When Politeness With AI Backfires

AI Doesn't Need Your Manners, it needs precise instructions and language

Author

Jason DePerro
April 24, 2025

In partnership with

This week Sam Altman mentioned that every "please" and "thank you" used in prompting is costing OpenAI tens of millions of dollars in tokens. While politeness seems a trivial nicety that costs nothing, the computational expense adds up dramatically at scale.

Researchers at Microsoft note that "When it clocks politeness, it’s more likely to be polite back. Generative AI also mirrors the levels of professionalism, clarity, and detail in the prompts you provide."

“Using polite language sets a tone for the response,” he explains. LLMs—large language models, a.k.a. generative AI—are trained on human conversations. In the same way that your email autocomplete suggests a likely next word or phrase, LLMs pick a sentence or paragraph it thinks you might want based on your input. Put another way, it’s a giant prediction machine making highly probabilistic guesses at what would plausibly come next. So when it clocks politeness, it’s more likely to be polite back. The same is true of your colleagues, strangers on the street, and the barista making your iced Americano: when you’re kind to them, they tend to be kind to you too. 

Microsoft’s Kurtis Beavers, Design Director Microsoft Copilot

Learning to communicate with precision is a required skill to thriving in an AI agent powered world. We will need to be able to succinctly give clear directions while also being respectful. These instructions will teach AI agents how we want them to communicate with us. So, we must find the right balance of clarity and kindness. We don’t want to be rude, nor do we want to be wishy-washy and imprecise.

Finding the right tone may sound like threading a verbal needle, but there are industries that have honed this craft for finding clarity and kindness.

Language Lessons from Life-or-Death Communication Systems

In domains where miscommunication costs lives, humans have developed specialized language systems that prioritize clarity and efficiency over social norms namely, air traffic controllers, military radio communication and medical professionals.

🛫 Air Traffic Control

Air traffic controllers use rigorously standardized phraseology to minimize ambiguity in time-critical scenarios. For example, the FlightGear ATC protocol mandates specific responses like "Wilco" (will comply) instead of informal acknowledgments, ensuring pilots and controllers share a common operational lexicon. Key principles include:

  • Elimination of superfluous words: Controllers avoid phrases like "please" or "could you," instead issuing direct instructions (e.g., "Delta 234, descend to 10,000 feet").

  • Mandatory readbacks: Pilots must verbatim repeat critical instructions (e.g., altitude changes) to confirm understanding.

  • Context-aware compression: Frequent terms like "holding point" or "vectors" replace longer explanations, leveraging shared situational awareness.

These protocols demonstrate how token conservation enhances both efficiency and safety—a model applicable to AI interactions where clarity and speed are prioritized.

🪖 Military Radio Communication

Military personnel use "prowords" that compress complex instructions into minimal syllables:

  • Clear Communication is Important: Just like the military uses specific words like "Wilco" and "Roger" to avoid confusion, AI systems need clear and consistent instructions to ensure they are understood correctly.

  • Brevity Drives Performance: Military radio operators keep messages concise without sacrificing meaning—AI agents similarly function best with direct, economical instructions that avoid unnecessary complexity or verbose explanations.

  • Confirmation Protocols: The military practice of reading back critical instructions has direct parallels in AI development; explicit confirmation steps help ensure mutual understanding between human and machine.

  • Structured Commands ≠ Disrespect: Military communication demonstrates how directness is not rudeness but efficiency—AI agents need instructions that are forthright and unambiguous while maintaining professional interaction patterns.

Military communication efficiency stems from universal training—everyone in the system understands the protocols and expectations. Similarly, developing shared vocabularies and interaction patterns between AI systems and their users creates a learned foundation for effective communication that outsiders may initially find opaque.

 🏥 Medical Lingo

Medical professionals use short-hand abbreviations to communicate complex procedures and lengthy medication names. These abbreviations can add simplicity where there is complexity to initiated professionals.

  • Standardized Terminology Saves Lives: Healthcare environments depend on universally understood abbreviations like "BP" and "NPO" to communicate efficiently in critical situations—AI agents similarly benefit from consistent, well-defined command structures that eliminate guesswork.

