The best Claude tips and tricks are about setup and briefing, not secret commands. Write prompts the way you would brief a smart new colleague: give context, explain why, and show an example. Put long documents first and your question last. Then set Claude up once so every chat starts smarter: account instructions, a Project for each ongoing job, memory you review, and only the connectors you need. Use artifacts when you want something finished you can share, and keep approvals on whenever Claude can send, share or delete anything in your accounts.
This guide is for people who already know the basics and want better answers with less back and forth. Every feature below links to Anthropic’s own help or documentation pages, checked as of October 2026.
Related reading: how to use Claude AI for beginners, how to use Claude projects, and Claude artifacts. Key terms used in this guide: prompt injection and few-shot prompting.
The Claude tips that matter most
If you only have five minutes, start with this table. Each row is one habit and where to find it.
| Tip | Where it lives in Claude | Why it helps |
|---|---|---|
| Brief Claude like a new colleague | Any chat | Clear, specific requests get better first drafts |
| Put long files first, question last | Any chat with uploads | Anthropic reports better answers on long inputs |
| Write account-wide instructions | Settings, Instructions for Claude | Your preferences apply to every chat |
| Use a Project per ongoing job | Projects | Files and instructions stay in one place |
| Review your memory | Settings, Memory | You decide what Claude keeps |
| Ask for an artifact | Any chat or the Artifacts tab | You get a document, deck or tool to reuse |
| Connect only what you need | Customize, Connectors | Less access means less risk |
| Edit Docs, Sheets and Slides in place | Claude for Google Workspace (beta) | No copying text back and forth |
The rest of this guide explains each habit, the mistakes that waste time, and a simple framework for picking the right feature.
Prompting tips from Anthropic’s own guide
Anthropic publishes a prompting best practices guide for its current models. It is written for developers, but most of it works just as well when you type into the Claude app.
1. Brief Claude like a smart new employee
The guide compares Claude to a brilliant new hire who does not yet know your norms. Its golden rule: show your prompt to a colleague who knows nothing about the task. If they would be confused, Claude will be too.
Compare these two requests:
- Weak: “Write an email about the meeting.”
- Strong: “Write a short, friendly email to our garden club members. The monthly meeting moves from the library to the community hall on 4 November. Ask them to reply if they need a lift. Keep it under 120 words.”
The second one names the reader, the facts, the action and the length. If you want something ambitious, say so plainly, because Claude will not guess that you wanted extra effort.
2. Explain why, not only what
The same guide says that giving the reason behind an instruction helps Claude aim better. “Avoid jargon” is fine. “Avoid jargon, because my readers are retirees who are new to computers” is better, because Claude can apply that reason to choices you did not list.
3. Show three to five examples
Examples are one of the most reliable ways to steer tone and format. Anthropic suggests three to five varied examples and recommends wrapping them in simple tags such as <example> so Claude can tell them apart from your instructions. If you want product descriptions in your shop’s voice, paste three you already like and ask for the next one in the same style.
4. Put long documents first and your question last
For long inputs, the guide advises placing documents at the top and your question at the end. It notes that queries placed last improved answer quality by up to 30 percent in its tests on complex, multi-document inputs. A related trick: for a long report, ask Claude to pull the relevant quotes first and then answer based on those quotes. This keeps the answer tied to the text and makes it easy for you to check.
5. Say what you want, not what to avoid
“Don’t use bullet points” often works less well than “Write in short, flowing paragraphs.” The guide also notes that the style of your prompt can rub off on the reply. If you write your request in plain paragraphs, you tend to get fewer lists back.
6. Use tags when a prompt mixes several things
When one message holds instructions, background and a pasted document, wrap each part in its own tag, such as <instructions>, <context> and <document>. This sounds technical, but it is just labeling. It stops Claude from treating your pasted notes as instructions.
Set Claude up once so every chat starts smarter
Prompting tips help one chat at a time. These settings help every chat.
7. Write your Instructions for Claude
Claude’s personalization guide describes three layers. Instructions for Claude apply to all your chats and sit under Settings. Project instructions apply only inside one project. Skills add specific behaviors, including tone and format, that you turn on when you need them.
