After the last article, a lot of you asked the obvious follow-up: "Okay, I'm saving tokens now. But how do I actually get better output?"
Fair question. Saving tokens and getting great results are two different skills. One is about efficiency. This one is about depth.
These are seven features sitting inside Claude right now — most of them free, all of them underused — that separate people who treat Claude like a chatbot from people who treat it like an actual collaborator.
1. Projects: Stop Re-Explaining Yourself Every Single Time
If you've ever pasted the same background information into Claude five times in one week, this is the fix.
Projects is a workspace feature that groups related conversations, uploaded files, and custom instructions under a persistent context. The project's instructions, files, and conversation history are accessible across every chat inside that project.
Set up one Project for your brand voice, your client's brand guidelines, or your codebase. Upload the reference files once. Every new conversation inside that Project already knows the context — no re-explaining, no re-pasting.
Where to find it: claude.ai → sidebar → Projects → New Project. File uploads are capped at 20 files per chat, 30MB each — more than enough for most workflows.
2. XML Tags: The Structure Trick Almost Nobody Uses
This is, without exaggeration, one of the highest-leverage formatting changes you can make to a prompt — and it takes about ten extra seconds to apply.
When your prompts involve multiple components like context, instructions, and examples, XML tags can be a game-changer. They help Claude parse your prompts more accurately, leading to higher-quality outputs.
Instead of writing one long paragraph mixing your context, your instructions, and your examples together, separate them:
<context>
This is a LinkedIn post for a B2B SaaS company targeting CTOs.
</context>
<instructions>
Write a 150-word post. Lead with a stat. End with a question.
</instructions>
<example>
[paste a post you liked the tone of]
</example>Claude recognizes XML tags more reliably than plain line breaks or fenced code blocks when the prompt contains multiple parts. The result: fewer misread instructions, fewer "that's not quite what I meant" follow-ups, and dramatically more consistent output.
3. Prefill: Skip the Preamble, Start Exactly Where You Want
This one is mostly known to developers using the API — but it solves a problem every single Claude user has run into.
Prefilling is the ability to guide Claude's response by starting the Assistant message yourself. This powerful technique allows you to direct Claude's actions, skip preambles, and enforce specific formats like JSON or XML.
Translation for everyday use: if Claude keeps starting every response with "I'd be happy to help with that!" before getting to the actual content, you can short-circuit it. On the API, you literally write the first few words of Claude's reply and let it continue from there.
Prefill beats cleanup — make the model start from the right position instead of repairing the output afterward. It's the difference between editing out fluff after the fact and never generating the fluff in the first place.
On claude.ai directly, the equivalent move is simpler: explicitly instruct "Skip the preamble. Start your response with [X]." It won't be as precise as true prefill, but it gets you 80% of the benefit with zero technical setup.
4. Extended Thinking: Make Claude Actually Think Before It Answers
For genuinely hard problems — complex reasoning, multi-step analysis, anything where the first instinct might be wrong — Claude has a dedicated thinking mode most casual users never touch.
Extended thinking allocates a dedicated pre-response reasoning budget, visible as a "Thinking" section above Claude's response. Performance on complex tasks improves logarithmically with thinking tokens — meaning the more Claude is allowed to reason before answering, the better the output gets, especially on genuinely difficult problems.
It's particularly effective for complex coding problems, mathematical reasoning, and strategic analysis with multiple competing factors.
When to use it: Don't turn this on for "write me a tweet." Save it for the problems where you'd genuinely want a smart colleague to pause and think before responding — a tricky architecture decision, a multi-variable strategy question, debugging something subtle.
5. Multishot Prompting: Show, Don't Just Tell
This is the single fastest way to fix Claude producing output that's technically correct but stylistically off.
Instead of describing what you want in abstract terms ("make it punchy," "more professional," "less corporate"), give Claude 2-3 examples of exactly the tone and structure you're after. This is called multishot prompting, and it consistently outperforms pure description.
The reasoning is simple: "professional" means something different to every person. A concrete example removes all the ambiguity. Claude pattern-matches to your examples instead of guessing at your adjectives.
The fastest fix for inconsistent output: Build a small folder of your 3 best past outputs — emails, posts, reports — and paste one in as a reference every time you start a similar task. Pair this with Projects (tip #1) and you've built a permanent style reference Claude uses automatically.
6. The /btw Command: Ask Without Breaking Claude's Flow
This is specific to Claude Code, but the underlying idea matters for anyone running longer agentic tasks.
While Claude is actively working, use /btw to ask a quick question without interrupting it. It's single-turn with no tool calls, but has full context of the conversation.
Normally, interrupting a long-running task means Claude has to stop, address your question, and then figure out where it left off — losing momentum and sometimes losing the plot entirely. /btw sidesteps that. You get your quick answer. Claude keeps working uninterrupted in the background.
If you're not a developer, the underlying lesson still applies broadly: don't derail a productive thread with an unrelated tangent. Open a separate conversation for unrelated questions, and keep your working thread focused on the task at hand.
7. Push Back — Don't Accept the First Draft
This is less a feature and more a habit, but it might be the most underused "trick" of all seven.
Don't accept the first solution — push Claude to do better. A few prompts that work well: "Grill me on these changes and don't make a PR until I pass your test." Forces Claude to validate your understanding before shipping. "Prove to me this works." Or simply: "Knowing everything you know now, scrap this and implement the elegant solution" — useful after a mediocre first attempt.
Most people treat Claude's first response as the final answer. Power users treat it as a first draft. The single most impactful habit you can adopt, according to Anthropic's own Claude Code team, is verification — giving Claude a way to check its own output rather than trusting it blindly.
That habit alone — asking "are you sure?" or "what would make this better?" — consistently produces a second pass that's meaningfully sharper than the first.
The Pattern Behind All Seven
Look at what these seven tips have in common. None of them are about a secret setting buried in a menu. They're about treating Claude less like a search bar and more like context engineering — structuring everything Claude receives before and during a task, including system prompts, files, memory, examples, and constraints, rather than focusing only on the wording of a single message.
In 2026, model quality has improved to the point where the structure around a task matters more than how cleverly the task is phrased.
That's the real shift. The model is already smart. What most people are missing isn't intelligence — it's the scaffolding around the conversation that lets that intelligence actually show up.
Most people use Claude the same way they use every other AI: type a question, get an answer, rinse and repeat. Claude is built for more than that. These seven features are the difference.
Which of these are you already using — and which one are you trying today? Drop it in the comments. I want to know which one moves the needle most for you.
Discussion