Everyday storyThe well-read new joinee
Picture a new colleague who joins your office on Monday morning.
They have read an astonishing amount: textbooks, novels, newspapers, manuals, tax guides, cookbooks, code, in many languages. Ask them to explain compound interest, draft a polite reply to an angry customer, or turn your messy notes into a neat plan, and they do it in seconds, in clear sentences, without getting tired or irritated.
But they joined this morning. They have never seen your shop, your clients or your files. They don't know that your boss hates long emails, or that the Surat client always pays late. Unless you tell them, they will guess, and their guesses will sound very sure of themselves.
They also stopped reading the newspaper some months ago. Ask about something that happened last week and they may not know, or worse, they may fill the gap with something that sounds right.
And like any new joinee, you wouldn't hand them the office safe's combination on day one.
Now you know how to work with them. Tell them who you are and what you need. Give them the background they're missing. Check their work on anything that matters, especially facts, dates and numbers. Keep secrets to yourself. Do that, and this colleague becomes the most useful person in the building.
That colleague is a good way to think about Claude.
In a real productClaude on the web, desktop and phone
People across India already use AI assistants for everyday work, often quietly, without calling it anything special.
A kirana store owner in Indore asks for a short Hindi and English WhatsApp message announcing longer opening hours in Diwali week, then a second version for customers who buy on credit.
A Class 7 teacher in Jaipur asks for ten fraction word problems set in a sabzi market, with an answer key, at three difficulty levels.
An analyst at a Pune startup pastes a 30-page market report and asks for the five points that matter for a two-wheeler rental business, plus the questions the report doesn't answer.
A final-year student in Kochi asks Claude to explain a database topic three different ways until one finally clicks, then asks it to quiz them.
A junior at a CA's office in Nagpur asks for a plain-language explanation of a notice a client received, so they can brief the partner in five minutes instead of thirty.
What's at stake is simple to see. Done well, each of these saves somewhere between ten minutes and two hours, and the result is often better than a rushed version written alone. Done badly, the same tools produce confident mistakes: a wrong due date in a client reminder, a made-up statistic in a college project, or a customer list pasted somewhere it should never have gone.
The difference between the two is rarely the tool. It's knowing what Claude is, what it's good at, where it slips, and how to hold a good first conversation. That's this lesson.
The ideaHow it works
What Claude is
Claude is an made by Anthropic, an AI company. You talk to it in ordinary language, typing or speaking the way you would to a person, and it replies in ordinary language.
Under the hood, Claude is a . During training, it learned patterns from a huge amount of text: how sentences are built, how explanations flow, how a formal letter differs from a WhatsApp message, how code is structured, and a great deal of general knowledge along the way. When you send a message, it writes a reply piece by piece, each piece chosen to fit everything that came before.
Two things follow from that, and they explain most of Claude's strengths and weaknesses:
- It is very good with language. Explaining, drafting, rewriting, summarising, translating, organising messy notes, brainstorming and writing code are all forms of producing good text, and Claude does them well.
- It does not look things up by default. Unless a feature such as web search is switched on, Claude answers from what it learned in training. It isn't checking a database of facts as it writes.
Claude comes in a few versions. Anthropic has used family names such as Opus (most capable), Sonnet (a balance of ability and speed) and Haiku (fastest and lightest). The exact versions change every few months, so if an app lets you pick a model, choose from its current list rather than from an old article.
Where you use Claude
You can reach Claude in several places, and they share the same account:
- On the web, at claude.ai, in any browser. This is the easiest place to start.
- In the desktop app for Mac and Windows (with a Linux version in beta). It does everything the website does, plus a few extras covered below.
- In the phone apps for Android and iPhone, handy for quick questions, voice input and photos.
- Inside other tools. Claude can work inside Slack, inside Excel, Word and PowerPoint through an , and in Chrome through a . Developers also reach it through an , which lets their own apps send it messages.
There is a free plan and paid plans. Paid plans give you more use and some extra features. What each plan includes changes over time, so check claude.com for the current list rather than relying on anything written down here.
What a conversation is
Each time you open a new chat, you start a fresh conversation. You send a message, Claude replies, you reply to that, and so on.
Here is the part most people don't realise. Every time you send a message, Claude reads the whole conversation so far, from your first message to your newest one, and writes its reply with all of it in view. That is how it remembers what "make it shorter" refers to. The amount of text it can keep in view at once is called its . It is large, but it isn't unlimited, so a very long chat can eventually lose sharpness about the early parts.
A new chat, by contrast, starts with a blank page. Claude doesn't automatically carry over what you discussed yesterday. Some features change this: memory can bring in useful background from earlier chats, and projects (covered in a later lesson) hold files and instructions for a whole group of chats. But a safe habit is to assume a new chat knows nothing about you until you tell it.