A good account instruction is short and practical. For example: “I live in the UK and use metric units. I prefer plain English and short answers unless I ask for detail. When you are unsure, say so.” That one paragraph saves you from repeating yourself in every chat.
If you remember a Styles menu, note that as of October 2026 the personalization guide no longer lists styles. For reusable tone and format, it points to skills, which are folders of instructions Claude loads only when a task needs them. Skills are available on all plans, but they need code execution switched on.
8. Give each ongoing job its own Project
Projects are self-contained workspaces with their own chats, files and instructions. Free accounts can create up to five. On paid plans, when a project’s knowledge gets large, Claude switches to a retrieval mode that the help page says can expand capacity by up to ten times.
Practical tips:
- Make one project per real job, such as “Family history book” or “Shop product listings,” not one per topic you are curious about.
- Keep project files current. Delete old drafts so Claude does not quote a version you replaced.
- Put the role and the rules in the project instructions, and keep the account instruction for things that are true everywhere.
9. Check what memory has saved
According to Claude’s memory and chat search guide, memory is on by default for Free, Pro and Max users, and Claude saves it as separate topics while you chat. You can read, edit or delete each topic under Settings, then Memory. You can also turn memory off for a single chat before the first message, or use an incognito chat that is not saved to history or memory. Each project keeps its own memory, separate from your other chats.
Paid plans also get chat search, which lets Claude look through your earlier conversations. This is handy when you ask “what did we decide about the budget?” and the answer sits in an older chat.
10. Decide your training setting on purpose
On consumer plans, Claude’s privacy settings page explains a toggle under Settings, then Privacy, about helping improve Anthropic’s models. If you turn it off, new chats and coding sessions are not used for future model training. Make this choice deliberately rather than leaving whatever you clicked at sign-up.
Work smarter with files, artifacts and your apps
11. Upload the right file, the right way
Claude’s file upload guide lists PDF, DOCX, CSV, TXT, XLSX and common image formats among supported types. In a chat you can attach up to 20 files, each up to 500MB, and PDFs up to 1,000 pages. Files added to a project have a lower limit of 30MB each.
Three tricks from the same page are easy to miss:
- Claude reads charts and images inside PDFs of 100 pages or fewer. Longer PDFs are read as text only, so split a long report if the charts matter.
- Word files and other non-PDF documents are read as text only. If a diagram matters, export to PDF first.
- When you ask about a page, use the page number your PDF viewer shows, not the number printed on the page.
12. Ask for an artifact when you want a finished thing
The artifacts guide describes an artifact as something Claude makes that you would put in front of someone: a document, slide deck, design, dashboard or small tool. It opens beside the chat, and every artifact is saved in the Artifacts tab. As of October 2026 you can start from templates such as Docs and Slides, and decks export to PowerPoint, PDF or Google Slides.
Two details save frustration. Artifacts need the code execution and file creation setting turned on under Settings, then Capabilities. And artifacts start private, so check who can see one before you share a link. If an artifact stores data, the guide warns you to check whether it uses shared storage before entering anything sensitive.
13. Connect apps with the least access that works
Connectors let Claude search and act inside services like Google Drive, Slack or Linear. You add them under Customize, then Connectors, and switch them on per chat. Claude can only reach what your own account can reach, and Anthropic’s advice is to connect only services you trust and need.
For the Google Workspace connectors, Claude can do more than read. It can send Gmail messages, change Calendar events and share or trash Drive files. Sending and sharing ask for your approval by default. Keep it that way, and keep approvals on for anything that sends, deletes, moves money or changes filters. One more risk: when Claude reads emails or web pages written by other people, hidden text in them can try to give Claude instructions. This is called prompt injection. Approvals are your safety net, so read each action before you allow it.
14. Edit Google Docs, Sheets and Slides in place
Claude for Google Workspace is a beta sidebar for Pro, Max, Team and Enterprise plans as of October 2026. You install it from the Google Workspace Marketplace, then open it from the Extensions menu. It can rewrite long documents, write and fix spreadsheet formulas, and turn notes into slides, with edits saved in version history.
