Stop Using AI Like a Chatbot
Most professionals using AI are getting some value from it, but not nearly as much as they could be. They open Claude, type a question, get an answer, and move on. Sometimes it's impressive. Sometimes it misses the mark. Rarely does it feel like something you'd trust for serious work. So it stays a nice-to-have rather than a go-to.
If that sounds familiar, here's the thing: it's probably not the tool. It's the setup. Claude isn't just a chatbot you ask questions to. It's a configurable work environment, and like any environment, it works a lot better once you've made it yours. Skip that step and you're essentially using a new laptop without ever adjusting the settings, connecting your accounts, or telling it who you are.
This walk-through takes about 25-30 minutes. It's practical and hands-on, so follow along as you read, and by the end, you'll have Claude set up, not just open.
In our one-day course, Working Smarter with Claude Chat, this is how we set ourselves up for success before we learn how to write effective prompts, organize work (using projects, artifacts, and styles), evaluate output, and more.
The Setup That Changes How You Use Claude
Step 1: Take 60 Seconds to Look at the Interface
Before typing anything, orient yourself.
Image 1: Claude Chat Interface
Left sidebar → where your work lives
Bottom left (profile) → where Claude learns about you
Center panel → where you interact
Right panel (not shown yet) → appears when you create outputs like Word docs or slide decks.
Tip: If something feels confusing later, you’re almost always just looking in the wrong section.
Common mistake: Treating this like a single chat box and ignoring everything around it.
Step 2: Give Claude Context (Memory)
Right now, Claude knows nothing about you. That’s why responses feel generic.
Open Memory Settings Profile (bottom left) → Settings → Capabilities
You’ll see something like:
Image 2: Claude Memory Panel
Generate memory is available on all plans including free. Search and reference past chats is a paid-plan feature (Pro, Max, Team, and Enterprise).
Enterprise plan owners have an additional org-wide toggle under Organization settings → Capabilities to control memory for all users.
Tell Claude What to Remember
In a new chat, type: “Please remember that I work in [your role], and I prefer a short summary first, followed by detail.”
Claude will confirm it stored it.
What Memory Looks Like
Image 3: What Claude Remembers
How to get the most out of memory:
- Tell Claude who you are early. In a new chat, share your role, industry, and how you like to work. Claude will carry that context into future sessions so you're not starting from scratch every time.
- Be explicit about your preferences. Want shorter answers, bullet points, or a specific tone? Just say so. Claude stores those preferences and applies them going forward.
- You're in control. You can delete individual memories you don't want, or clear everything and start fresh, which is reassuring for anyone who wonders what's being stored.
Things to watch out for:
- Memory is a summary, not a transcript. Claude doesn't record your exact words. It distills key facts (things like "prefers concise answers") not the full conversation behind them.
- It updates nightly, not instantly. If you share something important today, Claude won't carry it into a new chat until the next refresh.
- Projects have their own memory. If you use Projects, context lives inside the project, separate from your general memory. The two don't mix.
- Incognito chats are memory-free. Want a private session with nothing stored? Use an Incognito conversation and memory is off for that chat. For Enterprise users Org admins control the master switch.
- If you manage a Claude Enterprise rollout, you can turn on or off memory for the whole organization from the Organization settings. This is not just a user setting.
Tip: Anything you repeat across multiple chats belongs here.
Common mistake: Trying to store everything. Keep it general.
Step 3: Use Incognito Mode (When Privacy Matters)
Top right of chat → click 👻 (ghost icon). Now your window looks like this.
To exit incognito mode, click first X. The second x all the way to the right will close Claude completely.
Image 4: Claude Incognito Mode
What it is. Incognito mode stops data storage for that session. This means that you don't save anything to your chat history, you don't remember anything, and the conversation won't be used to train Claude. You start it by clicking the ghost icon when opening a new chat.
When to use it. It's ideal for one-off sensitive conversations. A contract question, a confidential client situation, or anything you simply don't want stored. Think of it as a clean room: useful in the moment, gone when you're done.
What it doesn't do. It's not a single "make everything private" toggle. Your regular chats and memories are unaffected. Incognito only applies to that specific conversation. Once you close it, it's gone. When you close an incognito chat, it's permanently deleted. There's no way to retrieve it later, so don't use it for anything you'll want to reference again.
