The Complete Guide

Guide

All of Claude

What Claude is, how it works, what it can do, what it can't, where you use it and how to start. Written assuming you know nothing; anyone who reads to the end starts out knowing exactly what they are working with.

Reference page · jump to any section

This is an independent guide written by FY. It is not an official Anthropic page and there is no partnership between them. Product names belong to their owners. For anything that changes — prices, limits, features — always check the provider's own page.

The short answer

Claude is an AI assistant built by Anthropic; it works with text.
It writes, summarises, translates, codes, reads documents and generates ideas — you get work done by talking to it.
It can be confidently wrong, it learned up to a certain date, and every conversation starts from scratch.
You use it from a browser, a phone, a desktop app and a terminal; there is a free tier.
The secret to using it well isn't intelligence, it's context: who, what, why, how.

01What Claude is, who made it

Claude is an assistant built by an AI company called Anthropic. Its technical name is «large language model»: software trained on an enormous amount of text that works with language. It is not a search engine, a database or a robot.

You get work done by talking to it. It reads what you write, works out what you want and produces an answer. You can explain things exactly as you would to a person — no technical vocabulary required.

02How it works

First it read

Books, articles, discussions, code. It memorised none of them; it learned how they are written — how ideas, sentences and code follow one another.

Then it predicts

It doesn't fetch the answer from somewhere. It looks at the text in front of it, works out what is most plausible next, then one more word. That's why answers stream in rather than appearing at once.

As much as it sees at once

It can look at everything accumulated in one conversation at the same time; this is called the «context window». It is very large but not infinite; once full, the oldest parts start to blur.

03What it can do

Almost anything that can be done with text. The most common:

Writing

Emails, listings, product descriptions, proposals, presentation copy, subtitles, lesson plans, social posts.

Rewriting

Shortening, expanding, simplifying, changing tone, making it formal, fixing grammar.

Reading and summarising

Long reports, contracts, email threads, articles. It flags the risky clause and works out what is being asked.

Teaching

It explains what you don't know at whatever level you want, answers your questions, quizzes you, builds a plan.

Translating

Between languages — and with tone and audience taken into account. Ask in Turkish, have it write in German.

Writing code

Programs from scratch, finding bugs, line-by-line explanation, data conversion scripts. You don't need to know how to code.

Putting things in order

Messy notes into a table, a conversation into bullet points, a jumbled list into a clean file.

Reading images

Screenshots, photos, handwritten notes, invoices, charts. «What does this error say?» is a valid question.

Interpreting data

Paste a table and ask what stands out, which month is bad and what might explain it.

Thinking with you

It generates ideas, argues against your own, asks what you've missed, builds a checklist before a decision.

04What it can't do, where it gets things wrong

It can look certain and be wrong. Rather than staying silent when it doesn't know, it may produce a plausible-looking answer; this is called «hallucination». You can't tell from its tone, only by verifying.
It doesn't know the present. Its training stopped on a certain date. Without a search tool it doesn't know yesterday's news, a live price or current regulation.
It doesn't remember. Every conversation starts from scratch; earlier ones aren't carried over by default.
It is weak at exact counting and long arithmetic. Use a tool for character and word counts and long chains of multiplication and division.
It can invent sources. A book title, article heading or page number that doesn't exist. Click every link.
It doesn't act in the world on its own. It won't send email, save files or make payments — unless tools are connected.
It doesn't replace a professional. On medical, legal and financial matters it prepares you, but it doesn't decide for you.

05Where you use it

Browser

claude.ai — create an account and start straight away. The right starting point for most people.

Phone

The mobile app. For taking a photo and asking about it on the move, dictating a note and picking up where you left off.

Desktop

An app installed on your computer. Opens instantly with a shortcut; like a colleague who stays open all day.

Terminal — Claude Code

The version that works directly with the files on your computer: it sets up projects, writes and edits code. Chapter 4 of the course is built on this.

In your own software

You can connect it to your own application through the API — chat on your site, automatic classification in your system.

Inside automation

It plugs into workflows through tools like n8n: reading an incoming email and drafting a reply, for instance. Chapter 3 of the course.

06Which version

There isn't a single Claude; there are versions of different sizes and their names change over time. Remember the tier logic, not the name:

Most capable

The Opus family

Hard reasoning, long and complex work, serious code. The best thinker; the slowest and most expensive.

Balanced

The Sonnet family

Most everyday work. The best balance of speed and capability; the right default for most people.

