The Art of AI Conversation

Common Sense Rules for Effective LLM Prompting

Introduction

The quiet revolution of large language models has transformed how we interact with artificial intelligence. Yet, a fundamental truth remains: the quality of their output depends heavily on the quality of our input. Over the years, I've collaborated with numerous professional services and legal firms to develop effective AI communication strategies. Through this work, I've discovered that successful approaches rarely rely on complicated frameworks. Instead, they embrace straightforward, common sense principles that treat the interaction as a genuine conversation.

The Conversation Paradigm

When we step back and consider what LLM prompting truly represents, we find it's fundamentally about clear communication with an intelligent entity that possesses both impressive capabilities and distinct limitations. The most effective approach treats prompting not as a programming exercise but as a conversation—a back-and-forth exchange where clarity, context, and iterative refinement lead to better outcomes.

Understanding this conversational dynamic helps users set realistic expectations and craft more effective requests. Just as we wouldn't walk up to a human expert and make vague, context-free demands, we shouldn't approach these sophisticated AI systems with poorly articulated requests and expect optimal results.

Core Principles of Effective Prompting

Be Specific and Clear

Vague requests invariably produce vague responses. I've witnessed this countless times across organizations—the executive who asks for "some thoughts on the market" receives a generic overview that offers little value, while the one who requests "an analysis of how recent regulatory changes in financial services affect mid-market lending institutions" receives targeted, actionable insights.

Specificity creates a clear target for the AI to aim at. This extends to all aspects of your request: length, tone, audience, format, and purpose. Consider the difference between "Write a blog post about marketing" and "Write a 1000-word blog post about content marketing strategies for B2B SaaS companies focusing on lead generation through educational content."

The first prompt leaves the AI to make countless assumptions about what you really want. The second provides a clear roadmap for producing relevant, targeted content. The difference in output quality is remarkable and consistent.

Provide Meaningful Context

LLMs cannot read your mind or understand your situation without explanation. They can't see your screen, access your files (beyond what you've shared), or recall previous projects unless you explicitly reference them. Including relevant background information dramatically improves output quality by giving the AI the context it needs to generate appropriate responses.

This became evident when working with a legal team initially receiving generic contract analyses from their LLM tool. When they began providing industry context, company background, and specific concerns in their prompts, the quality of insights improved dramatically. The AI wasn't suddenly smarter—it simply had the context needed to focus its capabilities more effectively.

Contextual elements include your audience and their level of expertise, the purpose of the content you're creating, industry-specific considerations, and references to previous work or conversations related to the current task. Think of it as briefing a new team member who's intelligent but unfamiliar with your specific situation.

Break Complex Tasks into Components

Even advanced LLMs can become overwhelmed by complex requests attempting to accomplish too much. Breaking larger projects into discrete steps produces better results and creates a more manageable workflow for both you and the AI.

This approach allows for more focused attention on each component, opportunities to course-correct between steps and better organization of the final output. I've seen this play out repeatedly in consulting environments, where attempting to generate comprehensive strategic plans in a single prompt produces disappointing results.

Instead, breaking the work into market analysis, target audience definition, channel strategy, content planning, and implementation timelines allows the AI to focus deeply on each element. The final product, assembled from these focused outputs, demonstrates greater coherence and quality than would be possible with a single, all-encompassing prompt.

Specify Format and Structure

LLMs can produce content in virtually any format, but they need explicit instructions to do so correctly. Being precise about structural elements ensures you receive immediately usable outputs rather than requiring extensive reformatting.

During a project with a management consulting firm, we discovered that simply adding format specifications to their prompts reduced post-processing time by over 60%. Their analysts went from requesting generic "market analyses" to specifying exactly how they wanted information presented—with executive summaries, clearly defined sections, specific data visualization suggestions, and appendices for methodological details.

Common format specifications include document structure (sections, headers, etc.), data presentation formats (tables, bullet points, etc.), technical formats (JSON, XML, markdown, etc.), and length constraints for individual sections. The more precise you can be about these elements, the closer the AI's output will match your expectations.

Real-World Examples: Bad Prompts vs. Good Prompts

Legal Research

When we examine specific examples, the difference between effective and ineffective prompting becomes crystal clear. Consider a scenario where a legal professional needs to research employment law cases.

A typical but ineffective prompt might simply state: "Tell me about employment law cases."

This approach leaves too much undefined. What jurisdiction? What time period? What specific aspects of employment law? What level of detail is needed?

A more effective prompt demonstrates how specificity transforms results: "Find key federal court decisions from the past 5 years addressing whether remote work is a reasonable accommodation under the ADA. Focus on cases from the 9th Circuit."

