I gave Claude an induction day. 9 settings later, it stopped acting like a chatbot.

You know that feeling when you sign up for a tool everyone's raving about, and it just... answers your question. Fine. Adequate. Nothing special.

That was me and Claude, for months. I'd open a chat, type something, get a fine reply, and close the tab wondering what the fuss was about.

Then I realised what I was doing wrong. I was treating Claude like a search box. It's not a search box. It's a new hire on day one, smart, capable, and completely useless to you until you actually onboard it.

You wouldn't hand a new employee a laptop, no intro, no access, no context, and expect brilliant work by lunch. But that's exactly how most people use Claude. No induction. No access to their stuff. No idea how they like things done. Then they call it "just a chatbot" and go back to Google.

I want to walk you through the induction I gave mine. Nine settings, most owners have never touched. Quickest wins first, highest leverage last, the way I'd walk a new hire through their first day.

Before the settings, let me name the mistake, because I made it too.

You open Claude. You ask it something. You get an answer that's competent but generic, the kind of answer anyone could get, because it is the answer anyone gets. Then you decide Claude just isn't that powerful, and you go back to typing questions into it the same way, forever.

Here's the thing nobody tells you. Claude out the box has no idea who you are, no access to your inbox or your files, no memory of what you told it last week, and it's running on the wrong settings by default. It's not that the tool is weak. It's that nobody's onboarded it.

That's what an induction fixes. That's what a chatbot never gets.

These are the settings you flip in the first ten minutes on the job. Small changes, biggest immediate jump in quality.

1. Switch the model and the effort.

Look at the top-left of any chat window and you'll find the model picker. If you're on Pro, choose Opus 4.8, and set the Effort to High when the task is complex. That single combination is the biggest quality jump you can make in ten seconds, and most people never make it. They're getting Haiku-tier replies from an Opus-tier tool because they've never opened the dropdown.

One caveat before you leave it on all day: Opus eats through your usage allowance much faster than the default model, so save it for the work that deserves it, like your proposals, your analysis, anything you'd have to redo if it came back mediocre. For everyday questions and mundane tasks, drop back to the default. You want the right model for the job, not the biggest model for everything.

2. Turn on Web search and Research.

Hit the + button in the chat box and you'll see the toggles for Web search and Research. Turn both of them on. Without them, your new hire is working from memory that's months out of date; with them, it actually goes and reads live sources before it answers you.

3. Pick your workspace mode.

Claude recently redesigned this, so let me describe what you'll actually see today. Code now sits on its own on the left side, and Chat and Cowork live together as a toggle inside the chat box itself, so you flip between them depending on the task. Most people never leave Chat.

Here's the difference. Chat answers you: you ask a question, it responds. Cowork actually does the work: it builds the real spreadsheets, accesses your folders, creates real PDFs, and saves the finished work back to your folder.

If you're a non-technical business owner, Chat and Cowork are your home, and Cowork is where Claude stops talking about the work and starts doing it for you. That's the difference between a new hire who talks about the task and one who quietly gets on with it.

4. Turn on memory.

Click your name at the bottom of the left bar and open Settings, then go to the Capabilities tab. In there you'll find "Search and reference chats". Toggle it on. From that moment, Claude remembers what you told it last week, and you stop re-explaining yourself every Monday morning. The longer you use it with memory on, the sharper it gets, because it's actually learning how you work.

That's four settings in ten minutes, and your new hire's first morning is done.

This is the actual induction. The bit where you sit your new hire down and tell them who you are.

5. Write ONE about-me file.

This is the highest-leverage ten minutes you'll ever spend in Claude. Write one file that says who you are, what you do, who your customers are, how you sound, and what your non-negotiables are. It doesn't have to be fancy. Then drop it into a Project or paste it into your Global Instructions. From then on, every output comes back in your voice, with your context, without you re-explaining a thing. You can boost it by adding a few of your past emails or posts, so it captures how you actually sound.

Most owners skip this step entirely, and it's the reason Claude sounds like Claude instead of sounding like them.