You can also add more than words. You can attach PDFs, photos, screenshots and spreadsheets, and Claude will read them as part of the conversation.
What Claude is good at, and where it slips
Claude is at its best when the job is mostly about language and thinking:
- First drafts of emails, messages, letters, reports and posts.
- Explaining a topic at whatever level you need, as many times as you need, without judging you.
- Summarising long documents and pulling out the parts that matter for you.
- Rewriting in a different tone, length or language.
- Organising messy notes into lists, plans, step-by-step guides and checklists.
- Brainstorming options you hadn't considered.
- Reviewing your own writing, plan or code and pointing out gaps.
It slips in a few predictable ways:
- Confident mistakes. Sometimes Claude states something that is not true, with the same calm tone it uses for true things. A made-up book title, a wrong section number, a statistic that doesn't exist. This is called a . It is rarer than it used to be, but it still happens, especially with precise details such as names, numbers, dates, quotes and citations.
- Old knowledge. Training stops at a date called the . Anything after that, such as today's prices, a rule that changed last month or yesterday's match score, Claude may not know. With web search on, it can look things up and show its sources.
- Missing context. Claude doesn't know your business, your boss, your city or your deadline unless you say so. A vague request gets a generic answer.
- Long calculations. It handles reasoning about numbers well, but a long chain of arithmetic can contain a slip. For money and marks, check the totals with a calculator or ask Claude to show each step.
The rule of thumb: let Claude do the drafting and thinking, and keep the checking of facts, figures and anything with consequences for yourself.
For experienced engineers
Why do confident mistakes happen at all? Claude writes the most fitting next piece of text, given everything before it. Usually the most fitting continuation is also the true one, because true things appear often and consistently in what it learned. But when a question asks for a precise detail it never saw, or saw rarely, the most fitting-sounding continuation can be a plausible invention. Asking Claude to say when it's unsure, to quote from a document you've attached, or to search the web all help, because they give it something real to lean on instead of guessing.
The desktop app's extras
If you use Claude a lot on a computer, the desktop app adds some conveniences the website can't offer:
- Quick entry on Mac: a keyboard shortcut (by default, a double tap of the Option key) opens a small Claude box over whatever you're doing, so you can ask a question without switching windows.
- Screenshots and window sharing, so Claude can see what's on your screen when you ask about it.
- Dictation, so you can speak instead of type.
- , which let Claude use things on your own computer, such as a folder of files, once you install and allow them.
- Cowork and Claude Code, for longer tasks where Claude works through many steps for you, on paid plans. Both get their own lessons later, such as Claude Cowork and Claude Code.
None of these are needed to start. The website is enough for everything in this module.
Privacy basics
Treat a chat like a conversation with a helpful outside person: useful, but not a vault.
- Never paste passwords, OTPs, PINs, card numbers or private keys. No task needs them.
- Be careful with other people's , such as customer phone numbers, patient details, Aadhaar numbers or salary sheets. If it isn't yours to share, don't share it. You can often replace real names with "Customer A" and still get the same help.
- Follow your employer's rules. Many companies use Claude through a work plan with its own settings and policies. Use the work account for work, and don't paste office documents into a personal account.
- Know your settings. On personal plans, Settings, then Privacy, has a choice about whether your chats may be used to improve Anthropic's models. There is also an , which stays out of your chat history and memory and isn't used for training.
Policies and settings change, so look at the current privacy page in the app when it matters to you.
The big idea: Claude is a skilled, fast, well-read helper that knows only what you tell it and what it learned up to a date. Give it context, use it for drafts and thinking, and check what matters.
Step by stepA teacher's first real chat
Kavya teaches Class 7 maths at a school in Jaipur. She has 32 students, a 40-minute period on Thursday and a chapter on fractions. She has never used Claude before. Here is how her first chat went, step by step.
- 1
YouClaude
Who you are, what you need, for whom
- 2
ClaudeYou
A first draft, plus one question back
- 3
YouClaude
Answer the question, ask for a change
- 4
ClaudeYou
A better second draft
- 5
YouClaude
Which parts should I double-check?
- 6
ClaudeYou
The facts and numbers worth verifying
Step 1: She says who she is and what she needs. Instead of typing "fractions worksheet", she writes:
I teach Class 7 maths at a CBSE school in Jaipur. My students are
mostly comfortable with adding fractions but get stuck on word problems.
Please make a worksheet of 10 word problems on adding and subtracting
fractions, set in an Indian sabzi market. Students have 25 minutes.
- Problems 1 to 4 easy, 5 to 8 medium, 9 and 10 harder
- Simple English, short sentences, Indian names
- Put the answer key at the end, with one line of working each
Before you start, ask me anything you need to know.
That message took her two minutes to write. It names the audience, the topic, the setting, the time, the difficulty spread, the language level and the format of the answer key.
