10 Claude Tips That Get Better Results and Cut Your Usage in Half

Most people use Claude the same way they use a search engine: type a question, read the answer, type a follow-up, repeat. It works. It is just not the most effective way to work with a large language model — and if you are on a subscription plan or watching usage closely, it is also significantly more expensive than it needs to be.

Claude is not a search engine. It is a model that reads your entire conversation — every message, every response — from the beginning each time it generates a reply. That means the habits that make you efficient with Google (short queries, lots of follow-ups, open tabs for later) actively work against you with Claude. The way you structure your inputs, which model you choose, what settings you leave on, and when you start a new conversation all have meaningful impact on both the quality of what you get back and how much of your monthly usage those results consume.

These ten habits will make you better at both. They are organized into two categories: getting sharper, more useful results, and doing it without burning through your message allowance. Many of them overlap — good prompting tends to be efficient prompting.

Part 1: Getting Better Results

Tip 1: Structure Every Prompt with GCAO

Most people fire off prompts the same way they send a text message — short, context-free, hoping the other party figures it out. With Claude that produces generic, often unusable output. GCAO is a four-part structure that turns any prompt from a guess into a clear brief. It takes 60 seconds to apply and eliminates most of the back-and-forth that wastes messages and burns through your usage.

G — Goal: What do you actually want to happen as a result of this conversation? Not the task — the outcome. “I want small business owners who feel overwhelmed to reach out about working together” is a goal. “Write me some content” is not.

C — Context: What does Claude need to know about your situation to give you a relevant answer? Your role, your industry, what you have already tried, constraints you are working with, your audience. Context is what separates a useful response from a generic one.

A — Action: The specific thing you are asking Claude to do. Not the topic — the task. “Identify the type of content that attracts buyers rather than followers” is an action. “Help me with content” is not.

O — Output: What the response should look like. Format, length, structure, what to include, what to leave out. “Give me 3 content angles with a one-sentence explanation of why each attracts buyers specifically. No generic advice about consistency or hashtags.”

Together, those four elements produce a prompt that is specific, grounded in your situation, and constrained to exactly what you need. Here is what a complete GCAO prompt looks like in practice:

Goal: I want to attract small business owners who are already making money but feel overwhelmed running everything by themselves. I’m not trying to get famous — I want the right people finding me and reaching out about working together.

Context: I run a coaching business doing about $15K/month. I post on Instagram and LinkedIn but I get generic engagement — likes from people who will never buy. I’ve tried posting tips and motivational content. It hasn’t translated to clients. I have about 45 minutes a day to spend on this.

Action: Identify the specific type of content that attracts buyers rather than followers for a service-based business. Explain why that content works differently than content designed for reach.

Output: Give me 3 content angles with a one-sentence explanation of why each one attracts buyers specifically. No generic advice about consistency or hashtags.

✅  For recurring tasks — weekly reports, email drafts, client summaries — save your role-and-goal framing as a User Preference in Claude's settings. It carries into every conversation automatically. See Tip 5.

Tip 2: Be Specific About Format Before You Ask

Claude will decide how to format its response if you do not tell it otherwise. Sometimes that works well. Often it produces a long, structured answer when you wanted a short paragraph, or a paragraph when you needed a bulleted list you could paste into a slide.

Specify format in the prompt itself: 'respond in three bullet points, each under 15 words' or 'write this as a two-paragraph email, professional but direct' or 'give me a table with three columns: risk, likelihood, and mitigation.' Claude follows explicit format instructions reliably — but it cannot read your mind about what would actually be useful.

Length is part of format. If you want a short answer, say so explicitly. 'In one sentence' or 'in under 100 words' or 'give me the quick version first, then I will ask for detail if I need it' — these constraints do not limit Claude's quality, they direct it toward what you actually need.

Tip 3: Give Claude Examples of What Good Looks Like

Describing what you want is harder than showing it. If you have an example of the tone, format, or style you are going for — a past email, a sample paragraph, a table from a previous project — paste it in and say 'write something like this.'

This works especially well for writing tasks where tone is important. 'Write a LinkedIn post about our new service' leaves a lot of room for interpretation. 'Write a LinkedIn post about our new service in the style and tone of this one: [paste example]' produces something much closer to what you actually want on the first attempt.

One good example beats five adjectives. If you find yourself piling on descriptors like 'professional, warm, concise, authoritative, not too salesy' — stop and find an example instead. The example communicates all of those things simultaneously.

Tip 4: Use Extended Thinking for Hard Problems — But Only Then

Extended Thinking is a mode where Claude reasons through a problem step by step before writing a response. For complex analysis, multi-step logic, difficult coding problems, or decisions with many interdependencies, it produces significantly better results. For straightforward questions, it is overkill.

