Meet both workspaces

What you'll walk away with: working accounts in both ChatGPT and Claude, the same small real thing built in each, a read on both products' menus — and the first line of your own answer to "which one," written by you.

You need: an email address, your phone (one of the sign-ups sends a text-message code), about 55 minutes — the two sign-ups and the double build are most of it — and one small real task to carry through both tools: a checklist you actually want, like onboarding the next hire. Free tiers finish this lab; the caveats are below, before you rely on them. Nothing to install, no card.

What these two are

First, thirty seconds of retrieval, because saying it back is how it sticks: from What AI actually is, what happens in the moment you press send? Say your version before reading on — then check it: everything in front of the model is handed over at once, and a frozen guessing machine writes a likely continuation. And the board: OpenAI makes ChatGPT, Anthropic makes Claude, and inside each app sits a family of models. Today you open both front doors and learn to read either one's menus on sight.

Why both, when you're not yet sure you need one? Three practical reasons. The two products keep arriving at the same ideas and naming them differently, so a person fluent in both vocabularies can read any tutorial, any colleague's setup, any announcement. A second window is a second opinion — the previous lesson's most useful move, and it works best across two products with different instincts. And both companies rename, repackage, and reshuffle constantly; someone who knows what the thing is survives every rename of the button.

If your company lives in Microsoft: Copilot is Microsoft's interface inside Word and Excel. Under its hood run the same kind of engines you are about to meet — which ones, and when, moves too fast to print here — so you are not off to one side learning the wrong thing; the ideas below are the ideas behind your company's tool. Gemini is Google's; excellent, not covered here, and every idea below has a Gemini twin.

And to say it plainly: this site will not tell you which of the two to prefer. Not today, not later — that answer is yours to earn, and by the end of this lab you will have started earning it with your own hands.

What you'll see when you log in

Both products share a floor plan — six fixtures that anchor every menu you will ever open in either one. And they are less six features than six answers to one question from the last lesson: what ends up in the window when you press send? Every fixture below is a way of loading — or sparing — the same finite window.

• Chats — each one a fresh, empty window; the list down the left is history, not memory. Nothing from an old chat rides along unless something below puts it there.
• The model picker — a dropdown naming the family: which engine reads your window — deeper and slower, or faster and cheaper. The decoder keeps the current names, dated.
• Projects — Projects (both): a scoped home for related work. Chats started inside it begin already briefed from its files and standing instructions — loaded whole while the pile is small, searched for the relevant parts once it isn't.
• The document panel — writing & code blocks (ChatGPT) · Artifacts (Claude): a live document the AI writes — beside the chat in one product, as editable blocks inside it in the other — which you edit in place, so revisions stop piling copies into the conversation.
• Memory — Memory (both): notes the app keeps about you between chats and loads into the window on your behalf; you can read them and delete what you don't want carried.
• Long-run modes — Deep Research (ChatGPT) · Research (Claude): the AI spends minutes gathering sources — fetched into the window — and returns a cited brief instead of a fast answer.

THE PLUMBING — WHAT FEEDS THE WINDOWSCHEMATIC — THE SHAPE, NOT THE SIZES

• project filesPROJECT
• standing instructionsPROJECT
• memory fileAPP
• this chat so farCHAT
• your newest messageYOU

CONTEXT WINDOWfinite — everything competes for it

MODEL

the app decides how much of each pipe fits — big piles arrive selected or condensed, not whole.

You will touch the first five today; the sixth — the long research run — you will meet by name in Step 4, and a later lab hands it to you properly.

What today costs

Nothing, honestly arrived at. Both free tiers cover this lab, with a few asterisks, stated now rather than after you've relied on them: Projects has drifted in and out of the paid tiers on both sides — if yours turns out locked, run the lab in ordinary chats and retype your two sentences of context; everything else works. And Claude's free tier meters messages in rolling windows of a few hours — if it pauses you mid-lab, that is the meter, not a malfunction; finish that half later. ChatGPT's free tier meters differently: everyday chat currently runs on the family's big engine, with limits that shift — past the cap it quietly continues on a smaller one — and the extras (uploads, images, the long research mode) carry their own caps.

Both sign-ups will offer you a paid plan. Decline everything today. The paid tiers mostly buy longer meters and earlier, longer access to the newest engines, and that is a decision to make with a week of experience, not on a landing page. One mechanical note: Claude's sign-up currently texts a code to a real mobile number, so have your phone out.

