Getting started, properly

You don't need Claude explained. You need to feel, once, what it does with real work, and then you're off. So here's the whole orientation in three sentences.

Claude is a conversation that can read what you give it (documents, spreadsheets, screenshots, whole folders), make things for you (files, reports, designs, small working tools), and increasingly do whole jobs on its own while you review. Quality in, quality out: the more real material and honest context you hand over, the better it performs. Everything else in this series is just those three abilities, deepened.

Don't "have a play". Run these three jobs with your own material. They cover the three abilities, and each teaches you something the next one uses.

Job 1 · Interrogate a document

Upload something real and dense: a contract, a report, meeting minutes, a policy. Then push past "summarise this".

Read this properly. Then:
1) the three things I actually need to know,
2) anything unusual or risky compared to what's standard
   for a document like this,
3) the questions I should be asking that I probably
   haven't thought of.

Point 3 is the one that changes minds. Asking Claude what you should be asking is the cheapest expertise you'll ever buy.

Job 2 · Make a real file

Ask for something you'd actually use this week: a one-page summary of that document as a Word file, a comparison spreadsheet, a slide outline. Claude builds it in a side panel and you download the finished file. Don't polish it in chat forever; get to 90% and finish it in your own tools.

Job 3 · Rewrite something in your voice

Paste two or three emails you've written and like, then a draft you're unhappy with, and say "rewrite the draft to match how I write in the examples". This is the pattern behind half of all daily Claude use: showing beats describing.

The model dropdown is the only setting worth understanding on day one. Same conversation, different engine.

The habit that matters: match the model to the stakes, not to habit. Drafting fifty product descriptions on Opus wastes your usage allowance; making a hiring decision on Haiku wastes something worse.

Plans & limits, in one honest table

How limits behave: think of a budget that refills on a short cycle (a few hours), with an overall weekly ceiling on the higher tiers. Heavier models and agentic work drain it faster, which is the practical reason the model habit above matters. Exact numbers shift too often to print; the pattern doesn't.

The single biggest upgrade to Claude's output is upstream of any prompt: upload the actual material instead of describing it.

• Documents: PDF, Word, Excel, PowerPoint, text, code. Drag them into the chat.
• Screenshots: Claude reads images properly. A screenshot of an error, a chart, a form, or someone else's layout beats a paragraph describing it.
• Several at once: upload the contract and the amendment and last year's version, and ask what changed and what it means.

Claude doesn't just answer; it produces. Documents, spreadsheets, presentations, PDFs, and working interactive pieces, built in a side panel while the conversation continues.

• Ask for the format you'll deliver in. "As a Word document", "as a spreadsheet with one tab per region", "as a one-page PDF". You get a file, not a wall of chat text to copy out.
• Iterate on the artifact, not from scratch. "Make section two harder-edged", "add a column for renewal date". It edits the thing in place.
• Small tools count too. A mortgage comparer, a rota builder, a quiz for your team: Claude can build little working apps that run right in the panel. Guide 05 goes deep.

Claude's built-in knowledge has a cut-off date. For anything where "as of when?" matters, make it look things up.

Twenty minutes of setup, repaid on every conversation after. Three layers, from broadest to most focused.

1. Preferences. In settings, tell Claude once how you like to work: "British English, no waffle, tell me when you're unsure". Applies everywhere.
2. Memory. Claude builds a picture of you and your work across chats, and you can tell it things to always remember. You can view, edit and delete what it holds, and it stays out of private chats.
3. Projects. A workspace per client, product or topic. Upload the reference files once, write project instructions once, and every chat inside starts fully briefed. This is the feature that separates casual users from people getting real leverage; it has its own guide (03).

Some standing instructions, please remember these:
I write in British English. When I ask for a document,
default to concise and plain-spoken. If my request is
ambiguous, ask me one clarifying question instead of
guessing. When you're not confident about a fact, say
so plainly.

Claude is now a family of surfaces, and the trick is simply knowing which one owns which job. Here's the whole map, with the guide that goes deep on each.

You don't need most of these yet. You need to know they exist, so that the day you catch yourself doing something manual and repetitive, a bell rings.

Before you paste client work into anything, you deserve straight answers. Here they are, plus the one habit that keeps you right.

• Training is a setting, not a default you can't see. On personal plans, you choose whether your chats help improve future models, and retention follows that choice. On Team and Enterprise plans, business data isn't used for training. Either way it's visible in Settings → Privacy, not buried.
• Private chats mean it. They skip your history, never feed memory, and are retained only briefly. Use them for anything you don't want carried forward.
• What Claude "knows about you" is a page you can open. Memory is viewable, editable and deletable (guide 03). Nothing accumulates behind glass.
• Connected apps see exactly what you granted and nothing more; audit the list quarterly (guide 04).

Claude is genuinely good, and it fails in patterns. Learn the patterns and you'll never be caught out; skip this section and you eventually will be.

• It can state wrong things confidently. Rare, but real, and most likely on precise facts: numbers, dates, names, citations, quotes. The tone gives no warning. Anything that must be right, verify or make it search and cite.
• It fills gaps helpfully. Leave something ambiguous and it guesses rather than interrogating you, and the guess is plausible. Say "ask me questions first" for anything with real stakes.
• It's agreeable by default. Present a plan and it looks for merit. When you want the holes, demand them: "be the critic, tell me why this fails".
• Each chat starts fresh except what memory and Projects carry. Chats don't see each other; long chats can lose their thread. New topic, new chat.

Habits that separate week-one users from year-one users. Steal them.

1. Upload, don't describe. The real file, the real screenshot, the real thread. Every time.
2. State the goal, audience and format in the first message. Three commas' worth of effort, double the quality.
3. Never settle for draft one. "Shorter", "warmer", "more sceptical", "give me three versions". Iterating is the normal way to use it.
4. Ask what you should be asking. On any unfamiliar ground, Claude-as-advisor beats Claude-as-answerer.
5. Make it search when currency matters, and glance at the citations.
6. Set up one Project this week for your most repeated context. You'll never go back.
7. New topic, new chat. Context is a workbench, not a wheelie bin.
8. When output disappoints, add an ingredient, don't re-roll the same prompt and hope.
9. Graduate from answers to work. The moment a task feels mechanical and repeated, check the map in section 08; there's probably a surface that just does it.