Fastest

The Haiku family

Simple, short and highly repetitive work. Answers instantly and costs the least.

The rule: start with the default. If answers feel shallow, move up a tier. Don't decide in advance; the work will tell you.

07How to start

1Go to claude.ai and create an account. The free tier is enough to begin.
2Open a new conversation and ask what it does: «Tell me in five points what you can do, I'm just starting out.»
3Give it real work. Describe your most boring, most repetitive task and do it together.
4Upload a document. A PDF, a screenshot or a table — say «summarise this in three points».
5Ask something you already know the answer to. See where it's strong and where it's weak on day one.

08How to write a good request

This is the single thing that determines the quality of what you get back. Four parts are enough:

Weak

Write me an email.

To whom, why and in what tone is unclear. It fills the gaps itself and an average text comes out.

Strong

I run a small furniture workshop (who). I need to email a corporate client who is three weeks late paying (what). I want a firm date without damaging the relationship (why). Polite, short, 120 words maximum (how).

This gives you an email you can actually send. The difference isn't intelligence, it's context.

Two more habits: add «say when you're not sure» to the end of your request — it lowers the chance of invention. And if the first answer misses, don't open a new conversation; say what's wrong: «too formal», «drop the third paragraph», «shorter».

09Privacy and safety

Don't type data you aren't allowed to share. Passwords, ID numbers, card details, customer lists. The test: would you show this to an intern?
What happens to your data depends on the plan and the settings. Open your account's privacy settings once and read them; the rules get updated over time.
What you upload can steer the assistant. A document or web page may contain instructions aimed at it. If an answer drifts somewhere you didn't ask for, look at the source, not the assistant.
Responsibility sits with the user. You are the one using the output; verify every sentence that touches a decision, money or a person.

10Frequently asked

Is it free?

There is a free tier and it is enough to begin. Paid tiers exist for more usage and more capable versions. Because prices and limits change, the current position is on the provider's own page.

Does it know Turkish?

Yes. Ask in Turkish and get Turkish back; you can also explain in Turkish and have it write in German, or have an English document summarised in Turkish.

Can it look things up online?

Not on its own. It will if the app you're using has given it a search tool. If you're unsure, just ask: «are you saying this without searching?»

Is what I write used for training?

It depends on the plan and your account settings, and the rules get updated. The correct answer is in your privacy settings and the provider's current privacy text.

How long can a conversation be?

Very long, but not unlimited. Once the limit is reached the oldest parts blur. In a long conversation, having it summarise and moving to a new one is the cleanest method.

Why does it give different answers to the same question?

Because it isn't reciting a stored answer; it produces one afresh each time. That isn't a fault — «try again» is a legitimate option when you don't like what you got.

Why does it refuse some requests?

It stops at work that could cause harm. That isn't personal; usually there is a workable version of what you want, and it will tell you why.

Does it know the facts about itself?

Not always. It can be wrong about which version it is, when its training stopped or what the daily limit is; it can't see itself from the outside. Get those from the app's own page.

11Glossary

The words that come up when AI is discussed. All of them appeared on this page:

Prompt: The text you write to the assistant. A «good prompt» means a request with the context filled in.
Model: The assistant itself; the trained software. Opus, Sonnet and Haiku are model families.
Token: The small piece of text the model actually sees. Smaller than a word — which is why it can't count letters.
Context window: The total text it can look at in one conversation. When it fills up, the oldest parts weaken.
Hallucination: Producing a plausible-looking but wrong answer where it doesn't know. The most common and most treacherous kind of error.
Training cut-off: The moment its learning stopped. It doesn't know anything after that date on its own.
Tool: An external capability connected to the model: searching, reading files, sending email.
Agent: A setup that carries out multi-step work on its own using tools. Chapter 3 of the course.
API: The way to connect the model to your own software. You talk to it in code instead of a chat window.
System prompt: A standing instruction given up front: its role, tone and rules. Chapter 2 of the course.
Multimodal: Being able to read images as well as text.
Fine-tuning: Specialising a model for a particular job with extra data. Most work doesn't need it; a good request is enough.

In one sentence

Claude isn't an all-knowing oracle but a very well-read, very fast assistant: give it context, give it what you have, speak plainly, verify what comes back.

Chapter 1 · The Wake-Up

This page told you what you're dealing with. The course teaches you to use it in your own work — from zero, step by step, with exercises.

Open Chapter 1