The right workflow: attempt the task normally first. If the result is shallow, misses something important, or feels like it cut corners on the reasoning, then enable Extended Thinking and try again. Do not leave it on by default — it uses significantly more tokens per response, which adds up quickly across a day of regular use.

Extended Thinking is available on Claude.ai Pro, Team, and Enterprise plans. Enable it per-conversation from the model selector, not as a persistent setting. Reserve it for the 10–20% of tasks that genuinely require deep reasoning.

⚠️  Extended Thinking is not faster — it takes longer to respond because Claude is doing more work. If you are under time pressure or asking a simple question, it will only slow you down.

Tip 5: Use XML Tags to Separate Instructions from Content

When your prompt mixes instructions, context, and the actual content you want Claude to work on — a document to summarize, data to analyze, an email to rewrite — Claude can sometimes blend them together or miss where one ends and the other begins.

XML tags solve this cleanly. Wrap different types of content in descriptive tags and Claude parses them without ambiguity:

<instructions>

Summarize the following client email in 3 bullet points.

Focus on action items and deadlines only.

</instructions>

<email>

[Paste the email here]

</email>

You can use any tag names — the names do not have to match anything specific. The structure is what matters. Common tags: `<instructions>`, `<context>`, `<document>`, `<data>`, `<example>`, `<task>`. This technique becomes especially valuable when you are working with long documents or complex multi-part requests.

Part 2: Getting More Out of Every Message

The tips in this section address how Claude actually processes conversations — specifically, the fact that every message in a thread costs more than the last one. Understanding this helps you make different decisions about when to start fresh, when to edit instead of reply, and which features to leave switched off.

Tip 6: Edit Your Original Message Instead of Sending a Follow-Up

This is the highest-impact usage habit most people have never heard of. Claude re-reads your entire conversation history every single time it generates a response. Every message — yours and Claude's — gets processed again from the start.

When Claude gives you something that misses the mark, the instinct is to send a follow-up message correcting it. Do not do this if you can avoid it. Instead, click the edit button on your original message, revise the prompt, and regenerate. The old response gets replaced rather than added to the history. You get a better answer from a cleaner starting point — and over a 10-message back-and-forth, this habit alone can cut token usage by 80–90%.

✅  The edit button appears when you hover over any message you have sent. On mobile, tap and hold. This is one of the most underused features in Claude.ai.

Tip 7: Start a New Conversation Every 15–20 Messages

This connects directly to the point above. Your first message in a conversation might cost a few hundred tokens to process. By message 30, that same simple question costs tens of thousands of tokens — because Claude is re-reading the entire preceding conversation to generate each response.

Long conversations accumulate cost and can actually degrade quality: older context competes with your current question for Claude's attention. The solution is deliberately simple — start a fresh chat. If you need continuity, ask Claude to summarize where things stand before you open the new tab. Paste that summary in as the first message of the new conversation.

💡  Claude can generate its own handoff summary on request: 'Before I start a new conversation, give me a concise summary of the key decisions, context, and open questions from this thread.' Copy it, open a fresh chat, paste it in.

Tip 8: Use the Right Model for the Right Task

The Claude model family has three tiers, each designed for a different class of work. Using the most powerful model for every task is like using a heavy truck to run errands — the capability is there, but you are burning resources you do not need.

A practical rule: use Haiku for anything you would ask a capable assistant to do in two minutes. Use Sonnet as your everyday model for substantive work. Use Opus when you have tried Sonnet and the result is not deep or careful enough for what you need.

On the Claude.ai interface, the model selector appears at the top of any new conversation. The model you choose stays until you change it — it is worth building the habit of checking it before you start.

Tip 9: Set Up Memory and User Preferences Once, Save Tokens Forever

Every conversation you start without stored context burns several messages just establishing who you are, what you do, and how you like responses formatted. Across dozens of conversations a week, that is a significant amount of wasted back-and-forth.

Claude has two tools for this. Memory captures facts about you from your conversations and surfaces them automatically in future ones — your role, your industry, the tools you use, recurring projects. User Preferences let you set explicit instructions that apply to every conversation: your preferred tone, formatting preferences, whether you want Claude to ask clarifying questions or just make reasonable assumptions.

• Navigate to Settings → Memory to review what Claude has stored and add or edit facts you want carried forward.
• Navigate to Settings → User Preferences to write standing instructions — things like 'I work in IT for Canadian SMBs,' 'be direct and skip unnecessary caveats,' or 'format code examples in code blocks.'
• Memory and User Preferences together mean Claude already knows the baseline context every time you start a new conversation. You spend your messages on the actual task.

