coco
10 KiB
title, source, author, published, created, description, tags
| title | source | author | published | created | description | tags | ||
|---|---|---|---|---|---|---|---|---|
| 10 Claude features that 90% of users ignore | https://x.com/0xbobaaa/status/2044475727030984900 |
|
2026-04-16 | 2026-04-26 | that’s “how to not waste gas” level advice. i want to talk about something different - how to actually drive faster.if you already use Clau... |
|
that’s “how to not waste gas” level advice. i want to talk about something different - how to actually drive faster. if you already use Claude daily for content, research or code here are 10 things that turn it from a chatbot into a full-blown work tool. most of them are free.
1. projects - a separate workspace for every task
most people open a new chat, dump context, explain who they are and what they need. every single time from scratch. projects fix this. it’s a folder with persistent memory: you upload files, write custom instructions, and every new chat inside the project already knows the full context.
here’s what it looks like in practice. you create a project called “X content”. in the instructions you write: what post style, what tone of voice, what data to use, what words are banned. you upload examples of your best posts. done every new chat in this project starts not from zero, but with full understanding of the task.
i keep separate projects for different workflows: one for posts, another for research, a third for technical work. Claude behaves differently in each because the instructions are different.
projects used to be paid-only. since march 2026 they’re free for everyone.
2. memory - set it once, use it forever
memory is Claude’s long-term recall that works across chats. Claude automatically remembers key things: who you are, what you do, how you prefer to receive answers. but beyond the automatic stuff, you can directly tell Claude what to remember. for example: “remember that i write content for crypto audiences, prefer short paragraphs and hate phrases like ‘in the world of cryptocurrency’ and ‘let’s dive in’”. set it up once and Claude never again opens with “of course! let’s dive into this topic!” there’s also user preferences in settings static instructions that apply to all chats. perfect place for baseline settings: language, format, response style. the token savings are small, but the sanity savings are massive. you stop burning the first 3-5 messages of every chat on calibration.
3. styles - switch formats in one click
few people know this, but Claude has a custom styles feature. these are writing style presets you can switch between. why this matters. you have at least three different formats: a telegram post (short, punchy), an X thread (hooks, structure, CTA), a blog longread (depth, examples, data). typing “write in the style of…” every time is wasteful. instead you create a style, describe the rules, and then just select it before your prompt. Claude instantly switches between “aggressive crypto thread” and “analytical longread” with zero extra prompting. you can create styles manually, or ask Claude to analyze a text sample and generate a style from it.
4. artifacts - not text in chat, but a finished product
when Claude answers in chat that’s text you then copy somewhere. when Claude creates an artifact it’s a separate object: a document, table, interactive widget, HTML page, even a mini app.
for a content creator this means: instead of “write me a comparison table of traders” you get an actual interactive table you can download or share. instead of “sketch me a landing page” a working HTML with design.
artifacts can render React components, SVG graphics, markdown documents, mermaid diagrams. if you need a yield calculator, portfolio tracker or data visualization Claude can build it right in the chat.
most users treat Claude as a text generator. artifacts turn it into a finished product generator.
5. deep research - research on autopilot
standard web search in Claude is a quick lookup across 1-3 queries. deep research is a completely different beast.
you give it a task, Claude formulates dozens of search queries on its own, reads sources, cross-references data and delivers a structured report with links. this isn’t “google it for me” it’s full-blown research that would take you a couple hours.
crypto context examples: “analyze the last 3 months of prediction market activity on Polymarket which categories are growing, which are declining, what trends are emerging”. or: “compile info on the top 10 copy-trading platforms in crypto, compare fees and mechanics”.
the result is a ready-made report you can use as the foundation for an article, thread or internal doc.
the feature is available on paid plans. if you make data-driven content this is one of the most useful tools out there.
6. structured prompts - prompt engineering for people tired of redoing things
the advice “write specific prompts” is correct but way too vague. here’s what actually works: structured prompts with markup. Claude understands XML tags perfectly. instead of one long paragraph you break your prompt into blocks:
<role>you're a copywriter for a crypto telegram channel</role>
<task>write a post about trader {name} on Polymarket</task>
<data>
win rate: 73%
profit: $47,000
best trade: "Trump wins" @ 0.62
</data>
<format>
- hook in the first line (question or hard statement)
- 3-4 short paragraphs
- concrete numbers in every paragraph
- CTA at the end
</format>
<constraints>
- no phrases like "in the world of crypto", "let's dive in"
- no emojis at the start of lines
- 150 words max
</constraints>
<example>
[paste an example of a good post]
</example>
the difference is massive. with a prompt like this Claude nails the result on the first try 8 out of 10 times. and every saved iteration is both tokens and time. spend 20 minutes building a template once then use it for months, only swapping out the data.
7. extra usage - don’t overpay for Max if you don’t need it
the Pro plan costs $20/mo. Max 5x - $100/mo. 5x the price, and most people think the choice is binary: either suffer with limits or pay a hundred. but there’s a third option: extra usage. you turn it on in settings, set a monthly cap (how much you’re willing to pay on top of $20), and when your regular limit runs out Claude keeps working at API prices. for most users this is cheaper than Max. you only pay for actual overuse, not a permanent subscription. if you have one heavy week per month and Pro covers the rest extra usage saves you $60-70 monthly compared to Max. important: set a spending cap. without one you can accidentally rack up a solid bill, especially if you use extended thinking.
8. connectors - Claude inside your ecosystem
Claude can connect to external services: Google Drive, Gmail and others. this isn’t just “upload a file”. it’s direct access to your documents from inside the chat. connect Google Drive and you can say “find my content plan doc for april and suggest posts for next week”. Claude will find the file, read it and deliver the result. for those working with a team of content creators, this removes an entire layer of manual work: no need to download a file, upload it to the chat, explain context. Claude works with your data directly. setup takes a minute. settings -> connectors -> pick a service -> authorize.
9. peak hours and prompt editing - two tactical hacks
two quick tips worth knowing.
first: since march 2026 Claude burns through your limit faster during peak hours. peak is 5:00-11:00 AM pacific time, weekdays. for Europe that’s roughly 2:00-8:00 PM. the same request during peak hours costs more.
practical takeaway: heavy tasks (longreads, research, generating large volumes of content) are better done in the morning or late evening European time. light tasks whenever.
second: if Claude gave you something you didn’t want don’t write a new message saying “no, i meant something else”. hit edit on your original prompt and rephrase.
every new message in the chat exponentially increases the cost: message number 30 costs 31 times more than the first one, because Claude re-reads the entire conversation history. fix the prompt, don’t feed the history.
10. multimodality - drop screenshots, stop describing things in words
Claude can see images. this sounds obvious, but 90% of people keep describing in words what they could just show. instead of “on the BTC/USDT chart on the 4H timeframe there’s a double bottom forming with support at 62,400” just drop a screenshot of the chart and write “what do you see on this chart?”. Claude will break down patterns, levels, volumes.
this works for: Polymarket screenshots (position analysis, order book), TradingView charts, on-chain dashboard screens, PDF reports (drop the whole thing, Claude reads it), even thread screenshots if you need a quick translation or summary.
Claude processes multiple images in one message. you can drop 5 screenshots of different traders and ask for a comparative analysis all in one prompt. stop describing in words what you can show with an image. it’s faster, more accurate and saves tokens.
bottom line
Claude is not a chatbot where you throw prompts and hope for the best. it’s a work tool with a ton of settings that most people simply never opened.
open your settings. seriously.