You already have an account. Or you are about to make one. The first half hour is where most people either get a useful page of output, or bounce between model names, app stores, and a half-written prompt that never becomes something you can send.
This is Part 3 of Learn Claude from scratch. Part 1 covered what Claude is in plain English. Part 2 covered Free, Pro, and Max in practical terms. Here we stay off Code installs, API keys, and product rabbit holes. The goal is boring on purpose: open Claude, know which surface you are on, run one real work task, and leave with a result you can edit and ship.
If you need LLM vocabulary first, read What are LLMs, ChatGPT, generative AI, and more. For the wider habit stack (context, verification, paste safety), the Practical AI series is the companion track. Learning paths also sit on Learn.
What you will leave with
- A map of Claude on web, desktop, and mobile (and what each is good for on day one)
- Account gates that matter: age 18+, supported location, plan you are on
- Where the model name, effort, and thinking controls live so you stop guessing
- A timed first-session path that produces one editable deliverable
- A starter prompt pack you can paste without inventing a second career in prompt craft
- What to skip today so you do not derail into Code, Cowork, or API setup
Before you open the tab
Claude is Anthropic’s assistant product. You reach the everyday chat experience at claude.ai. Anthropic also ships a desktop app for Mac, Windows, and Linux (Linux is still in beta) and mobile apps for iOS and Android. Same account can move across those surfaces.
Two gates people hit late and blame on “the model”:
- Age: you must be at least 18 to use the services.
- Location: you need to be in a supported location. If login fails or features look missing, check Anthropic’s supported locations article before you rebuild your prompt for the fifth time.
Pick one device for this session. Web on a laptop is the easiest teaching surface because the layout shows model controls and file attach options without fighting a tiny keyboard. Mobile is fine for follow-up and capture. Desktop is useful when you want a dedicated window and later product flavors. Day one does not require all three.
Choose one job for the half hour
Vague goals create vague chats. Write a one-line job on paper or in a sticky before you type anything into Claude:
- “Turn my messy notes into a 400-word project update for stakeholders.”
- “Outline a 20-minute teach-back on weekly revenue for non-analysts.”
- “Rewrite this cold email so it is shorter and less salesy.”
- “Build a study checklist from this PDF chapter (text I am allowed to use).”
If the job needs confidential customer data, stop. Use a redacted sample or synthetic numbers. Paste discipline gets a full post later in this series. For now: if you would not put it on a shared slide, do not paste it into a free-form chat on day one.
Where Claude lives: three surfaces
Think of surfaces as doors into the same account, not three different products with three different personalities. Features still vary by plan and by how mature a feature is on each device.
| Surface | How you open it | Best day-one use | Watch-outs |
|---|---|---|---|
| Web | claude.ai in a browser | First real task, file upload, model and effort controls, Projects later | Browser extensions and password managers can clutter the UI; use a clean profile if something looks wrong |
| Desktop | Download from Claude’s download page for Mac, Windows, or Linux (beta) | Dedicated window, less tab noise, later Cowork-style work on paid plans | Skip deep desktop-only features until chat feels boring and useful |
| Mobile | Claude app for iOS or Android | Capture notes, continue a thread, short rewrites on the train | Harder for long PDFs and careful editing; model selector lives at the top of the screen |
Sign in with the same account on each surface. Chats generally follow the account, not the laptop. That is the whole point of starting on web and finishing a polish pass on the phone later.
What you are not installing today
Claude’s product map has grown: chat, Code, Cowork, Design, Science, browser connectors, and more. Those deserve their own series parts. For this half hour, treat anything that wants a terminal, a local folder grant, or an API key as out of scope. If the UI offers a shiny new product door, close it and stay in ordinary chat.
The first 30 minutes, timed
Use a kitchen timer if you like. The point is not speed-running. The point is finishing one loop: input → draft → edit → usable file.
Minutes 0 to 5: land and orient
- Open claude.ai and sign in (or create the account if you have not).
- Confirm you can see a blank chat and a text box.
- Find the model name. On web and desktop it usually sits near the text input. On mobile it sits toward the top. Click it once so you know the selector exists. Leave the default unless you already know you need a different model for a paid workflow.
- If you see effort or thinking controls, leave them at defaults for this first pass. You can tighten later when a task is actually hard.
