Claude Tips | Boost Answer Quality with Extended Thinking, Projects, and Files

3 Basics That People Underusing Claude Tend to Overlook

Top-ranking explanatory articles on SERP (reference: SHIFT AI / Samurai Engineer) commonly point to these three things:

1. Assign a role at the start — Simply writing "You are an experienced editor" at the beginning changes both the writing style and perspective
2. Declare the output format upfront — Specifying a format like "in 3 bullet points" or "as a Markdown table" eliminates post-processing entirely
3. Re-explaining context every time — Using Projects eliminates this problem at its root

With these three points in mind, proceed to the tips below.

Tip 1 — Use Projects to Maintain Context and Eliminate Repetitive Explanations

Claude.ai's Projects feature is a dedicated workspace that bundles custom instructions, conversation history, and files together. If you find yourself repeatedly explaining "Our company is a SaaS business…" every session, setting up a Project will dramatically reduce your input effort.

Setup steps:

1. Click "Projects" in the Claude.ai sidebar
2. Enter a project name and description under "New project"
3. Fill in roles, tone, restrictions, and other details in "Project instructions"
4. Upload reference files (specifications, brand guides, past data)
5. From then on, simply have conversations within this project — instructions and files are automatically referenced every time

SHIFT AI's testing reports that "using Projects improves answer quality even as questions get shorter" (source: https://shift-ai.co.jp/blog/11518/). Since the custom instruction limit runs to several thousand characters, you can paste in an entire style guide, FAQ, or prohibited expressions list without issue.

Use case examples:

• A marketer sets up brand tone and competitor lists as standing instructions
• A developer keeps coding standards and testing policies in Projects
• A writer registers proofreading rules and reader personas

Tip 2 — Use Extended Thinking to Tackle Difficult Problems Head-On

With Claude Sonnet 4.6 and above, you can use Extended Thinking mode, which produces answers after a longer internal reasoning process than standard responses. The more complex the question, the more noticeable the effect.

How to enable it:

1. Toggle on "Extended thinking" at the bottom left of the Claude.ai input field
2. Or write "Take your time before answering" or "Think step by step" in the input field
3. When thinking mode is active, a gray thinking block appears before the response

Situations where it's effective:

• Complex math or logic problems
• Consultations involving legal or ethical judgment
• When you want to compare multiple options from multiple angles
• Code bug analysis (adding "Consider all possible root causes" makes it go deeper)

The Claude Code usage guide on maku.blog reports that "giving Claude the instruction 'Think harder' is effective for deeper reasoning" (source: https://maku.blog/p/qj9wydo/). The same principle applies to regular conversations on claude.ai. Extended Thinking takes longer to respond, but the more important the decision, the more worth it.

Tip 3 — Use Files, PDFs, and Images as Input to Cut Work Time

Claude can accept PDFs, Word documents, Excel files, PNGs, JPGs, and more as direct input. Just saying "Summarize this PDF" can process hundreds of pages of material in a short time.

Effective use cases:

1. PDF summarization: Attach a contract or paper and ask "Give me a 3-point summary and flag any clauses to watch out for"
2. Excel/CSV analysis: Paste a data file and ask "Point out monthly sales trends and outliers"
3. Text extraction from images: Attach a screenshot and ask "List all text in this UI"
4. Document comparison: Attach two files at once and ask "Tell me the differences and what changed"

The key is "pairing a specific question with the file attachment." Simply saying "Please take a look" will only return a generic summary. Writing out "who wants to know what, in what format" makes a big difference in accuracy.

Attaching multiple files at once and asking Claude to "compare these comprehensively" is also effective. Claude reads relationships between files and organizes the information accordingly.

Tip 4 — Specify Output Format for Ready-to-Paste Results

Claude uses Markdown by default, but simply specifying a format suited to your purpose eliminates all post-processing.

Samurai Engineer's testing reports that "specifying the output format in advance significantly reduced post-processing costs" (source: https://generative-ai.sejuku.net/blog/5423/). In particular, getting responses back as CSV or JSON means you can paste them directly into a system or other tool.

The most reliable placement for format instructions is either at the beginning of the prompt or as the last line. Embedding them in the middle of the text makes them easy to miss.

Tip 5 — Iterative Improvement — Design Conversations Without Trying to Get It Right in One Shot

Trying to get a perfect answer from the first prompt makes the prompt too complex and actually reduces accuracy. The professional approach is "narrowing down step by step."

Recommended flow:

1. Confirm direction ("Give me an overview of ○○ and three approaches")
2. Choose your preferred approach and dig deeper ("Tell me the details for approach B")
3. Refine specific parts ("Shorten the second paragraph and add one example")
4. Do a final check ("Review the whole thing and fix any contradictions")

Using these 4 steps produces higher quality than a single one-shot prompt, and makes it clear where things went off track. Since Claude retains context within the same conversation, instructions like "revise based on the previous draft" work well.

If a conversation starts to feel long, copying a good response and pasting it into a new conversation with "Use the following as a starting point…" is also effective. It resets any misunderstandings that can accumulate from long context.

Tip 6 — Additional Tips for Claude Code Users

If you use Claude Code (the CLI tool), different tricks are useful compared to regular claude.ai (reference: maku.blog).

1. Interrupt with ESC: If you feel things are "going in the wrong direction" while it's working, press ESC to stop and give corrective instructions. Stopping mid-task is fine
2. Launch Plan mode with Shift+Tab: Before starting implementation, review and approve "how to proceed." The bigger the change, the safer it is to see the plan first
3. Set up completion notifications with Hooks: For long-running tasks, configure notifications for when they're done so you can use the waiting time for other work
4. Parallel work with Git worktree: When running multiple tasks simultaneously, use Git worktree to avoid conflicts

Tip 7 — Overlooked Useful Features in Claude.ai

The web UI and iOS/Android apps have features beyond prompting that improve usability.

• Voice conversation: The voice button on the smartphone app enables hands-free interaction. Can also be used for dictating meeting minutes
• Conversation sharing: Use the "Share" button at the top right of a conversation to generate a URL and share it with your team (paid plan)
• Custom Instructions (System Prompt): In Settings under "Custom Instructions," you can set a persona and tone that applies to all conversations
• Saving favorites: Star a good response to quickly reference it later

Custom Instructions are especially powerful when combined with Projects. Setting something globally like "I am a freelance writer and my readers are office workers in their 20s" means you never have to write it again.

5 Common Failure Patterns and Quick Fixes

Finally, here is a summary of failure examples and fixes that are commonly cited across multiple top-ranking articles.

All of these can be solved by the basic principle of "conveying constraints and format first." Rather than making prompts longer, narrowing the format tends to improve accuracy more.

Summary — Systematizing Practical Tips

Here is a review of the 8 tips introduced in this article.

1. Cover the 3 basics: role, output format, and context
2. Keep custom instructions and files on standby with Projects
3. Use Extended Thinking for complex questions
4. Actively use files, PDFs, and images as input
5. Declare output format for ready-to-paste results
6. Aim for iterative improvement rather than one-shot perfection
7. Use ESC, Plan mode, and Hooks with Claude Code
8. Solve failure patterns by leading with constraints and format

The shift from "using Claude vaguely" to "using Claude intentionally" doesn't require trying all these tips at once. Simply setting up Projects first, then making it a habit to specify output format, will noticeably improve the quality of your everyday Claude use.