Step 2: Claude asks one question back. Because she invited questions, Claude asks whether her students have learned mixed numbers (like 2 and a half) yet, since harder problems often use them. Kavya replies: "Yes, but only with denominators up to 12."
Step 3: The first draft arrives. Ten problems, about a vendor named Ramesh selling tomatoes and a girl named Anjali buying spinach, with an answer key. Kavya reads it through. Problems 1 to 8 look right. Problem 9 is long, three sentences more than the others.
Step 4: She asks for a change. "Problem 9 is too wordy for my weaker readers. Rewrite it in two sentences. Keep the maths the same."
Claude rewrites it. She doesn't start a new chat; she keeps going in the same one, so Claude still has the whole worksheet in view.
Step 5: She checks what matters. Kavya solves all ten answers herself, which takes eight minutes. Nine match. In problem 7, the answer key says 7/12 kg, but the correct total is 11/12 kg; the working line added the numerators wrongly.
Step 6: She points out the mistake. "Problem 7's answer looks wrong to me. Can you recheck it step by step?" Claude works through it, agrees the answer should be 11/12 kg, and corrects the key.
Step 7: One last useful question. "Which parts of this should I double-check before printing?" Claude suggests confirming the answer key, checking that the vegetable prices it invented look realistic for her city, and reading the problems aloud once for clarity.
Total time: about 20 minutes, including her own checking, for a worksheet that would usually take her an hour. And she caught the one mistake before 32 students did.
Notice what made it work: a first message with real context, an invitation to ask questions, small follow-ups in the same chat, and her own check of every answer.
ScenariosYou're the engineer
Real situations you could face. Think about what you'd do before reading on.
You need a fact from this week
You want to know the last date to file something this month, or today's petrol price in Bengaluru. Claude's training knowledge may be months old. Turn on web search so it can look it up and show the pages it used, then open the official source and confirm. If search is off, treat any date or price it gives as a starting point, not an answer.
Your manager asks you to summarise a client contract
It's tempting to paste the whole contract into your personal Claude account. First check whether your company has a work plan and a policy on AI tools. If it does, use the work account. If it doesn't, ask your manager before sharing a client's confidential document anywhere. You can still ask Claude general questions, such as what a "limitation of liability" clause usually means, without pasting anything private.
A family member asks about their medicine
Your father wants to understand the blood pressure tablets the doctor prescribed. Claude can explain in plain Hindi what that type of medicine generally does and what questions are worth asking at the next visit. It shouldn't replace the doctor's advice on dose or changes. Use it to understand and to prepare questions, then let the doctor decide.
You don't know which app to install
Start with the website. Everything in this module works there. Install the phone app when you find yourself wanting quick questions on the go, and the desktop app when you use Claude every day on your laptop and want quick entry, screenshots or the longer-task features.
What happens when…Situations
Real situations this idea runs into, and how well-built systems handle each one.
The case that didn't exist
Arjun, a junior at a small law office in Lucknow, asked Claude for court judgments supporting a client's argument. The reply listed three cases with names, years and one-line summaries. He copied them into a draft. The senior partner searched for the second case and couldn't find it anywhere. It had never existed.
Treat every precise reference Claude gives, such as a case, a section number, a statistic or a quote, as unverified until you've found it in a trusted source. Ask Claude to mark anything it isn't sure about. Better still, give it the real documents (attach the judgments you already have) and ask it to work only from those, quoting the lines it relies on.
The password in the chat
Sneha was trying to fix a problem logging into her company's billing portal. To explain the problem fully, she pasted the error message along with her username and password, and the OTP she'd received.
Change the password right away, and tell your IT team if it was a work account. Then describe problems without secrets: "I enter my username and password and get error code 403" gives Claude everything it needs. No helper, human or AI, needs your password or OTP to explain an error.
Last year's rule in this year's notice
Imran used Claude to draft a notice to his shop's customers about a tax change. The draft quoted a rule that had been updated a few months earlier. Two regular customers who were accountants pointed it out on the WhatsApp group.
For anything that depends on current rules, rates or dates, either turn on web search and check the sources Claude shows, or confirm against the official government page yourself. Ask Claude to flag which sentences in the draft depend on facts that might have changed, so you know exactly what to check.
Watch outCommon mistakes
Typing keywords like a search box
People used to search engines type "cheap hotel Goa December" and get a generic list of tips back. Claude isn't a list of web pages; it's a conversation partner that does better with full sentences and reasons.
Write the way you'd brief a person: "We're a family of four, two kids under 10, going to Goa from 20 to 24 December. Budget about 6,000 rupees a night. We want a quiet beach, not a party area. What should we look for, and what questions should I ask hotels?"
Taking the first answer as final
The first reply is a draft. Many people read it, feel it's not quite right, and give up on the tool.