✅  If you are part of an organization using Claude for Work (Team or Enterprise plan), your IT administrator can set organization-level context through the system prompt — so Claude knows your company, your tools, and your preferred working style without individual employees needing to configure anything.

Tip 10: Turn Off Features You Are Not Using

Several Claude features add tokens to every response, even when you are not actively using them. Left on by default, they silently consume usage across every conversation.

• When Web Search is enabled, Claude may search the internet to supplement responses — even for questions where its existing knowledge is sufficient. Turn it off when you are working with internal documents, doing writing tasks, or asking questions that do not require current information. Enable it specifically when you need recent data.‍
• Research Mode runs multiple searches and synthesizes results from many sources. It is powerful for genuine research tasks but uses many times the tokens of a standard response. Leave it off unless you are specifically doing research that requires current, cross-referenced information.Research Mode:
• If Extended Thinking is enabled by default in your settings, turn it off. Use it selectively, as described in Tip 4. Running it on routine tasks is one of the fastest ways to exhaust a message allowance.  
• If Claude has integrations connected — Google Drive, Microsoft 365, Slack, and others — those Connectors add context to responses whether or not you are working on anything related to those systems. Review which connectors are active and disable any you are not actively using.

📋  A quick settings audit: open Claude Settings, review Memory, User Preferences, Connected Apps, and the default model. Five minutes of configuration pays off across every subsequent conversation.

Bonus: Chain Tasks Instead of Combining Them

One last habit worth building: when you have a complex task, break it into steps rather than asking Claude to do everything in a single prompt.

'Research the topic, summarize the key points, write a 500-word blog post, suggest a title, and format it for our website' is a real request people send. The problem is that Claude has to make a large number of judgment calls without knowing your preferences for any individual step — and if the research takes a direction you would have redirected, the entire output is built on a foundation you didn't choose.

The better approach: research first ('summarize the five most important points about X'), review the summary, redirect if needed, then ask for the draft. You stay in control of each step. The output at the end is more reliably aligned with what you actually wanted — and because you can edit at each stage rather than starting over from scratch, it is faster.

💡  Chaining tasks also gives you natural checkpoints to use the edit-instead-of-follow-up habit from Tip 6. Each step in the chain is its own small conversation you can redirect cleanly.

Quick Reference: The 10 Habits at a Glance

Frequently Asked Questions

Which of these tips has the biggest impact on usage?

Editing your original message instead of sending follow-ups is the single highest-leverage habit for most users — it directly prevents the conversation history from growing, which is the root cause of token cost escalation. Combined with starting a fresh conversation every 15–20 messages, these two habits alone account for most of the efficiency gains available.

Does using a lower-tier model mean worse results?

For the tasks each model is suited to, no. Claude Haiku is genuinely excellent at what it was designed for — quick answers, reformatting, grammar, brainstorming. Using Haiku for those tasks and reserving Sonnet for substantive work does not feel like a downgrade; it feels like having the right tool. Where you notice a difference is when you try to use Haiku for complex analysis or advanced coding — those are Sonnet or Opus tasks.

What is the difference between Memory and User Preferences?

Memory is automatically generated from your conversations — Claude notices facts about you (your job, your tools, recurring projects) and stores them so it has context in future chats. User Preferences are explicit instructions you write yourself, like standing orders that apply to every conversation. Both live in Settings. Memory is passive and accumulates; User Preferences require you to write them once. Using both together means Claude starts every conversation knowing who you are and how you like to work.

Is there a risk that editing messages will lose something useful from the original response?

Only if you want to keep it. If Claude's first attempt had something useful but missed the main goal, copy the useful parts before you edit and regenerate. The edit-and-regenerate workflow replaces the response entirely — the old one is gone. Treat it like a first draft: extract anything worth keeping, then improve the prompt to get a better version.

Does Always Beyond help organizations set up Claude for Work?

Yes. For organizations using Claude for Work on Team or Enterprise plans, Always Beyond can configure the system prompt and organization-level settings, set up integrations with your existing Microsoft 365 environment, and train your team on effective prompting practices for your specific workflows. If you are evaluating whether Claude for Work is the right fit for your organization, reach out for an assessment.

Getting the Most From Your AI Tools

Claude is a genuinely powerful tool for business productivity — but like most powerful tools, how you use it matters as much as whether you use it. The habits in this list are not advanced techniques. They are the basics that experienced users internalize quickly and that most new users figure out slowly and inconsistently.

Building them early saves time, reduces frustration, and makes the difference between AI that occasionally impresses you and AI that reliably helps you get work done faster.

Want help rolling out AI tools effectively across your team? Always Beyond helps Canadian businesses implement Claude for Work, Microsoft Copilot, and AI governance policies that get real results — not just licences. Reach out to start the conversation.