This revised prompt works because it specifies a clear timeframe (past 5 years), a specific legal question (remote work as ADA accommodation), a jurisdiction (federal courts, specifically the 9th Circuit), and implicitly communicates the level of detail needed (key decisions rather than comprehensive analysis). The output becomes immediately more valuable because the request was crafted with intentionality and clarity.

Contract Review

Legal professionals regularly need to evaluate contracts for specific issues. A vague request like "Check if this NDA is good" fails to provide the necessary parameters for meaningful analysis.

Contrast this with a well-crafted prompt: "Review this NDA for enforceability under California law, specifically analyzing: (1) the scope of confidential information, (2) the duration of obligations, and (3) whether the non-solicitation provisions comply with Section 16600 of the California Business Code."

This prompt succeeds because it specifies the governing law (California), identifies the exact aspects of the agreement requiring scrutiny, references specific legal code provisions, and implicitly establishes the perspective from which to analyze. The result is a focused, legally sound analysis rather than a generic evaluation.

Document Drafting

When creating legal documents, the quality of prompt engineering directly affects the quality and utility of the output. A request like "Write a cease and desist letter" is far too broad to produce a helpful draft.

A more effective approach would be: "Draft a cease and desist letter regarding trademark infringement of our client's mark 'TechPro' by competitor 'TechPros Solutions.' Include: (1) specific instances of infringement, (2) demands to cease use within 14 days, and (3) reference relevant sections of the Lanham Act."

This prompt provides the specific parties involved, the exact legal issue at stake, the required components of the letter, and references to relevant law. The difference in utility between these approaches is immense—one produces a generic template requiring substantial revision, while the other creates a nearly client-ready document tailored to the specific legal situation.

Common Pitfalls to Avoid

Ambiguity and Vagueness

One of the most prevalent issues I observe across organizations is the tendency toward ambiguous requests that force the LLM to make assumptions. These assumptions are often misaligned with expectations, leading to frustration and wasted effort.

Consider the request to "write a good marketing email." What constitutes "good" in this context? Is it conversion rate, open rate, brand alignment, or something else entirely? Without this clarification, the AI must guess at your success criteria. Similarly, phrases like "interesting content" or "high-quality analysis" without further specification create a definitional vacuum that the AI must fill—often incorrectly.

I worked with one professional services firm that consistently received disappointing content until they began explicitly defining their quality criteria in each prompt. Replacing vague qualifiers with specific requirements dramatically improved their results without changing the underlying AI system.

Overloading the Prompt

Just as human conversations become confusing when we attempt to cover too many topics at once, LLM interactions suffer when prompts become overloaded with requirements. The signs of prompt overload include multiple unrelated tasks, excessive qualifiers and conditions, contradictory requirements, and lengthy, convoluted instructions.

A financial advisory team I consulted with initially struggled with this problem. Their prompts resembled complex legal documents, attempting to address every possible angle and contingency in a single request. The result was muddled, unfocused responses that addressed everything superficially and nothing thoroughly.

We restructured their approach to focus on one clear objective per prompt, with appropriate context and constraints. This simple shift—breaking complex requests into focused conversations—improved their results dramatically without requiring any technical changes to their LLM implementation.

Neglecting to Provide Examples

One of the most powerful yet underutilized techniques in effective prompting is the inclusion of examples. Examples dramatically improve an LLM's understanding of your expectations by providing a concrete reference point rather than relying solely on abstract instructions.

Even a brief example of the desired output format can significantly enhance results. Examples are particularly valuable for unusual or specialized content formats, industry-specific terminology usage, desired tone and style, and complex analytical approaches.

A legal firm I worked with struggled to get their LLM tool to generate client advice in their preferred format until they began including snippets of previous, anonymized documents as examples. The improvement was immediate and substantial—the AI now had a clear pattern to follow rather than attempting to infer the desired format from verbal descriptions alone.

Advanced Techniques for Professional Users

Role and Perspective Assignment

More sophisticated techniques can enhance your results as you become more comfortable with basic prompting principles. One particularly effective approach involves assigning a specific role or perspective to the LLM.

For instance, rather than simply requesting an analysis of marketing data, you might frame it as: "Analyze this marketing data as if you were a senior marketing executive with expertise in consumer behavior and digital acquisition channels. Focus particularly on identifying trends that might not be immediately obvious but could indicate emerging opportunities."

This technique works because it provides the AI with a conceptual framework for organizing its knowledge and capabilities. It's not that the AI literally "becomes" a marketing executive, but rather that this framing helps it prioritize and structure relevant information in a way that aligns with your needs.