- Glance at plan indicators (Free versus paid) so usage limits do not surprise you mid-draft.
You do not need a tour of every menu. You need to know two things: which model is answering, and that you can change effort or thinking when a job deserves it.
Minutes 5 to 12: load context the honest way
Gather the raw material for your one job. Prefer short and real over long and “complete.”
- Bullet notes from a meeting
- A draft email you hate
- A public-safe PDF or doc you are allowed to use
- A screenshot of a slide outline (if the surface accepts images)
On web, use the + control near the composer to attach files or photos when that is faster than pasting a wall of text. Supported document types commonly include PDF, DOCX, CSV, TXT, HTML, and several others. Images usually include JPEG, PNG, GIF, and WebP. Spreadsheets (XLSX) may need code execution and file creation enabled in settings. Limits change over time; if an upload fails, shrink the file or paste a slice.
If you are moving from another assistant and care about long-term preferences, Free, Pro, and Max users can import memory from other AI providers on web and desktop. That is not a full chat history import. Anthropic’s own get-started docs are explicit: conversation history from another provider does not import wholesale. Memory import is the supported path, and it is experimental. You can ignore it for this half hour. A good first result does not depend on imported lore.
Minutes 12 to 22: one strong first prompt
Paste a structured ask, not a vibe. Here is a template that works for almost any first deliverable:
Role: You are helping me draft for human review, not publishing for me.
Context:
- Audience: [who will read this]
- Goal: [decision or action I want]
- Tone: [plain, short, no hype]
- Constraints: [word limit, no invented stats, flag unknowns]
Source material:
[paste notes or describe the attached file]
Task:
1) Draft the deliverable in the format I asked for.
2) List assumptions you made in bullets.
3) List anything you could not verify from the source.
Refuse:
- Do not invent numbers, quotes, or sources.
- Do not claim the draft is final.Run it. Read the whole answer once without editing. Then do a second pass as a human editor:
- Delete anything you would not defend in a meeting.
- Replace vague verbs (“optimize,” “enhance”) with the real action.
- Check every number against your notes or file.
- Ask a follow-up only for the broken pieces: “Shorten the second section to 80 words and keep my three risks.”
Two or three tight follow-ups beat one giant “make it better” message.
Minutes 22 to 30: export and plant the next habit
Copy the final draft into the place it actually lives: email, doc, ticket, slide outline, LMS note. Title the Claude chat with the deliverable name and date so you can find it later. If the product offers renaming, use it. “New chat” is how useful work disappears.
Optional five-minute mobile check: open the mobile app, find the same chat if it synced, and try one small edit request (“make the subject line less corporate”). You are proving the account travels. You are not redoing the whole task on a phone.
Worked example: stakeholder update from messy notes
Suppose your notes look like this (toy example, not a real company dump):
- Q2 pilot: 3 stores, East region
- Dashboard v1 live Tuesday; still missing returns
- Support tickets up ~12% week over week (ops estimate, not finance)
- Next: freeze metric defs Friday with analytics + ops
- Risk: store managers using spreadsheet shadow metrics
- Ask: 2 hours from data eng for returns field in warehouseA weak first prompt:
Write a project update about the pilot.That invites invented timelines, fake “success,” and numbers that sound precise because language models like precision theater.
A stronger first prompt using the template:
Role: Draft for human review. I will edit before sending.
Audience: VP Retail Ops and analytics lead.
Goal: Clear status, risks, and one ask. No celebration copy.
Tone: Plain. Short paragraphs. No buzzwords.
Constraints: 350-450 words. Do not invent metrics beyond my notes.
If a number is approximate, keep it approximate.
Source notes:
[paste the bullets above]
Task:
1) Draft an email body with subject line options (3).
2) Sections: Status, Risks, Next steps, Ask.
3) Assumptions list.
4) Unknowns list.
Refuse: no invented ROI, no fake quotes from store managers.What a solid draft should feel like after your edit: the ~12% ticket rise stays labeled as an ops estimate, returns stay missing, the ask is two hours of data eng time, and nothing claims the pilot “drove growth.” If Claude softens a risk into a virtue, put the risk back. You own the email.
