Claude's Built-In Tools: Projects, Artifacts, and When to Use Each
Here’s a question worth sitting with: when you open Claude, do you type into a box and hope for the best — or do you actually turn on the tools that do the heavy lifting?
Most people do the first thing. They treat Claude like a very smart autocomplete. But over the last year the Claude app quietly grew a set of built-in capabilities that change what it can produce — real Excel files, persistent workspaces, live web results, interactive documents you can edit in place. Almost none of it is hidden behind code. It’s toggles and plain-English requests.
There’s one thing to unlearn first, though. Unlike a developer tool, the Claude app has almost no commands to memorize — no slash-this or type-that. You turn a capability on in settings, then ask for what you want in ordinary words. That’s the whole interface. So this isn’t a list of commands to learn; it’s a list of switches to know exist. Here are the four that matter most.
File creation: when you want a file, not a chat reply
This is the big one, and the one most people have never switched on.
Claude can now build and edit Excel spreadsheets, Word documents,
PowerPoint decks, and PDFs — not a description of a spreadsheet, an
actual .xlsx file you download and open. Behind the scenes it’s running
inside a private computer environment where it can
It doesn’t “know” how to make an Excel file the way it knows facts — it
writes a little program that builds the file, runs that program on a
sandboxed computer, and hands you the result. You never see the code; you
just get the file.
to produce the finished document.
How to turn it on: Settings → Features → Experimental → enable “Upgraded file creation and analysis.” (It rolled out to paid plans in late 2025 and later reached free users.)
Use it when: you’d otherwise open a blank template and start typing. “Turn these meeting notes into a formatted one-page PDF.” “Build a budget spreadsheet with formulas that total each category.” “Make a 10-slide deck from this report, one idea per slide.” Upload the raw material, describe the shape you want, and download the result.
The good: it collapses the tedious middle — the formatting, the formulas, the slide layout — that eats your afternoon. You can also save straight to Google Drive.
The catch: to do this, the feature gives Claude internet access inside that private environment, and Anthropic’s own warning is blunt — “this feature gives Claude internet access to create and analyze files, which may put your data at risk. Monitor chats closely.” The concern is Hidden instructions buried inside a file or web page you hand the model — text you can’t see that the model reads and obeys. In a tool with internet access, a booby-trapped document could in theory tell Claude to send your data somewhere. Rare in practice, but the reason for the “watch it closely” warning. . Practical rule: it’s excellent for your own data and documents; be thoughtful before feeding it sensitive files pulled from untrusted sources.
Projects: when you keep re-explaining the same context
A A saved workspace inside Claude that bundles reference files and standing instructions, so every chat you start inside it already knows your setup. Think of it as a room where Claude already remembers the brief — as opposed to a fresh chat, which is a stranger every time. is a container that holds two things: uploaded reference files (your style guide, past reports, product specs — up to 30MB each) and a set of standing instructions that apply to every conversation inside it. Nothing carries between Projects; each one is its own sealed room.
Use it when: you do the same kind of task repeatedly and you’re tired of pasting the same background into every new chat. A “Weekly Report” project with your format and last month’s examples. A “Customer Emails” project that already knows your tone. On paid plans, Projects also build a memory that summarizes past chats every 24 hours, so continuity like “we’re mid-way through the Q3 review” carries forward.
The good: the quality jump over a cold chat is real — Claude writes in your actual voice and stops asking setup questions you’ve answered ten times.
The catch: Projects don’t read every uploaded file top to bottom on every question. They use The tool pulls only the slices of your uploaded files that seem to match your question, rather than reading everything. Fast and cheap — but if your wording doesn’t match how the document is phrased, the relevant passage may never reach Claude at all. , so for “what does clause 4.2 say, exactly?” verify against the source. Memory stores summaries, not verbatim text — it’s good for “remember what I’m working on,” not “remember this number precisely.”
Artifacts: when the output is a thing you’ll keep editing
Ask for something substantial — a full document, a chart, a working calculator, a chunk of formatted content — and Claude opens a side panel and builds it there instead of burying it in the chat. That panel is an Artifact: a standalone, editable piece of work that lives beside the conversation, so you can revise it turn by turn without scrolling back through a wall of text.
How to invoke it: it often triggers on its own for anything longer than ~15 lines. If it doesn’t, just say so — “build that as an artifact,” or “put that in the side panel.” (If you don’t see the feature at all, check Settings → Capabilities.)
Use it when: you’re drafting something you’ll iterate on — a policy doc, a landing page, a table you’ll tweak five times. “Make the tone warmer.” “Add a third column.” The Artifact updates in place instead of regenerating a fresh copy in the chat each time.
The good: it keeps the artifact and the conversation separate, which matches how real editing works — you talk about the change, and the document changes.
The catch: it’s built for things with structure. For a quick answer or a two-line email, forcing an Artifact just adds a panel you don’t need. Reach for it when the output has shape and staying power.
Web search: when the answer depends on this week
By default, Claude answers from what it learned during training, which has a cutoff date — it doesn’t automatically know today’s news. Flip on web search and it queries the live web in real time and cites the URLs inline, so you can click through and check the source yourself.
Use it when: the question is time-sensitive or specific — recent figures, a current price, “what changed in the new version,” a company’s latest announcement. Anything where a plausible-but-stale answer is worse than no answer.
The good: it’s the single best guard against confident-but-outdated replies, and the inline citations let you verify rather than trust.
The catch: the open web includes junk. Citations tell you where a claim came from, not that the source is right — treat a random blog differently from a primary study. And for questions about timeless concepts, search adds noise without adding accuracy.
How to actually reach for these
Because there are no commands to learn, the skill here is smaller than it looks: know the four switches exist, then say what you want in plain English. Need a file? Turn on file creation and describe the document. Doing this task weekly? Put it in a Project. Editing something iteratively? Let it become an Artifact. Depends on current facts? Turn on search. The capability does the work; your job is just to pick the right one — and to stop treating a tool this capable like a chat box.
The takeaway
Claude’s most useful features aren’t buried in a manual — they’re four toggles most people never flip. File creation gives you real documents. Projects stop you re-explaining yourself. Artifacts make iteration sane. Web search keeps answers current. None of them require memorizing a single command. Turn on the one that fits the job in front of you, and describe what you want like you would to a capable colleague.
Sources: Anthropic, “Claude can now create and edit files” (file creation for Excel, Word, PowerPoint, and PDF; the “Upgraded file creation and analysis” setting; and the internet-access security warning quoted above). Details on Projects (30MB-per-file knowledge, standing instructions, retrieval, and the 24-hour memory summary), Artifacts (the side-panel behavior and ~15-line auto-trigger), and web search (live results with inline citations, available across tiers since May 2025) reflect the current Claude app as documented in Anthropic’s product materials and corroborating 2026 feature guides. The framing of these as “four switches, no commands” and the when-to-use guidance are my own.
Related: Commands, Skills, and Plugins: Which One Do You Actually Need? covers the same ideas in Claude’s developer tool, where you do build custom commands — a useful contrast to the app’s toggle-and-ask model. And Build a Reusable AI Assistant for the Task You Do 20 Times a Week goes deep on the Projects idea introduced here.