Anthropic releases method to 10× Claude Code / Opus 4.5
A comprehensive guide showcasing 10 concrete prompting techniques to maximize results from Claude AI models, based on official Anthropic recommendations.
Published Dec 10, 2025 by Greg Isenberg
Key Insights
Using friendly, clear, and firm language in prompts yields better results than vague requests or overly polite language.
Adding explicit constraints (length, style, audience, banned words) forces Claude to produce more creative and focused outputs.
Breaking complex tasks into multiple steps (outline → refine → execute) produces higher quality results than attempting to get everything in one prompt.
Structured output formats (tables, markdown, specific schemas) help organize information in a more useful and digestible way.
Using 'power phrases' like 'think step by step' and 'critique your own response' triggers more sophisticated reasoning from Claude.
Including the 'why' behind your request helps Claude understand context and generate more relevant responses.
Providing templates and examples as scaffolding guides Claude's output structure and prevents 'AI slop'.
0:56
Rule #1: Tone of collaboration
“The tone of collaboration is really important. You're going to want a friendly and clear and firm tone because that yields better results and more direct results.”
Greg explains that the tone used when prompting Claude significantly impacts the quality of responses. He contrasts vague, demanding requests ("fix this grammar in this now") with what he calls "architected briefs" - prompts that are direct but respectful and provide context.
The ideal approach treats Claude like a teammate rather than a servant, using clear language while maintaining a collaborative tone. This approach yields more direct and helpful responses rather than overly cautious or chatty outputs.
Takeaways
Friendly, clear, and firm prompting produces better results than vague demands
Treating AI as a collaborative teammate leads to more direct and useful responses
Overly demanding prompts can cause Claude to become defensive or cautious
Providing context in a respectful way helps Claude understand what you need
2:16
Rule #2: Principle of explicitness
“State your request as a clear action-oriented command with all the necessary details. Use action verbs, specify quantity, and define your target audience.”
The second rule focuses on making prompts explicit and detailed rather than vague. Greg demonstrates how a vague request like "I need a bunch of blog post ideas" leads to generic "AI slop," while an architected brief with clear parameters produces targeted results.
The key elements of explicit prompts include using action verbs (like "generate"), specifying exact quantities ("10 blog posts"), and defining the target audience ("city officials and real estate developers"). Each specific constraint adds a useful layer of direction.
Takeaways
Use action verbs rather than passive language in prompts
Specify exact quantities to get the right amount of information
Define your target audience to make the output relevant
Adding specific parameters prevents generic "AI slop" responses
3:20
Rule #3: Define the boundaries with clear constraints
“A well-defined box produces a more creative result than an empty field.”
Greg explains the counterintuitive idea that adding constraints actually increases creativity rather than limiting it. He contrasts a vague request like "write a short story about a detective in the future" with a highly constrained brief that specifies word count, style, character details, setting, and even words to avoid.
By creating boundaries around length, style, character, setting, and vocabulary, the AI is forced to be more creative within those parameters rather than defaulting to clichés. This approach takes more upfront effort but yields more original and focused results.
Takeaways
Adding constraints paradoxically increases creativity rather than limiting it
Specify length constraints (word count, paragraph limits) for appropriate sizing
Define style parameters by referencing specific authors or genres
Ban certain words to force more creative language choices
Creating boundaries prevents the AI from defaulting to clichés
4:26
Rule #4: Draft, plan, then act
“Don't try to get a perfect final product in one go. Working with the AI to create and then refine a plan or outline is a way more reliable path to a high-quality result.”
The fourth rule advocates for a multi-step approach rather than attempting to get perfect results in one prompt. Greg outlines a three-step process: first proposing an outline, then refining the plan with specific adjustments, and finally executing the full task based on the revised outline.
Although this process might seem more time-consuming initially, it actually saves time by reducing the need for multiple corrections later. By allowing for course correction early in the process, the final output becomes significantly more aligned with the user's intentions.
Takeaways
Break complex tasks into a sequence: outline → refine → execute
Start by asking for a plan or outline before requesting full execution
Refine the plan with specific adjustments before final execution
Early course correction prevents major revisions later
The multi-step approach actually saves time despite seeming longer
6:39
Rule #5: Demand structured output
“The AI is fluent in many formats beyond prose. Requesting specific formatting like markdown tables ensures information is organized exactly as needed.”
Greg demonstrates how requesting structured output formats dramatically improves the usability of Claude's responses. He contrasts a vague request ("List Apollo missions and some facts about them") that would yield unstructured paragraphs with a structured request for information presented in a specific format with defined columns and elements.
By specifying formats like markdown tables, bullet points, or other structured layouts, users can ensure information is presented in a digestible, scannable way. This approach makes Claude's outputs significantly more useful, especially when dealing with datasets or comparative information.
Takeaways
Request specific output formats like markdown tables for organized information
Specify exactly what elements/columns should be included in structured data
Structured formats make complex information more scannable and usable
Claude supports various output formats beyond simple prose
8:00
Rule #6: Explain the why behind your request
“Explaining the why behind an instruction helps the AI understand your true intent.”
The sixth rule emphasizes the importance of providing context about why you need something rather than simply requesting it. Greg illustrates this with an example about marketing slogans, showing how providing background information about brand values and target audience significantly improves relevance.
By explaining that coffee beans are "ethically sourced from small independent farms" and the target audience is "environmentally conscious millennials," Claude can generate far more targeted and appropriate slogans. This context helps the AI align with the underlying intent rather than just the surface-level request.