Activity: Try it yourself — incognito mode
Open an incognito chat and type: "Help me prepare for a difficult conversation with a team member." Notice how freely you write when you know it won't be saved. When you're done, close the chat and check your Recents and you’ll see it's gone.
Tip: If you hesitate before typing something, that's your cue to switch to incognito.
Common mistake: Assuming a normal chat is private enough. On Enterprise, your admin can see your chat history — incognito is the only mode that keeps a conversation out of your personal record.
Step 4: Test What Claude Remembers
Now verify how this works. Ask: “What did I ask you to remember earlier today?” Claude should retrieve it.
Then ask: “What did I say in my earlier sensitive conversation?” If that was incognito, Claude won’t know.
The Mental Model
Memory → stays across conversations
Chat history → searchable
Incognito → disappears completely
Step 5: Compare Models (this is where it clicks)
Claude isn't one system. It's a family of models, each built for a different kind of work. Try this prompt "In three bullets, name the biggest trade-off a consulting firm faces when deciding whether to train all employees on AI vs. only client-facing consultants." Run it three times, once per model, in a fresh chat each time. Here's what to look for:
Image 5: Claude Models: Sonnet, Haiku, Opus.
What you'll notice:
- Haiku answers fast and keeps it simple.
- Sonnet is balanced and ready to use.
- Opus slows down and goes deeper and you'll feel the difference in how it frames the trade-offs.
Tip: Don't ask "which model is best?" Ask, "which fits this task?" A quick email draft doesn't need Opus. A board-level recommendation might.
Common mistake: Defaulting to one model for everything. Once you run this comparison, you'll stop doing that.
Step 6: Make the Trade-Off Clear
You've run the same prompt three times. Now the real learning begins by comparing what landed in front of you. Look at your three results and ask:
- Which one would I send to a client or manager?
- Was Opus worth the extra seconds, or did Sonnet get there?
- When would Haiku's speed matter more than depth?
There's no universal right answer. A quick internal Slack message and a board-level recommendation are different tasks and they deserve different tools. This is where you stop guessing about models and start having an instinct for them.
Optional: break it on purpose
Try this: Stay in the same chat, switch models using the selector, then type: "Tighten your previous answer." You'll likely notice something feels slightly off, like a shift in tone, structure, or framing. That's because each model doesn't have memory of what the other one wrote. It's working from the conversation history but interpreting it through a different lens.
Takeaway: Models aren't interchangeable mid-conversation. If the model matters for a task, start a new chat with the right one from the beginning. Switching partway through is like handing a half-finished document to a different writer and asking them to continue.
A quick rule of thumb:
| Task | Model to reach for |
| Quick draft, summary, or lookup | Haiku |
| Most everyday work | Sonnet |
| Complex analysis, strategy, nuanced writing | Opus |
When in doubt, start with Sonnet. It handles most work well, and you'll quickly develop a sense for when something needs more horsepower — or less.
Knowledge check
- What does memory do — and what doesn't it do?
- You're drafting talking points about a sensitive reorganization. Normal chat or incognito?
- You need a fast summary of a meeting transcript. Which model?
- You've been using Claude for two weeks, but it still doesn't feel consistent. What's the most likely thing you skipped?
Answers: 1. It stores context about you across conversations — but it doesn't record transcripts, and it updates overnight, not instantly. 2. Incognito — and remember, on Enterprise your admin can still see chat history, but it won't be stored in your personal record or memory. 3. Haiku. 4. Setup — memory, preferences, and understanding which model fits which task.
By now Claude shouldn't feel like a black box. It should feel more predictable, like a tool you've configured rather than one you're just hoping works. Not magical. Just more yours. Want this to be second nature? This is the first section of the full Claude Foundations course. From here, you start using all of this for real work, like analysis, communication, and decision-making. The environment is already set up the way you need it.
View all our Claude training courses, including our introduction to Claude, Working Smarter with Claude Chat. We also teach Claude Cowork, Claude Design, Claude Code, and more. For a comparison of these Claude tools, read our guide, Everybody Loves Claude.
One last thought
Most people try to get better results by tweaking their prompts. That works, but it's the second move, not the first. Set up the environment. Get the memory working. Pick the right model for the job. Once that's in place, the answers start improving on their own and you stop feeling like you're fighting the tool to get what you need.
