Getting More Out of Claude — What Most People Don't Know
Most people use Claude like a search engine. There's a lot more to it — if you know where to look.
I've been working with Claude almost daily for over a year. At first, I used it like most people: question in, answer out. At some point I started going through the settings — and realized how much I'd been leaving on the table. Here's what actually made a difference for me.
Which Model for What?
Claude has three models. The differences matter because they directly affect how quickly you burn through your quota.
Sonnet is the default. Fast, reliable, enough for the vast majority of tasks. Writing, summarizing, organizing ideas, reviewing code — Sonnet.
Opus is the most thorough model. I use it when I'm analyzing long documents, trying to understand complex relationships, or polishing a text where every sentence has to land. It takes more time per response and uses more quota — but it's worth it when the task calls for it.
Haiku is the fastest. Good for quick questions, simple rephrasing, fast fact-checks. Uses the least quota.
My daily split: 80% Sonnet, 15% Opus, 5% Haiku. If you always run on Opus, you'll hit your limit faster than necessary.
Setting Up Projects
The feature that made the biggest difference. A Project is a dedicated workspace with its own instructions and documents. Everything you put in there applies to every chat within the Project.
In practice: I have a Project for this blog. It contains my style guide, tone of voice, target audience. When I start a new article, I don't need to explain any of that — Claude already knows the context.
Another Project I use for travel planning. It has hotel bookmarks, flight data, notes. When I ask "What works next week in Lisbon?", Claude knows what I'm talking about.
Setting one up: Left sidebar → New Project → Name it → Write instructions → Upload documents. Done in two minutes.
Personal Instructions — Write Once, Always Active
There are two places for instructions that complement each other:
Global (Personal Preferences): Click your initials in the lower left → Settings → enter your preferences under "Personal Preferences". What you write here applies everywhere. For example: "I work in marketing, respond in English, get to the point, no introductions."
Per Project: More specific. In my blog Project, for instance: "Write in the first person, factual but not dry, no lists with more than five items."
Claude takes instructions very literally. If you write "maximum three paragraphs", you get three paragraphs. That's good — but only if you think beforehand about what you actually want.
Saving Tokens
Your quota is limited regardless of your plan. A few things that actually help:
Pack everything into one prompt. Instead of "Make the text shorter" and then "Now more formal", go directly with: "Cut the text by a third and make the tone more formal." Every additional message costs tokens.
New chats for new topics. A chat with 40 messages across five different topics drags the entire context along. That slows things down and burns quota. For a new topic, just open a new chat.
Documents go in the Project, not the chat. If you upload the same PDF in every conversation, you're wasting quota. Upload it to the Project once, done.
Choose your model deliberately. Using Opus for a simple question is like driving a truck to the bakery. Sonnet or Haiku will usually do.
Writing Good Prompts
No secrets, no tricks. The difference between a usable result and a good one almost always comes down to how precise the question was.
What works well: give context, state the task clearly, name the desired format.
An example:
Instead of: "Write me something about hybrid work."
Better: "I'm writing an article for an HR trade magazine. Target audience: team leads in mid-sized companies. Topic: Why hybrid work models fail. 600 words, three sections, one concrete example per section."
The first prompt delivers something generic. The second delivers something you can actually use.
Features Most People Overlook
Artifacts: Claude can generate things directly in the browser — working code, diagrams, tables, presentations. All in the side panel, directly editable. If you only use Claude for text, you're missing out.
Connectors: Over 50 integrations — Gmail, Notion, Google Drive, Slack, GitHub, Figma. Connect once, then Claude can access your data directly. I mainly use it with Google Drive and Notion.
Voice Mode: Available in the app, works bidirectionally. Several voices to choose from, currently English only. Handy when you're on the go and typing isn't practical.
Memory: Since fall 2025, Claude remembers things across chats — your name, your preferences, what you work with. You can view and edit it under Settings.
Which Plan?
| Free | Pro (~$20/month) | Max (from ~$100) | |
|---|---|---|---|
| Models | Sonnet + Haiku | All incl. Opus | All |
| Quota | Limited | 5× Free | 5× or 20× Pro |
| Projects | Limited | Unlimited | Unlimited |
| Voice Mode | Yes | Yes | Yes |
| Connectors | Restricted | Yes | Yes |
Free is enough if you use Claude a few times a week. If you work with it daily, you'll hit the limit fast — that's where Pro makes sense. Max is for people who use Claude as their primary tool and never want to see the limit.
That's it. No magic. Most improvements come from taking twenty minutes to set up Projects and write Custom Instructions. After that, it just works better.
