Home / Blog /AI Automation/ How to set up Claude AI before you start prompting

By Traci Gurney | Last updated: May 2026 | 9 min read

THE SHORT ANSWER

Most people open Claude and start typing, then wonder why the answers feel generic. The problem is setup, not the tool. Before your first prompt, you need to configure privacy settings, custom instructions, memory, and optionally projects, and that ten-minute investment changes how useful Claude actually is.

KEY TAKEAWAYS

• Turn off training data sharing first. Go to Settings > Privacy and disable “Help improve Claude,” otherwise every conversation you have feeds their training data.
• Custom instructions eliminate repetition Write who you are, how you want answers formatted, and what Claude should never do. Every conversation then starts from that baseline.
• Projects keep different work from bleeding into each other. Each project has its own instructions, files, and memory, so client A’s context never pollutes client B’s output.
• Artifacts are built things, not just text. When you ask Claude to create a form, report, or calculator, it builds it in a side panel you can download or interact with, not just describe it.
• Connectors let Claude take actions, not just generate text. Link Google Calendar, Gmail, Drive or other common biz tools and Claude can read, create, and edit inside those tools directly.

Watch & Read · Your Call!

This comes straight from my YouTube video. Same breakdown, just written out so you can follow it at your own pace. Pick whichever way you learn best.

Prefer to watch instead? I walk through the whole thing on my channel.

Go to claude.ai and sign up with Google or email. The free plan works, but I recommends at least the $20 Pro plan, it gives five times more usage, which you will notice quickly. Claude is also available as a desktop app and on mobile. The interface may change as Anthropic adds features, but the underlying functions stay the same.

Go to Settings > Privacy and turn off "Help improve Claude." When this is on, every conversation you have is used to train their AI. On the free plan you may not be able to disable this. Pro plan users should turn it off before doing anything else.

Click the dropdown at the bottom of the chat box to choose your model. Sonnet handles most day-to-day work quickly. Opus is for heavy tasks, untangling legal language, summarizing a 90-page technical report, or anything with a lot of moving parts. Haiku is for speed when you just need a quick answer. Claude also has a thinking feature (called extended thinking or adaptive thinking depending on the model), turn it on for complex decisions, leave it off for routine tasks since it costs more usage and takes longer.

Go to Settings > Capabilities and enable both memory options. "Search and reference chats" lets Claude pull relevant details from past conversations. "Generate memory from chat history" builds an automatic profile of your preferences, work style, and patterns, updated roughly every 24 hours. When memory is off, Claude forgets everything the moment you close the session. If you want Claude to remember something immediately, just say "remember this" in the chat rather than waiting for the 24-hour cycle. You can view, edit, or delete individual stored memories from the same settings page.

Go to Settings > General and fill out the custom instructions field. Cover four things: who you are, what you care about, how you want answers formatted, and what Claude should never do. I find the most valuable parts are the format preferences and the "never do" list, for example, stopping Claude from giving long-winded answers when you just want a yes or no. Fill this out once and it applies to every conversation automatically.

Click Projects in the left sidebar and create a separate project for each client, workflow, or topic area. Each project gets its own custom instructions, its own uploaded files, and its own memory. This means Claude knows the full context for that specific project without you re-explaining it every time, and nothing from one project crosses into another. Inside a project you can write a brief, upload client documents, and then prompt without repeating background details.

Go to Customize > Connectors to link external tools. Claude can connect to Google Calendar, Gmail, Drive, Canva, and others. Once connected, Claude can take real actions, checking your calendar, creating events, finding scheduling conflicts, not just describe what you could do. Traci recommends setting read-only as always allowed and requiring approval for any action that creates or deletes something.

Go to Customize > Skills to store processes you repeat often. You teach Claude the format or workflow once, save it as a skill, and Claude follows that process automatically whenever it's relevant. Skills work across all your projects. You activate a saved skill by typing a forward slash followed by the skill name.

Make sure to setup Claude properly from the start.

Go through the Claude AI Setup Guide here →

I connected my Google Calendar through the Connectors settings. Then asked Claude "what's on my calendar tomorrow?" and Claude pulled the actual events. I then asked it to rename a 1:00 pm meeting, clicked Allow when prompted, and the calendar event updated in real time, no manual editing.

I created a demo project called "Sweet Life Bakery website relaunch," added custom instructions covering tone, target audience, and goals, then uploaded a client brief. When I then asked Claude to write copy, it already knew the location (Riverside), the upcoming spring specials, and the brand voice, no re-explaining required in the prompt.

I asked Claude to create a marketing report on whether the bakery should expand. Instead of returning plain text in the chat, Claude built a formatted document in a side panel. I could read it, download it, and use it directly, the artifact panel on the left side stores everything Claude has built so you can find it later.

I then asked Claude to help create a skill for writing Facebook posts using a specific format: hook, main point, call to action. Claude used its built-in skill creator to draft the process, let me test it, then saved it. Now whenever I types /facebookpost, Claude follows that exact structure automatically across all projects.

To test how Claude responds without any personalization, use the incognito option (the small icon near the top of the interface). This opens a session with zero memory, zero custom instructions, and no context, useful for seeing Claude's default behavior or working on something you don't want saved to history.

Turn your content into a system that grows your business