  • Dangerous Ambiguity Has Consequences: Healthcare's "Do Not Use" list of forbidden abbreviations demonstrates how similar-looking instructions like ""Q.D." (once daily) vs. "QID" (four times daily): Similar abbreviations risk dosing mistakes if poorly handwritten) can cause life-threatening errors—AI systems require similarly careful attention to preventing command ambiguity.

  • Context-Dependent Understanding: Medical shorthand varies across specialties yet maintains core consistency—AI interactions should similarly adapt to user contexts while preserving fundamental instruction patterns.

Medical communication reveals the tension between brevity and precision—AI agents face identical challenges where overly abbreviated instructions risk dangerous misinterpretations, while excessive verbosity slows critical processes.

A Framework for Precision Communication

Increasingly AI agents are capable of learning, retaining memory in the form of reference documents. You can see this many of popular AI tools:

  • Perplexity’s Spaces links, instructions and files allow associated for a perpetual research topic.

  • Windsurf to attach instructions and connections for it to recall throughout projects

  • Claude allows you to add personal communication preferences in settings

These forms of persistent information can apply to our communication patterns to create more efficient communication. Use this template of communication principles and add to it for your own.

Core Communication Principles

Communication Instructions:
principles for how humans and agents will correspond to accomplish tasks. It is a guide for efficient context window management. By committing to these principles tokens and server time will be efficiently utilized. It is to be added to the persistent communications settings in your favorite LLM, IDE or AI Agent.

User Prompting

  1. Clarity Over Niceties

    • Skip "please" or "could you." Use direct verbs like "Generate," "Debug," or "Summarize."

    • AI doesn’t care about politeness—save tokens and time.

  2. Structured Prompts

    • Use a consistent format: Action | Context | Parameters | Confirmation.

    • Reduces guesswork for AI and humans alike.

  3. Just Enough Context

    • Give the bare minimum background needed. Overloading wastes resources; under-explaining risks errors.

  4. Feedback Loops

    • Ask AI to confirm understanding or clarify ambiguity—like a pilot’s readback to a controller.

  5. Iterate as You Go

    • Treat prompting as a dialogue. Refine based on AI’s output for better results next time.

AI Responses

AI shouldn’t just obey—it should communicate. Build these into your workflow:

  1. Summarize First: AI restates the task (e.g., "I’ll design a login wireframe with social and biometric options").

  2. Ask If Unsure: AI flags ambiguity (e.g., "Social login priority over biometric?").

  3. Update Progress: For big tasks, AI checks in (e.g., "Layout done, adding auth now").

  4. Deliver Clearly: Final output matches your parameters, flagged as done.

This two-way clarity cuts frustration and rework.

The Precision Prompting Template

Craft prompts with this simple structure:

  • Action: What to do (e.g., "Design," "Analyze").

  • Context: Where it applies (e.g., "Mobile app for fitness").

  • Parameters: Specifics (e.g., "Include social login").

  • Confirmation: Verify understanding (e.g., "List what you’ll include").

Why It Works

  • Fast: Fewer words, less processing, lower costs.

  • Accurate: Structure kills ambiguity.

  • Flexible: Adapts to design or dev tasks.

  • Collaborative: Feels like a teammate, not a black box.

Building a Common AI Communication Language

As AI becomes more deeply integrated into our professional workflows, developing consistent communication patterns becomes increasingly valuable. Rather than awkwardly phrasing requests in overly formal language, professionals across all domains can benefit from purpose-built frameworks that respect both human cognitive patterns and AI processing capabilities.

By embracing structured prompting approaches, teams can dramatically improve their AI interactions while reducing friction and inconsistency. The future of human-AI collaboration lies not in learning to speak to machines like humans, but in developing shared protocols that leverage the strengths of both.

Learn AI in 5 minutes a day

This is the easiest way for a busy person wanting to learn AI in as little time as possible:

  1. Sign up for The Rundown AI newsletter

  2. They send you 5-minute email updates on the latest AI news and how to use it

  3. You learn how to become 2x more productive by leveraging AI

💬 Suggestion box

A newsletter exploring the principles that will help us design AI agent experiences and startups that amplify human creativity.

Subscribe to join a community of designers and developers shaping purposeful AI agents.

How'd we do this week?

Login or Subscribe to participate in polls.

New Referral Rewards

Until next time, keep innovating and stay curious!