Good habits here:
- Leave the default “Ask before edits” mode on while you learn how it behaves.
- The add-on asks for access to the single file it is opened in. If any add-on, from anyone, asks to read all your Drive files or your Gmail when you only need one document, decline it.
- The help page says the beta should not be used with protected health information. Keep medical records, ID numbers and card details out of it, and out of every AI chat.
15. Pick the model and plan for the job
As of October 2026, Anthropic’s models overview lists Claude Fable 5.1, Claude Opus 5.5, Claude Sonnet 5.5 and Claude Haiku 5.5 as current models. It suggests Opus 5.5 as the starting point for most work, Fable 5.1 for the most demanding reasoning, Sonnet 5.5 for a balance of speed and intelligence, and Haiku 5.5 for fast, simple tasks. The pricing page lists Opus access under Pro, so the models you can pick in the app depend on your plan.
On cost, the Claude pricing page lists a Free plan, and Pro at $20 per month billed monthly or $17 per month billed annually, as of October 2026. Verify current pricing on the official site before you buy. Pro adds more usage, Research, Claude Code and Opus models.
16. Keep the age rule in mind for family use
Claude’s minimum age page says every user must be at least 18 to create and use an account. If you are a parent or grandparent, use Claude to prepare your own material, such as a quiz you print out, rather than handing the account to a child or teen.
What to know before deciding: common mistakes and fixes
Most frustration with Claude comes from a handful of habits. Here are the common mistakes and the fix for each.
- Starting a new chat for every question about the same job. Context is lost each time. Fix: use a Project so files and instructions carry over.
- One endless chat for everything. Very long chats mix old and new instructions. Fix: start a fresh chat inside the same project when the task changes.
- Trusting a summary without checking it. Claude can misread a table or miss a page. Fix: ask for quotes with page numbers, then spot check two or three.
- Pasting sensitive data. Fix: never paste passwords, ID numbers, card numbers, health records or children’s details. Replace names with roles such as “Client A.”
- Turning on every connector. Fix: connect one service, test it on a low stakes task, then add the next. Keep approvals on.
- Fighting the format. Fix: describe the format you want, give one example, and save it in your instructions so you do not repeat it.
- Assuming features never change. Claude’s features move fast. Help pages change too: the personalization guide no longer mentions styles. Fix: when a menu looks different, check the help center before assuming something is broken.
Decision framework: which Claude feature fits the job
Use these questions in order. The first “yes” tells you where to start.
- Will I come back to this job again? Use a Project with instructions and files.
- Do I want something to hand to another person? Ask for an artifact, such as a doc or deck.
- Is the work already in a Google Doc, Sheet or Slides file? Use Claude for Google Workspace and keep “Ask before edits” on.
- Does Claude need to look up my email, calendar or files? Add one connector, with approvals on.
- Do I want the same tone or format every time? Put it in your instructions or a skill.
- Is it a quick one-off question? A plain chat is enough. Brief it well and move on.
A hypothetical worked example. Imagine Maria, 58, who runs a small online shop selling handmade candles. She creates a project called “Shop listings” and adds her ten best product descriptions as examples. Her project instruction says: “Write in a warm, plain voice. Mention burn time and scent notes. Keep each listing under 100 words, because buyers read on phones.” First she uploads a supplier price list as a PDF and asks Claude to quote the rows that changed before writing anything. Next she asks for an artifact: a one-page price sheet she can share with a stockist. Then she opens her stock spreadsheet in Google Sheets, asks the sidebar to fix a broken formula, and approves the change after reading it. She never connects her email, because she does not need to. Each step uses one feature for one clear job.
Next step
Pick one job you do every week and try three habits on it: a better brief, a Project with instructions, and one artifact you can reuse. Keep notes on what changed in the results, and adjust your instructions after a few uses.
If you would rather learn these skills step by step, a guided route helps. Our lessons are short and bite-sized, so you can take them at your own pace, with practice on popular AI assistants. Explore Coursiv AI lessons and build the habit one small lesson at a time.
Related guides: Claude memory and Claude skills.