Step 1 — Open both front doors

Go to chatgpt.com and claude.ai and make an account on each — email or single-sign-on, decline every upgrade screen, and expect the text-message code on the Claude side. Put the two tabs side by side.

First, a ten-second look in either tab: click the model dropdown at the top of the chat. That is the family from the game board, under this season's names — the control that decides which engine reads your window. Leave it on the default and close it — knowing the control exists is today's whole job with it.

Then pick either tab to start in. Not a commitment — whichever is already in front of you. You'll do the full build there first, and repeat it in the second tab in a few minutes, where it will be the same three clicks under different names. That repeat is the lesson.

Step 2 — Build in your first tab

Give the work a home. You met Projects (both) on the floor plan: a scoped workspace with its own files and standing instructions. In the terms of What AI actually is, a project is a pre-loaded window: everything you put in it becomes the briefing your chats start with — loaded whole above your message while the pile is small, searched for the relevant parts once it outgrows the window. Each chat begins briefed instead of empty; you decide what is on the shelf, the app decides what rides. The lever you hold, made standing. The idea has its own shelf entry: projects and folders.

Make one in your first tab, named for the real task you brought.

One honest note: menus move. Both companies reshuffle their sidebars often, so if the word is not where this says, look for the nearest thing that holds files and standing instructions.

Now build the thing you actually came with. Inside your new project, ask for it:

Write me a one-page checklist for [your task]. Ask me up to five questions first — I would rather answer questions than get something generic. Put the result in an editable document I can revise in place — not just a chat reply.

What opens is the feature this whole site keeps pointing at: writing & code blocks (ChatGPT) · Artifacts (Claude). One idea, two names — a live document you edit directly instead of asking for "the same thing again, but with point four fixed." One product currently opens it as a panel beside the chat, the other as editable blocks inside the thread; where it sits moves with redesigns. Edit-in-place is the fixture.

Use it like a document, for a minute: change a line by hand, then ask for one section to be rewritten and watch it update in place rather than the chat growing a fourth copy.

The lesson of this lab arrives early here: the job of this feature is identical in both products — what differs is the word on it, where it sits this season, and the trims around it: what exports, what shares, what it can run. Learn the job once; check the trims per product. Tutorials written a year ago call this fixture Canvas in one product — a name already retired — and the other product's name for it still stands, for now; the decoder keeps the current pair, dated. The names move. The thing itself has not.

Step 3 — Mirror it in the second

Now the second tab. Same project setup, same checklist prompt, same document-panel request — go. It should take a third of the time, because you are not learning the ideas again, only the words: Projects is still Projects, and the editable doc is writing & code blocks (ChatGPT) · Artifacts (Claude) under the other tab's name.

When both builds exist, put the two checklists side by side and actually read them. Which would you hand the new hire tomorrow? Then score more than taste, because the taste question hides the useful ones: which tool asked you the better questions before writing? Which showed you what it was using — your files, its search, anything you could check? Which was easier to revise in place? Did anything fail, or sit locked behind your tier? Note which tool made the better artifact and one sentence on why — plus anything the other did that you'd miss. Keep the note — it is the first line of your own answer to "which one," and it will change as you use both, which is exactly why nobody here will write it for you.

(And if Monday's standup asks what this was: the ideas you just exercised are the ones under your company's Copilot, and this lab — one checklist through both tools — is a half-hour exercise anyone on your team can run.)

Step 4 — Read the rest of the menu

This lab ends with seven name-pairings written onto a card (that is Step 5). You have built with two of them already; the third — memory — you will open in a minute; the decoder keeps the complete dated table, these and more. The other four take one honest sentence each, and come with explicit permission: nothing below has homework today. Later labs put each one in your hands.

• GPTs (ChatGPT) · Skills (Claude) — saved behavior you switch on: a bundle of instructions, sometimes with files attached, that changes how the AI works on demand.
• Codex (ChatGPT) · Claude Code (Claude) — an AI that works inside a software team's files (a "repository" is the shared folder where a team's code lives), returning finished changes instead of advice. Built for developers; two later labs borrow it.
• Work (ChatGPT) · Cowork (Claude) — an AI that drives real applications and files on your behalf, taking many steps on its own before reporting back.
• Deep Research (ChatGPT) · Research (Claude) — the long research run from the floor plan: slower on purpose, and it returns a cited brief instead of a fast answer.

