How to use Claude AI: projects, files, artifacts and research

To use Claude AI well, give it real context (files, a Project, clear instructions), pick the right model for the job, and ask for a concrete output you can check. This guide walks through Claude.ai feature by feature: plans, models, prompting, files, Projects, artifacts, Research, memory and privacy settings.

How to use Claude AI: the short version

Using Claude AI well comes down to three habits: give it the material it needs, tell it exactly what output you want, and check the result before you rely on it. Claude is Anthropic's AI assistant, available at claude.ai, in desktop apps for macOS and Windows, and in mobile apps for iOS and Android.

Here is the quickest path from sign-up to useful work:

1. Create an account at claude.ai with an email or Google account.
2. Open Settings and fill in Instructions for Claude with a few lines about who you are and how you like answers (more on that below).
3. Start a chat, attach the document or data you are working on, and describe the task, the audience and the format you want.
4. Read the answer critically, then ask for specific changes ("shorter", "add a table comparing the two options", "cite the page for each claim").
5. When a task repeats every week, move it into a Project so the background is always there.

The rest of this guide covers each feature in the order most people need it.

Which Claude plan do you need?

Most people should start on the Free plan and upgrade only when they hit usage limits or need Research or Claude Code. At the time of writing (October 2026), the Claude pricing page lists these plans:

Usage limits reset on a rolling five-hour window, and paid plans also have weekly limits. Long conversations, big files and Research all use more of your allowance than short chats, so starting a fresh chat for a new topic makes limits last longer.

Which Claude model should you pick?

Pick Sonnet 5.5 for everyday work and change only when you have a reason. You switch models from the model menu in the message box, and the choice applies to that chat.

A practical rule: if Sonnet gives you a shallow or wrong answer on a hard problem, retry the same prompt with Opus before rewriting your prompt. If you are burning through limits on simple tasks, drop to Haiku. For how Claude stacks up against other assistants by task, see ChatGPT vs Claude vs Gemini.

How do you write prompts that get good answers from Claude?

Claude gives its best answers when the prompt says what the task is, who it is for, what material to use and what the finished output looks like. Vague prompts get generic answers because the model has to guess the missing parts.

Compare these two requests:

Write an email about the project delay.

Write an email to our client, Dana (head of marketing at a retail chain),explaining that the new loyalty app launch moves from Nov 3 to Nov 17.Reason: the payment provider's review took two weeks longer than planned.Tone: direct and calm, no grovelling. Offer a 20-minute call this week.Keep it under 150 words and end with one clear next step.

The second prompt works because every guess the model would have made is now a fact. A few habits make the same difference across all kinds of work:

• Paste or attach the source material. "Summarize the attached contract's termination clauses" beats "What do contracts usually say about termination".
• Show an example of the output you like. One good sample does more than a paragraph of adjectives.
• Ask for structure when you need it: a table, numbered steps, a JSON object, a one-paragraph summary followed by details.
• Iterate in the same chat. Follow-ups such as "make section 2 more concrete" keep the context Claude already has.
• Ask Claude to check itself. "List any claims above that you are not sure about" surfaces weak spots fast.

Our prompt engineering guide covers these techniques in more depth, with templates.

How do you use files and connectors in Claude?

You can attach files with the paperclip or "+" button, or drag them into the chat, and Claude reads them as part of your message. PDFs, Word documents, spreadsheets, CSVs, images, code and plain text all work, and Claude can analyze charts and scanned pages in images and PDFs.

Good file tasks look like this:

• "Compare these two vendor proposals and give me a table of price, scope, timeline and risks."
• "This CSV is last quarter's orders. Which five products grew fastest, and how did you calculate it?"
• "Here is a screenshot of an error dialog. What does it mean and what should I try first?"

With code execution and file creation enabled under Settings, Capabilities, Claude can run code to analyze data and produce real files you can download, such as spreadsheets, Word documents and slide decks.

Connectors go one step further: they link Claude to services such as Google Workspace so it can search your documents, email or calendar when you ask. Connectors are available on every plan. Only connect accounts you are comfortable letting Claude read, and remember that anything Claude retrieves becomes part of the conversation.

How do Claude Projects work?

A Project is a workspace with its own instructions, knowledge files, chat history and memory, and every chat you start inside it gets that context automatically. Projects are the single most useful feature for work you repeat.

Typical Projects:

• "Client: Acme" with the contract, brand guidelines and past reports as knowledge, and instructions like "Write in Acme's voice; they use British spelling."
• "Weekly newsletter" with three of your best past issues and a short style guide.
• "Thesis" with your outline, sources and the citation style your department requires.

