Claude setup: 6 configurations that 99% of people ignore

By default, every conversation with Claude starts from scratch. It does not know you have a dental clinic, a salon in Miami or a B2B consultancy. You have to explain it every time. That ends with Memory active.

Go to Settings → Memory, turn it on and paste a short, explicit profile of your business. Something like this:

I'm [YOUR NAME], owner of [BUSINESS NAME],
a [TYPE OF BUSINESS] in [CITY].

My clients are [PROFILE: age, economic profile, behavior].
My value proposition is [WHAT SETS US APART IN ONE SENTENCE].
My main channels are [WHATSAPP, INSTAGRAM, GOOGLE, ETC.].
The tools I already use are [CRM, SCHEDULING, BILLING, ETC.].

When I ask you for help, assume this context.
If something doesn't fit my business, ask me before making things up.

From that moment on, every chat starts with that context loaded. When you ask it for a welcome email, it already knows it is for your dental clinic. When you ask for content ideas, it knows your audience is local and specific.

Loose chats are the equivalent of having all your files on the desktop. It works for the first 5 days. Then you cannot find anything and you redo work. Projects solve that.

Create a Project for each active client, for each business you have, or for each big goal (launch, campaign, hiring). Upload the key documents there: service catalog, price list, brand guidelines, ICP, transcripts of important calls, etc.

Claude reads all those documents in every conversation inside the Project. You do not have to paste anything again. Any answer comes out already aligned with your brand and your product.

Even with Memory and a Project, if you ask poorly, you get poorly. The difference between a Wikipedia answer and an answer you can send to a client is in how you phrase it.

The amateur way:

How do I write an email for clients?

That gives you four generic sentences you would not use.

The professional way:

I own a dental clinic in NJ.
I have 80 clients who haven't come back in 9+ months.
I want to win them back with an email.

Goal: get them to book a $79 cleaning (discount from the usual $120).
Tone: warm, professional, no pressure. None of that "hello dear client".
Format: short subject (max 8 words) + body of 4 short paragraphs
        + clear CTA to the "Book my cleaning" button.
Constraint: no emojis. Don't mention inflation.

Give me 3 versions with different tones.

Same AI, same model, completely different result. The formula is: situation + goal + format + constraints. It does not have to be perfect: with a couple of lines in each block, you already multiply the quality of the output.

Claude has two main models that most people use backwards:

• Sonnet— fast, cheap, capable. 80% of daily work: emails, content, summaries, ideas, Q&A about your documents. By default, this one.
• Opus — slower and more expensive, but with deeper reasoning. Use it when there is a strategic decision, complex code, multi-step analysis, or when the answer truly matters.

Using Opus for everything burns through your limits in hours and frustrates you. Using only Sonnet when you need depth leaves you with shallow answers on decisions that matter. Having a clear criterion to choose is the shortcut.

If you do the same thing every week (generate content ideas, draft follow-up emails, summarize meetings), and each time you explain the format to Claude from scratch, you are working twice.

Skills are saved workflows: you define once the input you expect, the rules, the output format and the tone. Then you invoke the Skill in a single line and it comes out the way you want.

Useful Skills to start with:

• Weekly newsletter — takes 3 points, expands them into 600 words in your tone.
• Call summary — paste transcript, returns decisions, next steps and owners.
• Sales proposal — input of 4 client data points, output of a proposal ready for PDF.
• Instagram caption — input of topic and format, output of copy + hooks + hashtags.

Up to here Claude responds, drafts and reasons. But everything it proposes you have to copy and paste somewhere else. Connectors change that: they plug Claude directly into the tools where you work.

• Google Drive — reads and writes on your real documents without downloading or pasting them.
• Canva / Figma — design visual pieces without leaving the flow.
• n8n / GitHub — automate tasks and push code without opening three tools.
• Notion / Linear / Slack — create, move and comment on your real tasks and conversations.

The mental shift is important: Claude stops being a copilot that suggests and becomes an operator that executes. The difference in productivity is enormous once you get used to it.