Takeaways
Include the purpose or context behind your request
Specify brand values, audience characteristics, or other relevant background
The more specific the target audience description, the better the results
Context helps Claude understand intent beyond the surface-level request
9:05
Rule #7: Control brevity vs. verbosity
“Explicitly command the AI to be more or less verbose to match your needs. You're in control of the output length.”
This section focuses on controlling the length and complexity of Claude's outputs. Greg demonstrates three different approaches: the "expert" prompt for detailed technical explanations, the "brief" prompt for concise bullet points, and the "simplifier" prompt for easy-to-understand explanations.
By using simple phrases like "explain in detail," "be concise," or "explain like I'm 5," users can calibrate Claude's responses to match their specific needs. This control over verbosity ensures outputs are neither overwhelming nor insufficiently detailed.
Takeaways
Use "explain in detail" or similar phrases for comprehensive expert responses
Request bullet points and conciseness for brief overviews
Use "explain like I'm X years old" to control complexity level
Different verbosity levels are appropriate for different contexts and uses
10:21
Rule #8: Provide a scaffold and templates
“Give Claude a template or example to guide its structure and style.”
In this section, Greg explains how providing templates or structural frameworks significantly improves Claude's outputs. Instead of vague requests like "summarize this article," he recommends providing explicit formatting instructions with placeholders that Claude can fill in.
The example shows how a summary request can be structured with specific sections for main thesis, key supporting points, and concluding insight. This rigid structure ensures the output is not only accurate but formatted exactly as needed, making it more useful and aligned with the user's expectations.
Takeaways
Provide explicit templates with placeholders for Claude to fill in
Define the exact structure you want in the final output
Templates ensure consistent formatting and inclusion of all required elements
Scaffolding reduces the chance of Claude adding unnecessary information
11:21
Rule #9: Use power phrases and expert personas
“Using advanced prompting terms can trigger more sophisticated modes of operation. These are like cheat codes.”
Greg reveals special "power phrases" that can trigger more sophisticated reasoning from Claude. These include "think step by step" to force detailed reasoning, "critique your own response" to encourage self-correction, and "adopt the persona of an expert" to prime the model for domain-specific knowledge.
These phrases leverage Claude's training on AI-related texts to activate specific behaviors that lead to higher-quality outputs. Greg describes these as "magic words" or "cheat codes" that can significantly enhance Claude's performance on complex tasks requiring careful reasoning or specialized knowledge.
Takeaways
"Think step by step" forces Claude to show its reasoning process
"Critique your own response" enables self-correction and improvement
"Adopt the persona of an expert in [field]" activates domain-specific knowledge
These phrases trigger more sophisticated reasoning capabilities in Claude
Power phrases are based on Claude's training on AI-related texts
12:28
Rule #10: Divide and conquer complex projects
“For a complex task, act as the conductor. Prompt for each part separately, then prompt for the synthesis.”
The final rule addresses how to handle complex projects that are too large for a single prompt. Greg recommends breaking tasks down into logical subtasks and managing them step-by-step, comparing this approach to project management.
Using a business plan example, he demonstrates a three-phase approach: first creating a blueprint (table of contents), then developing each section individually, and finally synthesizing everything with a review for consistency. This methodical approach ensures that each component receives appropriate attention before being integrated into the whole.
Takeaways
Break complex projects into logical subtasks rather than one massive prompt
Start with a blueprint or outline of the entire project
Develop each section individually with focused prompts
Synthesize the parts with a final integration and review step
This approach mirrors effective project management techniques
14:09
Putting it all together with an example
“Instead of saying 'Tell me about stoicism,' you're going to say 'Act as a university professor of philosophy. I'm preparing a 1-hour intro lecture for students with no prior knowledge.'”
In the final section, Greg demonstrates how to combine all ten rules into a comprehensive prompt. Using stoicism as an example, he transforms a vague request ("Tell me about stoicism") into a sophisticated prompt that incorporates persona, context, structure, constraints, and clear expectations.
The example incorporates an expert persona (university professor), explains the purpose (1-hour intro lecture), uses the divide and conquer approach (outline with three main sections), adds structure (nested bullet format), includes explicit requirements (key figures and ideas for each point), and specifies tone (accessible and engaging). This comprehensive example illustrates how the principles work together in practice.
Takeaways
Combine multiple prompting techniques for maximum effectiveness
Start with a persona that matches your needs (e.g., "university professor")
Include context about why you need the information
Specify structure, formatting, and organization requirements
Define constraints and expectations for style and tone
Conclusion
This video presents a systematic approach to prompt engineering specifically optimized for Anthropic's Claude AI models. While many users default to vague, demanding, or overly simplistic prompts, Greg demonstrates that thoughtfully constructed "architected briefs" consistently produce superior results.
The techniques shared aren't just abstract theories but practical implementations of Anthropic's own recommendations, gathered from their documentation, blog posts, and public communications. By incorporating elements like collaborative tone, explicit instructions, purposeful constraints, structured formats, and power phrases, users can transform Claude from a basic text generator into a sophisticated thinking partner.
So what? By investing slightly more effort in prompt construction, users can dramatically improve the quality, relevance, and usefulness of AI outputs while actually saving time that would otherwise be spent on multiple correction attempts. These techniques represent a shift from treating AI as a simple query-response tool to engaging with it as a collaborative partner in creative and analytical work. For professionals looking to integrate AI effectively into their workflows, mastering these prompting techniques creates a significant competitive advantage in producing high-quality, tailored outputs.