To set one up, click Projects in the sidebar, create a new Project, add files to its knowledge base and write project instructions. Free accounts can have up to five Projects. On paid plans, when a Project's knowledge gets close to the context limit, Claude switches to retrieval, which the help center says expands capacity by up to 10x. On Team and Enterprise you can share a Project with colleagues as "Can view" or "Can edit".

Keep project instructions short and specific: who the audience is, the tone, terms to use or avoid, and the format you want by default. A dedicated walkthrough with examples is in our Claude Projects guide.

What are Claude artifacts and when should you ask for one?

Artifacts are finished outputs, such as documents, slide decks, designs, diagrams, code and small interactive tools, that open in a panel beside the chat so you can edit, reuse and share them. Ask for an artifact whenever the thing you want is a deliverable rather than an answer.

According to the artifacts help article, artifacts need the "code execution and file creation" capability turned on in Settings, Capabilities. You can then ask for one in any chat or start from a template in the Artifacts tab in the sidebar. You can edit by asking Claude or directly in the artifact, and in documents you can highlight text and choose "Edit with Claude".

Export options depend on the type:

Chat artifacts and the Docs, Slides and Design templates are available on all plans. Connecting apps to artifacts and storing data in them needs a paid plan. Artifacts start private; you choose who can open them.

A useful artifact prompt names the format and the parts:

Create a one-page project brief as a doc artifact with these sections:Goal, Scope (in and out), Timeline as a table, Risks with an owner each,and Open questions. Use the meeting notes I attached as the only source.

How do you use Claude's Research mode?

Research mode makes Claude run many web searches that build on each other and then write a sourced report, which suits questions where one search is not enough. It needs a paid plan and works on web, desktop and mobile.

To start, make sure web search is on, click "+" at the bottom left of the message box and select Research. A blue indicator shows it is active. The Research help article notes that Research runs on Opus regardless of the model you picked, can use your connectors (except on mobile, which searches the web only), and uses your limits faster than a normal chat.

Research works best with a question that has a clear scope:

Research the current requirements for registering a small online shopselling handmade goods in Portugal: business registration, VAT thresholds,and consumer return rules for EU buyers. Cite official government sourcesfirst, and flag anything that changed in the last 12 months.

Always open the cited pages for anything you will act on. Research reduces search time; it does not remove the need to check sources.

How do you set up memory and personal instructions in Claude?

Claude personalizes answers through three layers: account-wide Instructions for Claude, per-Project instructions, and memory that Claude builds from your chats. Setting the first one up takes two minutes and improves every chat.

Find Instructions for Claude by clicking your initials in the lower left, then Settings. Write a few lines such as:

I'm a product manager at a 40-person software company.Default to concise answers with the conclusion first.Use US English. Avoid jargon unless I use it first.When you're unsure, say so instead of guessing.

Memory is on by default for Free, Pro and Max. Under Settings, Memory you can turn off "Generate memory from chats", pause or reset memory, and open the Topics list to read, edit or delete what Claude has saved. Each Project keeps its own memory, and you can turn off "Use account memory" on a Project to keep it fully separate. Paid plans also get chat search, so you can ask "what did we decide about the pricing page last week?"

For tone and format beyond your instructions, Claude also supports Skills: reusable behaviors you can turn on for particular tasks, such as matching your writing style.

Which Claude privacy settings should you check?

On consumer plans (Free, Pro and Max), the setting that matters most is Model Improvement in your privacy settings at claude.ai/settings/data-privacy-controls. It decides whether your chats and coding sessions can be used to train future models.

What the Anthropic Privacy Center says, as of October 2026:

• Model Improvement on: new or resumed chats may be kept in de-identified form for up to five years in training pipelines.
• Model Improvement off: your chats are not used for future training.
• Deleted chats: removed from your history right away and from back-end storage within 30 days.
• Incognito chats: not saved to your history or memory, and never used for training even if Model Improvement is on.
• Thumbs up or down feedback: stored with the related conversation for up to five years.

Team, Enterprise and API usage fall under Anthropic's commercial terms, which do not use your data for training by default. If you use Claude for work, check your company's policy before pasting client data; our guide on whether it's safe to use AI at work covers what to look for.

How to start using Claude today

Spend 20 minutes on setup and you will get better answers from day one:

1. Sign up at claude.ai and install the mobile app if you want voice and quick questions on the go.
2. Write four or five lines in Instructions for Claude.
3. Check Model Improvement and Memory settings and set them the way you want.
4. Turn on code execution and file creation so you can get artifacts and downloadable files.
5. Create one Project for your most repeated task and add two or three reference files.
6. Run one real task end to end, then refine the Project instructions based on what you had to correct.

If your work involves code, Pro also includes Claude Code, Anthropic's coding agent; our Claude Code tutorial covers that path. And if you want coding agents to keep working after you close your laptop, Orkio runs them on an always-on cloud box using your own Claude account.

Frequently asked questions