Getting More From Claude: A Mid-Level User Guide

Most people start using Claude the same way — paste a question, read the answer, try again. That works, but there is a significant gap between using Claude as a fast search engine and using it as a genuine thinking partner. This guide is written for people who have moved past the basics and want to understand the commands, techniques, and structural approaches that make the biggest practical difference — without needing to touch the API or write any code.

Where Claude Lives — Choosing the Right Interface

Claude is available in several forms, and they serve different kinds of work. Using the right one for the task removes friction before you even type a word.

For most non-developer use, claude.ai with Projects enabled is the right starting point. For anything involving a codebase, Claude Code is in a different category entirely.

Writing Prompts That Actually Work

Prompting is not a mystical skill. It is the same discipline as briefing a colleague clearly: the more information you give about what you want, who you are, and what constraints apply, the better the output. These six patterns cover the majority of what separates useful responses from vague ones.

The single most common mistake is being vague about the output. "Tell me about X" gets you an encyclopedia entry. "Give me a three-sentence summary of X written for someone who has never heard of it, then list the three key risks" gets you something immediately usable. The length of the prompt is not the issue — the precision is.

Equally important: iterate rather than restart. If the first response is 80% right, correct it in the same conversation rather than reprompting from scratch. Claude can refine, rephrase, shorten, expand, or completely restructure output mid-conversation. Tell it what needs to change and it will change just that part.

Projects: Giving Claude a Persistent Memory

By default, every new conversation starts from scratch. Claude knows nothing about your organisation, your preferences, or the context you explained three days ago. Projects — available on claude.ai — solve this directly.

A Project is a persistent workspace. You give it a system prompt (standing instructions that apply to every conversation within it) and a set of knowledge files (documents, specifications, style guides, code snippets, or anything else Claude should always have access to). Every conversation you start inside that Project inherits all of it automatically.

A practical example: create a Project for a recurring work task. Set the system prompt to something like: "You are helping me with customer communications for Deriva Energy. Our tone is professional but direct. Never use jargon. Our escalation contact is the operations team." Add your email templates and product documentation as knowledge files. From that point, every conversation in that Project starts with Claude already briefed — no re-explaining the basics, no inconsistent tone.

Understanding the Context Window

The context window is Claude's working memory — the total amount of text it can hold in mind at once, covering the full conversation including your messages, its responses, and any files you have shared. Once a conversation grows long enough to fill the context window, earlier parts of the conversation begin to fade from view.

In practice, this means three things. First, very long conversations can cause Claude to lose track of details from the beginning — if something important was established early, restate it or paste it back in when it becomes relevant again. Second, pasting the actual text of a document into the conversation is more reliable than asking Claude to remember something from a prior session (use Projects for that). Third, when a conversation feels like it has drifted or gone stale, starting a fresh one is often faster than trying to realign a long one.

In Claude Code, the /compact command compresses the conversation history to free up context space — useful on long coding sessions without losing the thread of what you were building.

Slash Commands in Claude Code

If you use Claude Code — the CLI tool — slash commands are the fastest way to control your session. Type / at the prompt and a menu appears. The commands below are the ones that come up most often in practice.

One other shortcut worth knowing: prefix any message with ! to run it as a shell command directly in your terminal, without leaving the Claude Code session. ! git status runs git status and pastes the output into the conversation — useful for checking state without switching windows.

Choosing the Right Model

Claude is available in three capability tiers. The choice matters for both speed and quality — using the most powerful model for every task is like taking a lorry to the shops.

In Claude Code, switch models mid-session with /config. In claude.ai, the model selector sits at the top of the conversation. A practical rule: start with Sonnet, escalate to Opus if the output quality is not meeting the task, and drop to Haiku for any bulk or repetitive processing where speed matters more than depth.

Power Patterns That Change How You Work

Beyond the basics of prompting, there is a set of interaction patterns that experienced Claude users reach for repeatedly. None of them are difficult — they are habits of working that take a session from functional to genuinely productive.

What Claude Cannot Do — and Why That Matters

Understanding the limits is part of using any tool well. Claude has a knowledge cutoff — it was trained on data up to a point in time and does not know about events after that date unless you tell it. For anything time-sensitive, always provide the current context rather than assuming Claude knows it.

Claude does not browse the web in a standard conversation. It also cannot access files on your computer unless you paste their contents in or use a tool like Claude Code that is explicitly given file access. If you are asking about a document, share the document. If you are asking about a live situation, provide the current facts.

Claude will sometimes be confidently wrong, particularly on niche factual questions, precise numerical claims, or very recent events. On anything where accuracy matters — data, citations, statistics — treat Claude as a starting point and verify the specific claims before acting on them. The most effective users treat Claude as a reasoning partner that drafts and structures, not as an authoritative reference.

Making Claude Part of Your Workflow

The users who get the most from Claude are not the ones who use it most — they are the ones who use it deliberately. A few structural habits make a significant difference over time.

Build a small library of reusable prompts for the tasks you repeat. A prompt template for weekly status reports, one for email triage, one for summarising meeting notes — each one tuned over several sessions to the format and level of detail that actually works for you. Store them somewhere you can paste them quickly; the thirty seconds of setup pays back on every use.

Use Projects to hold standing context for the work that matters most. Not everything needs a Project, but anything you return to regularly — an ongoing initiative, a codebase, a client relationship — benefits from having its brief written once and held persistently rather than re-explained every session.

Treat your interactions as iterative rather than transactional. A single well-developed conversation produces better output than five separate one-shot queries. Stay in the conversation, build on what is working, and correct what is not. That is where the real leverage is.