Reply to it. "Shorter." "More formal." "Use an example from a kirana store." "You missed the part about refunds." Two or three small follow-ups usually turn an average answer into a good one.
Asking for facts you can't check, then trusting them
A student asks for "five statistics about India's smartphone users" for a project and pastes them in as they are. Some may be outdated, some invented.
Ask Claude to explain the topic and suggest where reliable numbers are published, then get the numbers from those sources yourself. Or turn on web search and check each figure against the page it links to.
One endless chat for everything
Some people keep a single chat open for weeks: a resume, then a recipe, then a tax question, then a birthday poem. The early topics crowd the conversation and confuse later answers.
Start a new chat for each new task. Keep a chat going only while you're working on the same thing.
Pasting other people's data without thinking
Pasting a sheet of 400 customer names and phone numbers to "write a personalised message for each" shares data that isn't yours to share.
Ask for a template with placeholders, such as "Dear NAME", and fill in the names yourself or with a mail merge. You get the same result without the data leaving your hands.
Trade-offsWhat it costs, and when not to use it
What Claude gives you: fast first drafts, patient explanations at any level, a second pair of eyes on your work, and help organising messy thinking. It works in many languages, at any hour, and on almost any topic.
What it costs you:
- Checking time. You still need to verify facts, figures and anything with consequences. Budget for it.
- Thinking about privacy. Every time you paste something, you're deciding to share it.
- Limits. Every plan has a , and long chats or big files use it up faster.
- The risk of not learning. If a student lets Claude write every answer, they skip the practice that builds skill. Use it to explain and quiz you, not only to produce.
When not to rely on it:
- Final decisions on health, law or money, where a qualified person should decide.
- Live facts such as prices, scores or today's rules, unless web search is on and you check the sources.
- Anything confidential that your employer or the data's owner hasn't allowed you to share.
Interview angleIn an interview
More and more interviews, even for non-technical roles, include a question like "Do you use AI tools?" A good answer shows you use them sensibly, not that you use them for everything.
You might say:
"I use Claude most days for first drafts and for understanding things quickly. For example, I'll paste a long report and ask for the five points that matter for our team, with the page numbers. I always give it context first, who it's for and what format I need, and I check any facts, numbers or dates against the source before I send anything. I never paste passwords or customer data into a personal account."
Likely follow-ups:
- "How do you know its answers are right?" For writing tasks, I read it as I'd read a colleague's draft. For facts, I check against the source document or an official site, or I turn on web search and open the links it cites.
- "Is it safe to use with company data?" It depends on the company's plan and policy. I'd use the work account if there is one and follow the rules on what can be shared. For anything sensitive, I'd remove names or ask in general terms.
- "Won't it make mistakes?" Yes, sometimes, especially with precise details. That's why I treat its output as a draft I'm responsible for, not a final answer.
When explaining Claude to a manager or a client, keep it simple: "It's a very capable assistant for writing, explaining and organising. It's fast and usually right, but it can be confidently wrong on details, so a person checks what goes out."
Quick checkCheck your understanding
Words in this lessonThe jargon, in plain English
Every technical word used above. In the lesson, tap any word with an ⓘ to see its meaning right there.
- AI assistant
- An app you talk to in everyday language that uses an AI model to answer questions, write, explain and help with tasks.
- API
- An Application Programming Interface: an agreed set of requests one program can send another, and the responses it will get.
- Browser extension
- A small add-on installed in a web browser such as Chrome that adds new abilities, like a helper that can read and act on the pages you open.
- Context window
- Everything an AI model reads on one call: the app's instructions, the chat so far, any documents and the question. It has a maximum size, counted in tokens.
- Desktop extension
- An add-on for the Claude desktop app that lets Claude use something on your own computer, such as a folder of files, after you install and allow it.
- Hallucination
- When an AI model states something confidently that isn't true, because it writes what sounds likely rather than looking anything up.
- Incognito chat
- A Claude conversation that is kept out of your chat history and memory and is not used to train models.
- Knowledge cutoff
- The point where a model's training text ends. The model knows nothing that happened later unless the app supplies it, for example through search or RAG.
- LLM
- Large language model: an AI model trained on huge amounts of text that writes answers one word piece at a time.
- Office add-in
- A plug-in that adds a helper panel inside Microsoft Excel, Word, PowerPoint or Outlook, so you can work with Claude without leaving the file.
- Personal data (PII)
- Any information that identifies a person or could be linked to them, such as a name, phone number, address or Aadhaar number.
- Usage limit
- How much of a service one person or plan may use in a set time, such as 40 messages or a number of AI tokens a day. It resets at a known time.
Enjoyed this? Keep going for free
This lesson is a free preview. Sign up free to save your progress and earn XP, and to unlock all 288 lessons, the simulator challenges and mock interviews.
Sign up free