Controls worth knowing on day one
Model selector
The UI shows the model you are talking to. Click it to switch among models available on your plan. For a first session, default is fine. Switching models mid-task can change voice and thoroughness. If you switch, say so in your own notes so you are not comparing apples to mangoes later.
Effort and thinking
Anthropic exposes settings for how much effort the model puts into a response and whether extended thinking is on. Use more effort or thinking when the problem is multi-step reasoning without needing a web crawl (tricky rewrite constraints, careful comparison of two options you already pasted). Use ordinary chat for straightforward drafting. Use web search or Research (paid, covered in Part 4) when the answer depends on current external sources. Matching the control to the job saves both time and usage.
Usage limits
Free plans use session-style limits that reset on a schedule (Anthropic documents a five-hour style reset for free usage, with other fairness limits possible under demand). Paid plans raise the ceiling. When you hit a wall, the product will say so. Do not interpret a limit message as “you prompted wrong.” Save your draft elsewhere and continue when the window opens, or upgrade if this is daily work. Part 2 of this series is the plan chooser.
Starter prompt pack for the first week
Copy these into a note app. Swap the brackets. Keep the refuse lines.
1. Meeting notes to action list
Turn these notes into:
- Decisions (only if clearly decided)
- Open questions
- Action items with owner if named, else "owner missing"
Do not invent owners or deadlines.
Notes:
[paste]2. Explain like I have ten minutes
Explain [topic] for a smart coworker who is new to it.
Use: plain words, one analogy, three pitfalls, a 5-item checklist.
No hype. Flag where experts disagree if you are unsure.3. Rewrite shorter
Rewrite the text below to [N] words.
Keep my meaning and numbers.
Remove filler and corporate padding.
Show a short "what I cut" list after the rewrite.
Text:
[paste]4. Study sheet from allowed material
Using only the attached / pasted material:
- 10 quiz questions (answer key after)
- Glossary of 8 terms in one sentence each
- What I should re-read if short on time
If the material does not support an answer, say "not in source."5. Plan a small project
I need a plan for [goal] with about [hours] of work.
Audience: [me / team].
Deliver: phases, risks, definition of done, what to skip.
Do not invent tools I did not mention.
Context:
[paste]Web versus mobile in practice
People ask which surface is “best.” That is the wrong question for week one. Match the surface to the friction you have.
| Situation | Prefer | Why |
|---|---|---|
| First structured prompt with a file | Web or desktop | Keyboard, multi-line edit, easier attach |
| Polish a paragraph between meetings | Mobile | Low setup cost |
| Long PDF with charts | Web, larger screen | Reading and citing pages is less painful |
| Voice-of-notes capture then cleanup | Mobile then web | Capture fast, structure later |
| Anything involving codebases or local folders | Not day-one chat | That is later product territory |
If mobile feels cramped, you are not failing. You are noticing that long-form drafting still likes a real keyboard. Use the phone as a remote control for chats you started on web.
Common first-session mistakes
| Mistake | What happens | Fix |
|---|---|---|
| No single job | Five partial chats, zero sendable output | One line goal before open |
| Prompt as wish | Pretty fiction | Role, source, format, refuse |
| Paste secrets “just this once” | Risk you cannot unsay | Redact or use fake sample |
| Switch models every reply | Inconsistent drafts, burned usage | Stay put until the draft stabilizes |
| Trust numbers without a source line | You become the error | Check against notes or file |
| Chase every product tile | No first win | Chat only for 30 minutes |
| Leave chat untitled | You cannot find the good one | Rename with deliverable + date |
| Expect full history import from another AI | Frustration | Memory import only; rebuild key chats as needed |
How this connects to the rest of the series
Part 4 covers chats versus Projects, files that stick across conversations, and Research for multi-source digs with citations. Parts after that go into everyday writing jobs, memory and privacy, light workplace connectors, and when Claude is the wrong tool. The product map series (Code, Cowork, Design, Science) waits until ordinary chat is already useful.
If your work is analytics-heavy, keep the Practical AI habits close: assist, do not replace; verify SQL and numbers; treat paste as a governance act. Claude on web does not cancel those rules. It just gives you a fast drafting partner.
What a finished first session looks like
You should be able to point at four artifacts when the timer ends:
- A deliverable outside Claude (email draft in your mail client, doc in Drive, ticket comment, outline in notes).