Now the third pairing, hands-on, one minute: open Memory — Memory (both). Yours is empty or nearly so; that is what a new account looks like, and the point today is knowing the file exists, where it lives, and that you can read it and delete lines. While you are in settings, find the other switch the last lesson promised you: the one governing whether your conversations are kept to help train future models. The wording moves — look for "improve the model" or "data controls" — and set it the way your work requires.

Step 5 — Write your translation card

The thing you keep from this lab is not an account. It is a card.

Pick the tool you expect to use most. Your seven rows: projects, the document panel, memory, saved behaviors, the coding agent, the computer agent, the long research run. Open a chat inside your project there and ask for the card in writing & code blocks (ChatGPT) · Artifacts (Claude) — one page, seven rows: the pairing on the left, what the thing actually does in the middle, and what it means in your work on the right — that last column written by you rather than by the model.

Make me a one-page table in an editable document beside the chat. Left column: the seven feature pairings I list below. Middle column: what the thing actually does, in plain language, using neither company's name for it. Right column: leave every cell blank for me to fill in. Here are the rows: [paste your seven].

Fill the right-hand column by hand for the three rows you touched today — "where the Q4 pricing files live," "the one I keep ignoring." The other four stay blank on purpose: each later lab that hands you one of those tools ends with you filling its row. A finished card, in your own words, is what completing this site actually leaves you holding.

Keep the card in the project, where every future chat can read it — or, if Projects is locked on your tier, anywhere you keep working documents.

Minimum viable context

The craft you started practicing today has a name, and it is the thing this whole site trains: deciding what goes into the window and what stays out.

Decide the goal first — what you want out is what tells you what must go in. Then split the load the way the last lesson's ledger splits it. What training already baked in, you never paste: it knows how a checklist is built, what onboarding is, how a polite email reads. What only you can supply, you always paste: the actual role, the actual constraints, the numbers, the way your firm actually runs — it cannot know any of that through any other door. And keep the window lean, because it is a budget: everything you add competes with everything else for it (you watched the odds move with the pile in the last lesson's x-ray).

Enough that it isn't guessing; little enough that nothing drowns. When an answer disappoints, before you blame the machine or the vendor, audit the window you handed it: what was missing that only you could have supplied? What was riding along that didn't need to be there? And did the load-bearing thing actually make the window — or did it sit in a big attachment the app read for you? That audit — not a different product — is usually the fix.

---

Check your understanding: name three things one product calls by a different word than the other — then say what each one actually does without using either company's name for it. Then the harder pair: which company makes which product, and what is the difference between an app and a model?

Check yourself first — then open one good answer+−

Three of the pairings, plainly: the editable work product beside the chat — writing & code blocks (ChatGPT) · Artifacts (Claude) — a live document you revise in place instead of re-prompting. The scoped workspace — Projects (both) — files and standing instructions your chats start briefed with. The saved behavior you switch on — GPTs (ChatGPT) · Skills (Claude) — a bundle of instructions that changes how the AI works on demand. (Any three fixtures count; the card you just wrote is the answer key.)

The harder pair: OpenAI makes ChatGPT; Anthropic makes Claude. And the app is the ordinary software around the text box — history, settings, folders — while the model is the machine that writes. When something remembers you across visits, that is the app; when something drafts, that is the model.

Your answer should have hit: three jobs named without vendor words, the maker–product pairs both ways, and app versus model as furniture versus engine.

Use this in every lab that follows. The fixtures you touched today — the project, the document panel, the window they all feed — are the furniture of every other lab here; when a menu looks unfamiliar, read it against your card, not against the buttons. Next up — Where it breaks: the failure shapes, and the window move that answers each one.

Remember that product terms change. Every vendor name on this page is accurate as of August 2026, and at least one of them will be wrong before this sentence feels old; both companies rename and repackage constantly. What does not move is underneath: a scoped workspace, an editable document beside the chat, saved behavior you switch on, an agent in your codebase, an agent at your computer, a long cited research run, and memory across chats. When a name changes, the decoder is what gets corrected — so trust the middle column of your card, not the label on the button. For any word either product hands you that this card doesn't cover, the AI glossary defines 296 of them in plain English.