Claude chat: how to use claude.ai

A practical guide to the Claude chat interface in 2026 - projects, memory, artifacts, research, connectors, voice, file uploads, and how to get better answers out of it.

Claude chat is the web app at claude.ai: you type a message, Claude answers, and everything else - file uploads, Projects, Memory, Artifacts, web search, connectors and voice - hangs off that one box. It works on the free plan, needs no password (Google sign-in or an emailed link), and the same account carries across the desktop app, iOS and Android.

This page is a hands-on walkthrough of the chat interface as it stands in August 2026: what each control does, which plan unlocks it, and what to type to get something useful out of it. It is an independent reference, not Anthropic - account and billing actions all happen on Anthropic's own site.

#How do you start a conversation, and what can you attach?

Open claude.ai, sign in, and type into the message box. No setup step, no template to fill in. Three controls sit around that box and they are the ones worth learning:

• The attachment button - pulls in files, images and screenshots, or content from a connected app.
• The tools menu (labelled “Search and tools”) - toggles web search, extended thinking, Research on paid plans, and any connectors you have added.
• The model picker - visible on paid plans only. Free accounts get whatever model Anthropic assigns and cannot switch. Paid accounts choose between Sonnet 5, Opus 5 and older versions; Fable 5 is Max and premium-seat territory.

Conversations autosave into the left sidebar. The ghost icon in the top right starts an incognito chat - excluded from memory, from chat search and from any training use, on every plan including Free. Sign-in trouble is covered on the Claude login page.

#Uploading files, images and screenshots

Attach a file and Claude reads it in the same turn. PDFs, Word documents, spreadsheets, CSVs, code files, plain text and images all work; the mobile apps additionally let you photograph a page or share a screenshot straight into a chat.

Two limits matter. A large attachment stays in context and costs usage on every subsequent turn, so start a fresh chat when you switch documents. And Claude accepts images but never produces them, nor does it accept video or audio as input - ask for a chart and you get code or an SVG artifact, not a picture.

File creation is a separate capability and it is on by default, including on the free plan: Claude can hand back a real .xlsx, .docx, .pptx or .pdf built by Anthropic's own document Skills. That switch, “code execution and file creation”, also gates Artifacts and Skills - if any of the three are missing, check Settings first.

#What are Projects, and are they worth using?

A Project is a folder with a memory: a set of chats that share uploaded knowledge files and a block of custom instructions. Use one when you keep re-explaining the same background - a codebase, a style guide, a client's contracts, a research corpus.

The plan difference is real and often misreported. Free accounts get Projects, capped at five, without RAG-enhanced project knowledge. Paid accounts get unlimited Projects, and when a project's knowledge base approaches the context limit Claude switches into a retrieval mode that Anthropic describes as expanding capacity by up to roughly 10×. Team and Enterprise add per-project sharing at view or edit level.

One efficiency note from Anthropic's guidance: project knowledge is cached, so re-referencing the same documents only bills the new or uncached parts. A Project is cheaper on usage than pasting the same 40-page document into ten chats.

#Memory, chat search and how Claude remembers you

These are two different features and people conflate them constantly.

Memory lives in Settings, where you can pause it (keeping what exists) or reset it (deleting it permanently). Chat search being paid-only is one of the sharper Free-plan gaps - on Free, an old conversation is only findable by scrolling the sidebar.

#Artifacts: when Claude opens a second panel

An Artifact is a side panel holding something substantial and self-contained - Anthropic's rule of thumb is content over roughly 15 lines that you are likely to edit or reuse. Documents, code, single-page websites, SVGs, diagrams, React components and small AI-powered apps all qualify.

Artifacts are on every plan, Free included, but two things surprise people. They do not appear in your sidebar automatically: open one and click Publish to keep it, after which it can be shared or embedded. And persistent storage plus MCP integration inside artifacts require Pro or above. Artifacts inside Claude Code are gated differently again, which is why you will see conflicting claims online.

#Research, web search, connectors and Skills

Web search is on every plan. Claude searches, cites, and you can click through. Research is the agentic version: Claude runs a chain of searches that build on each other, decides what to investigate next, and returns a cited report. Research is paid plans only, requires web search to be enabled, and eats usage noticeably faster than ordinary chat.

Connectors plug Claude into your own systems over the Model Context Protocol. Google Drive, Gmail and Calendar connectors work for everyone including Free. Directory connectors - Linear, Slack, Notion, Asana, Stripe, GitHub, HubSpot and 100-plus others - are added from claude.ai/connectors. Custom remote MCP servers also work on Free, but Free is limited to one custom connector. How to build your own is on the MCP and Claude agents page; the Claude API page covers the programmatic equivalent.

Skills are folders of instructions and scripts Claude loads only when a task calls for them. Anthropic's built-in document Skills fire automatically when you ask for a spreadsheet or a deck. You can upload your own under Customize → Skills on any plan, Free included; Team and Enterprise owners can push Skills to a whole organization.

Voice mode is a beta on all plans - two-way spoken conversation, hands-free or push-to-talk, on mobile, desktop and web, and optimised for phones. Two caveats: Fable 5 is unavailable in voice mode, and voice cannot drive Cowork or Claude Code. Voice counts against normal usage limits.

#Eight starter prompts that actually work

These are shaped to exploit what Claude is good at - long inputs, structured output, and being told what “good” looks like. Replace the bracketed parts.

More worked examples, by job rather than by feature, are on the Claude examples page.

#How do you get better answers out of Claude chat?

Four habits move the needle more than any template:

1. Give it the raw material. Claude's advantage is a large context window - up to 1M tokens on paid plans. Paste the whole document, not your summary of it.
2. Say what a good answer looks like. Format, length, audience, and what to leave out. “A table, one row per obligation” beats “summarise this”.
3. Start a new chat when the topic changes. A long thread is re-read every turn: it costs more usage and drags in stale context.
4. Ask for the reasoning to be checked, not shown. Thinking is on by default on current models; on older ones you can toggle extended thinking manually. Either way, “now find the weakest step in that argument” is the useful follow-up.

The full technique set - role framing, few-shot examples, XML tags, output schemas and the failure modes to avoid - is on the Claude prompt engineering guide.

#What Claude chat cannot do

Beyond that: chat search, Research, model selection, the 1M-token context window, Claude Code and Cowork are all absent on Free, and Anthropic publishes no absolute message count for any plan - only a rolling five-hour window plus a weekly cap. The complete breakdown of which plan unlocks what is in the feature and plan matrix, and current prices are on the pricing page. If you would rather pay per token than per month, pay-as-you-go pricing is often cheaper for light use.

#Frequently asked questions