- A named chat you can reopen tomorrow without scrolling mystery titles.
- A short note of what the model invented or softened, so you know where you spent editing time.
- A saved prompt template that worked better than your first try.
If you only have a long chat and nothing exported, the session is incomplete. The product is sticky on purpose. Your real job lives elsewhere: the inbox, the board deck, the LMS, the ticket queue. Claude is a station on the way, not the destination.
A second tell that the session worked: you can explain in one sentence what you asked for and what you still distrust. “I asked for a status email from my notes; I still need finance to confirm the ticket spike.” That sentence is more valuable than a confident paragraph you cannot defend.
FAQ for day one
Do I need a paid plan for this half hour?
No. Free is enough to learn the loop: prompt, edit, export. You may hit usage limits sooner on Free. That is annoying, not a signal that you failed. Save the draft outside Claude and continue later, or upgrade when the limit keeps blocking real weekly work. Plan choice was Part 2; do not reopen that decision mid-timer.
Should I turn on every tool in the + menu?
No. Day one is ordinary chat plus an optional file attach. Research, connectors, and product-specific modes can wait until you have a reason. Extra tools add extra ways to burn usage and attention. Part 4 covers Research and Project knowledge when you are ready.
The answer sounds smart. How do I know it is wrong?
Check the parts that can embarrass you: numbers, names, dates, claims about what “the team decided,” and any citation-like flourish. Language models are good at smooth certainty. Your notes are the source of truth for this exercise. If a claim is not in the source, it is either an assumption (which should be listed) or fiction (which should be deleted).
Can I import all my old ChatGPT threads?
Not as full conversation history. Anthropic documents memory import for Free, Pro, and Max (and related plan notes for Team in the memory articles), not wholesale history migration. Memory is a summary of durable preferences and facts, and the import flow is experimental. For day one, rebuild the one chat that matters by pasting the source material again. That is faster than fighting an import you do not need yet.
Web or mobile if I only have twenty minutes?
Web, if you have a keyboard. Mobile if you only have a phone and a short rewrite. Twenty minutes on mobile with a long PDF is a recipe for frustration. Match the surface to the task size, not to the marketing screenshot of people smiling at phones.
Practice for the next two days
- Repeat the 30-minute loop on a second real task with a different format (email → outline, or outline → study sheet).
- Install mobile if you have not, open one existing chat, and make a single constrained edit.
- Write a personal “never paste” list of three items (customer IDs, credentials, health data, whatever applies to you).
- Save your best first prompt as a template in a note app before you forget which wording worked.
Quick recap
- Use web for the first serious task; keep mobile for capture and light edits; treat desktop as optional day one.
- Confirm age and supported location before blaming the product.
- Know where model, effort, and thinking settings live; defaults are fine for the first win.
- One job, structured prompt, human edit, export out of the chat.
- Skip Code, API, and deep product installs until chat is already paying rent.
- Memory import is not full history import; ignore it if it slows the first deliverable.
Sources
Product behavior and limits change. Confirm details in Anthropic’s help center when something in the UI does not match this guide.
- Anthropic Help Center, Get started with Claude: https://support.claude.com/en/articles/8114491-get-started-with-claude (access surfaces, age 18+, model selector location, effort and thinking, memory vs history import)
- Anthropic Help Center, Where can I access Claude (supported locations): https://support.claude.com/en/articles/8461763-where-can-i-access-claude
- Anthropic Help Center, Upload files to Claude: https://support.claude.com/en/articles/8241126-upload-files-to-claude (types, limits, project vs chat files)
- Anthropic Help Center, Import and export your memory from Claude: https://support.claude.com/en/articles/12123587-import-and-export-your-memory-from-claude
- Anthropic Help Center, How do usage and length limits work?: https://support.claude.com/en/articles/11647753-how-do-usage-and-length-limits-work
- Claude product entry: https://claude.ai/ and desktop download: https://claude.com/download
- Analytics Made Simple, What are LLMs, ChatGPT, generative AI, and more: https://analyticsmadesimple.com/analytics/what-are-llms-chatgpt-generative-ai-and-more/
- Analytics Made Simple, Practical AI series: https://analyticsmadesimple.com/series/practical-ai/
- Analytics Made Simple, Learn: https://analyticsmadesimple.com/learn/
