No. A search engine gives you places to look. Claude gives you a result.

You describe what you need – Claude thinks, looks things up if it has to, calculates, writes, and puts the finished thing in front of you.

Nothing at all. Open a browser, sign in, start typing. There are apps for your computer and your phone too – but you need neither.

And you don’t have to learn how to ask properly. Put it the way you would say it to a colleague.

Yes – just drag them into the conversation. PDFs, photos, spreadsheets, text files.

Claude reads them and works on from there: summarising, comparing, pulling out numbers, answering questions about them.

That is an artifact. When the result is no longer a sentence but a thing, Claude puts it beside the conversation instead of into it.

A document, an overview, a diagram – or a small application that actually runs.

Say what should be different – Claude rebuilds it. You can also touch the text directly: highlight it and have it changed.

Every version is kept, so you can page back. You can download it, and publish it too: then you get a link to pass on.

Yes. Claude can build real files – spreadsheets with working formulas, presentations, documents, PDFs.

It is built in a sealed-off area at Anthropic, not on your computer. You download the finished file.

Better not. A conversation has a working memory, and it fills up – then things get sluggish and vague.

Rule of thumb: new topic, new conversation. It costs nothing and makes the answers better.

No. Claude remembers things across conversations and can look back at earlier ones. You can also just tell it “remember that”.

What it remembered you can look at, change and delete – or switch memory off entirely. For one-off things there is incognito: it shows up nowhere.

That is what projects are for. Each one is its own workspace: its own conversations, its own documents, its own instructions – and its own memory.

Set it up once and it applies to everything inside. You don’t have to explain the background again in every conversation.

Not by itself – its knowledge has a cut-off date. But it can look things up, and then it names its sources.

Two different buttons: search briefly fetches something from outside. Extended thinking fetches nothing at all but thinks harder – good for puzzles.

Then use research. Claude keeps searching on its own, follows leads and checks several sources against each other – for one to three minutes.

What comes out is a report with evidence, not a quick answer. For comparisons, market overviews, decision papers.

If you connect it, yes. You attach Drive, calendar, mail and a good deal more under Customize.

After that Claude fetches things itself. But only connect what you really need – and what you know.

Then change the style. There are ready-made ones – concise, formal, explanatory – and you can build your own.

For your own you simply give Claude a few lines that sound like you. From then on it writes that way.

The same conversation, everywhere: browser, desktop app, phone. You open it wherever you happen to be.

On the phone you can simply talk instead of typing. Handy while walking – and for sorting out your thoughts.

Usually not. The default fits almost everything.

If a job is genuinely hard, take a stronger model. If it just has to be quick, take the small one. As of August 2026 they are called:

State the goal, not the steps – Claude finds the way itself. And say how you can tell when it is done.

Give the context: who it is for, what for, and what must not happen. The clearer the task, the fewer rounds.

Usually yes – but Claude can also be convincingly wrong, and it sounds just as certain as ever while doing it.

So: for anything that matters, ask where it came from. And have numbers worked out again rather than believing them.

Good question – that is where the trap is. Claude reads web pages, e-mails and other people’s documents. None of that is an instruction from you.

Instructions come from you. Everything else is material to read. So with anything from outside, look more closely at what comes out.

That is in the settings, and you decide it. There you set whether your conversations may be used to improve Claude.

Incognito conversations are never used for it – and never appear in your history either. For delicate matters the rule still holds: only put in what you would also hand to a service provider.

Then it goes on in stages – the question is always how far you let Claude at your things.

Here, in the conversation, everything passes through you: you put in, you take out. That covers an astonishing amount.

Exactly. Start with a real job, not with a test.