6. Set up Projects with your business context.

Open the top-left bar, go to Projects, and create a new one. Name it, add your instructions, and upload the files that belong to that job; every chat inside the project follows those instructions and remembers everything in there. Build a separate Project for each use case: one for sales calls where you drop in your call recordings and your sales playbook, one for content with your brand voice document and your top posts, one for client comms with your tone examples. Each Project becomes a version of Claude that's been trained on that one job, and it works in both Chat and Cowork.

7. Connect Gmail, Drive, Calendar.

Go to the top bar and click Customize, then connect your apps. Search for the tools you use every day, your Gmail, your Drive, your Canva, your Slack, and link each one. Now Claude can read your actual inbox, your actual files, and your actual calendar, so when you type "read my email" it just goes and does it, without you explaining where to look.

One warning from me: only connect the tools you genuinely use regularly, because every connector you switch on eats tokens. Something you touch once a year has no business being connected.

8. Install your first skill.

A skill is a job Claude has learned to do end to end; not a prompt, not a checklist, but a saved workflow that runs from the starting instruction to the finished output. Go to Customize and click into skills, and you'll see three ways to create one: upload a skill someone shared with you, write the instructions yourself, or simply tell Claude what the skill should do and let it build one for you.

My favourite shortcut: when Claude does a task well in a conversation, tell it to turn that conversation into a skill. The whole job becomes rerunnable any time you need it again.

Start with just one. Inbox triage is the highest-ROI skill I've seen with business owners: build it once, run it every day.

That's the induction over. Your new hire now knows who you are, has access to your stuff, and has learned its first job. This next one is where it stops being a hire at all.

9. Set up your first scheduled task.

A scheduled task is exactly what it sounds like: you tell Claude "every morning at 6am, run this job," and it does, forever, without you opening anything.

You'll find Scheduled tasks in the top bar. Click new task and either describe the job to Claude and let it set everything up, or build it manually: name it, describe what it should do, pick the folder it writes to, decide whether it asks permission first or acts without asking, and set the frequency, hourly, daily, weekly or weekdays. Save it, and off it goes.

Mine runs at 5am every weekday. It pulls my calendar, my inbox, my recent invoices, and writes me a 1-page brief. By the time I'm on my first coffee, my morning is already planned.

That's what an employee does. A chatbot waits for you to open the app.

You set it up once, and it runs every day, unattended, whether you remember it exists or not. The compound effect over a month is enormous, because you stop spending your sharpest hour of the day on admin that something else could be doing for you.

That's what an employee does. A chatbot waits.

Honest take. You don't need all 9 on day one. Nobody does a full induction in an afternoon. The first 3 will get you 80% there. Add the rest as you find tasks that need them.

If you want to watch me set these up on real client work, the spreadsheet, the brief, the lead list, all built in front of you, I'm going Live on LinkedIn on Thursday 16 July at 6:30pm BST. 45 minutes. Free. I'll demo the inbox triage, the lead-gen job, and the morning brief.

Everyone who registers gets a free skill on signup. It's the one that makes you the answer when people ask ChatGPT or Claude about your niche.

Save your seat: https://bit.ly/abi-claude-live

Watch the full 22-min walkthrough on YouTube: https://youtu.be/iW-zwnpa-zQ

Reply and tell me which setting you're turning on first. I read every one.

Until next week, build something.

Abi O · CodeFreeIQ

PS. If reading about settings is less fun than having the whole thing done for you, that exists now. The AI Employee Installer interviews you, then builds your entire Claude setup: your voice, your guardrails, and 4 working systems, in about 15 minutes. Get it here: https://abiodedeyi.nestuge.com/ai-employee-installer

Claude's most powerful model is free until Sunday 12 July, then it moves to pay-per-use. I put the 4 jobs actually worth giving it, with the exact copy-paste prompts, on one page: https://links.codefreeiq.com/guides/fable5-prompts

Run the contract read first. It pays for itself.👉 Save your seat: bit.ly/abi-claude-liveSave your seat: https://bit.ly/abi-claude-live