional knowledge work.

The bottom line: for reading documents, careful analysis, and writing that goes to a client or a board, Claude is the stronger default. For image generation, voice, and broad general-purpose work, ChatGPT wins. Plenty of professionals keep both and use each for what it does best. Neither one removes your duty to check the work before it leaves your desk.

Your first ten minutes with Claude

You do not need a course to start, though a structured one shortens the path. This is the shortest route to a useful first result.

1. Go to claude.ai and create an account. The free tier is enough to test it on real work.
2. Pick one task already on your desk this week, not a toy example. A recap email, a document you have not had time to read, a memo you keep putting off.
3. Use the RCF check from the top of this guide: name the role, paste the real context, name the format you want back.
4. Read the first draft critically. Fix what is wrong by telling Claude what to change in plain words. It will revise.
5. Save the prompt that worked. Next week, reuse it. That is how the habit compounds.

If the first result comes back generic, the cause is almost always thin context, not the tool. Give it more of what you already know.

Projects: load your context once, stop re-explaining

A Project is a saved workspace that holds documents and standing instructions Claude reads in every chat inside it. You set it up once. After that you stop re-pasting the same background every time.

• A client or matter workspace holding the engagement letter, prior deliverables, and a standing instruction on tone and citation style, so every question you ask inside it answers in context.
• A recurring-deliverable workspace, for example the monthly board memo, seeded with the template and the last three versions so each new draft matches.
• A house-standard workspace holding your firm's policies and playbooks, so answers stay on-standard instead of generic.

Free plans allow up to five Projects. Paid plans remove the cap and hold far more reference material per Project.

Artifacts: when Claude builds the document, not just describes it

Artifacts are a side panel where Claude produces a finished thing you can use and refine, rather than a wall of chat. A document, a table, a chart, a simple interactive calculator, even a small web page. You edit it turn by turn, and you can share it by link without the other person needing an account.

• Draft a long memo or policy in a stable editing surface and refine it section by section, instead of scrolling back through a conversation.
• Turn a spreadsheet you uploaded into a clean chart or a one-page summary you can hand to a client.
• Build a simple client-facing calculator, for example a retirement or tax-scenario tool, without involving anyone technical.

What it costs, and the one privacy line that matters for client work

Pricing as of 2026:

• Free, $0. Enough to test Claude on real work, with limited usage and up to five Projects.
• Pro, $20 a month (about $17 a month billed annually). Higher usage, unlimited Projects, more model choice.
• Max, $100 or $200 a month for five times or twenty times the Pro usage, for heavy daily users.
• Team, about $25 per seat a month (minimum five seats), with shared Projects and admin controls.

Now the part that matters most for anyone handling client or patient information. On the consumer tiers (Free, Pro, Max), Anthropic uses your conversations to train its models unless you turn that off in Privacy Settings, and even with it off, conversations flagged for safety review can still be retained. For confidential client work that is the wrong setting. Team, Enterprise, and the developer API run under commercial terms that do not train on your content by default. The practical rule: test and learn on a consumer plan, but put real client matters on Team, Enterprise, or the API, and never paste identifying client data into any tool until you have confirmed where it goes.

That boundary, what goes in, what stays out, and how to set it up correctly, is the first thing our courses cover, because for a regulated professional it is not optional. Not sure which one fits your work? Take the two-minute selector.