Effective roles include industry experts with specific credentials, representatives of particular stakeholder groups, or process-oriented roles like critic, facilitator, or analyst. The key is choosing a role that naturally aligns with the type of output you're seeking.

Multi-step Reasoning Prompts

For complex analytical tasks, guiding the LLM through a structured thinking process often produces superior results. This approach creates a roadmap for the AI to follow, breaking down complex reasoning into discrete steps.

For example: "First, identify the key stakeholders in this project. Second, analyze their primary interests. Third, identify potential conflicts between these interests. Finally, suggest strategies to address these conflicts."

This technique is particularly valuable for complex analytical tasks, strategic problem-solving, risk identification and mitigation, and decision justification. By explicitly structuring the reasoning process, you help the AI organize its response logically and thoroughly rather than potentially missing important analytical steps.

I've seen this approach transform the quality of strategic analyses in management consulting settings. Where unstructured prompts often led to superficial SWOT analyses, multi-step reasoning prompts produced nuanced, thoughtful examinations of complex business challenges.

Implementation in Professional Settings

Creating Prompt Libraries

As organizations integrate LLMs into their workflows, developing standardized templates for common tasks becomes increasingly valuable. These prompt libraries serve as institutional knowledge, capturing effective approaches and making them available across teams.

One professional services firm I worked with created a comprehensive prompt library organized by task type, audience, and department. They maintained and refined these templates based on performance, creating a continuously improving resource that accelerated adoption and ensured consistent quality across the organization.

This approach is particularly valuable for organizations that repeatedly perform similar analyses or content creation tasks. Rather than having each team member reinvent effective prompts, a shared library leverages collective experience and best practices.

Training Team Members

Even with excellent prompt libraries, educating colleagues on effective prompting techniques remains essential. The organizations that have most successfully integrated LLMs into their workflows have invested in training programs that demystify these interactions and build practical skills.

Effective training programs typically include scenario-based exercises, clear guidelines for different use cases, and opportunities for practice with feedback. Some organizations have also established communities of practice where team members can share insights, innovations, and lessons learned from their LLM interactions.

The investment in training pays dividends in reduced frustration, more consistent results, and more creative applications of the technology across different business functions. As one executive told me, "We initially thought the technology would be the hard part. It turned out that helping our people communicate effectively with it was the real challenge—and the key to our success."

The Future of LLM Interaction

As LLM technology continues to evolve, we're already seeing more sophisticated interaction patterns emerge. Multimodal inputs are increasingly important, with systems that can process and respond to combinations of text, images, data visualizations, and other media types.

Future prompting will likely incorporate visual elements, diagrams, interactive components, and audio inputs. These developments will expand the possibilities for AI assistance while also creating new challenges for effective communication.

Nevertheless, the fundamental principles of clear prompting will remain relevant: specificity, context, appropriate structuring, and iterative refinement. The conversational paradigm—treating these interactions as thoughtful exchanges rather than programming exercises—will continue to offer the most productive framework for collaboration with these increasingly capable systems.

Conclusion

The art of effective LLM prompting isn't about mastering complex frameworks or technical jargon—it's about applying common sense principles of clear communication within a unique conversational context. By treating these interactions as thoughtful exchanges rather than programming exercises, professionals across industries can unlock the remarkable capabilities of these systems while avoiding their limitations.

As I've seen repeatedly across organizations, the difference between frustration and success often comes down to how we frame our requests. The teams that approach LLMs with clarity, specificity, and a conversational mindset consistently achieve better results than those who treat these systems as magical black boxes or simplistic command-line tools.

The good news is that these skills build naturally upon capabilities that most professionals already possess. Clear communication has always been valuable; LLM prompting simply applies these principles to a new context with its own unique characteristics.

As these technologies become increasingly embedded in professional workflows, mastering the art of effective prompting will emerge as a crucial skill—not just for technical specialists but for anyone who seeks to leverage these powerful tools to enhance their work. The organizations that invest in developing these capabilities today will be best positioned to benefit from the continuing evolution of AI technologies tomorrow.

Personal Notes

Two quick things. First - for those of you who don’t know - I spend most of my time working with companies doing AI training and workshops to help people really understand the practical use cases for these tools. I primarily with with professional services firms (law firms, PE firms, valuation firms, consulting firms) and have content that makes it really easy for companies to get started now and see real-world results. If you want to know more - please send me a note at steve@revopz.net or feel free to book some time on my calendar here:

Several people asked “Where’s Magnus?” in the last update. Here’s a few shots. He will be 19 weeks old this week and he’s already over 60 pounds. He’s so much fun and is super needy (every time I get up - he’s like “where are we going!”). Anyhow - here are a few pictures for those that like to see him.

Magnus