Politeness vs Performance – Why Saying Please may Killing Your AI’s Accuracy - The Health AI Brief

Summary3 min read

Episode Overview: "Politeness vs Performance – Why Saying Please May Be Killing Your AI’s Accuracy"

Purpose:
This episode of The Health AI Brief, hosted by Stephen A, focuses on a counterintuitive but vital concept for clinicians integrating artificial intelligence (AI) into healthcare workflows: how commonly-used polite language can impair the accuracy and efficiency of large language models (LLMs) when processing clinical data. The episode delivers essential practical advice for medical professionals on how to communicate with AI for optimal performance.

Key Discussion Points and Insights

1. The Politeness Problem in AI Prompts

In human teams—especially in high-pressure clinical settings—politeness (“please,” “thank you”) helps teams function smoothly.
But: When these habits carry over into medical AI interactions, they can degrade the AI’s effectiveness.

“But when we transition to using large language models, that social habit becomes a technical hindrance potentially. Does being nice to your AI actually make it less effective at processing clinical data? The short answer is yes."
— Stephen A (00:23)

2. Token Optimization: The Technical Challenge

Every word—including polite phrases—is converted into a “token” in AI models.
LLMs have a fixed context window (a limited “working memory”), so every unnecessary word eats into valuable processing space.
Metaphor: Adding polite phrases is like crowding out patient details in a trauma radio alert.

“You’re paying a token tax. You're literally crowding out the space available for the patient's actual clinical details.”
— Stephen A (00:41)

“You don’t say hello, hope you’re having a lovely morning. Would you mind awfully preparing for a patient? You say, trauma alert, 30 year old male penetrating chest wound. ETA five minutes.”
— Stephen A (00:51)

3. How AI Interprets Prompts

AI is not a person—it's a mathematical engine.
AI follows instructions best when given direct, imperative commands.

"The AI is a mathematical engine, not a person. It follows instructions best when they're framed as direct imperative commands."
— Stephen A (01:03)

Practical Takeaways for Clinicians

1. Use Imperative Language

Start prompts with clear, direct verbs: “summarize,” “extract,” “analyze.”
These verbs tell the AI exactly what to focus on.

2. Cut the Fluff

Remove all social or polite filler.
Every token saved can be used for more clinically relevant content.

3. Be Precise, Not Nice

Bluntness will not offend the AI.
Clear, direct prompts enhance the signal-to-noise ratio for more accurate outputs.

“Every token saved on politeness is a token earned for clinical accuracy.”
— Stephen A (01:23)

“The AI won’t be offended if you’re blunt. In fact, it will follow your instructions more accurately because the signal to noise ratio is higher with clear direct instructions.”
— Stephen A (01:28)

Notable Quotes & Memorable Moments

On the core message:
“Does being nice to your AI actually make it less effective at processing clinical data? The short answer is yes.” (00:25)
On direct communication:
“Direct verbs signal the core task that the AI’s attention mechanism should focus on.” (01:13)
On optimizing prompts:
“Remove all social filler. Every token saved on politeness is a token earned for clinical accuracy.” (01:20)

Important Segments & Timestamps

| Timestamp | Segment | |-----------|-----------------------------------------------------| | 00:00 | Introduction: Clinical politeness vs. AI interaction| | 00:23 | The technical hindrance of polite prompts | | 00:41 | Explanation of token optimization and “token tax” | | 00:51 | Trauma alert metaphor for concise communication | | 01:03 | Imperative commands and AI as a mathematical engine | | 01:13 | Takeaway #1: Use imperative verbs | | 01:20 | Takeaway #2: Cut the fluff | | 01:23 | Takeaway #3: Be precise, not nice |

Summary

This concise episode reframes an everyday habit—politeness—as a technical pitfall when prompting clinical AI tools. By replacing social courtesies with clear, direct language, clinicians can maximize the effectiveness of LLMs and ensure more clinically precise and accurate outputs.
Bottom line: Instruct AI like you’re relaying a trauma alert—not chatting at a team huddle. Every unnecessary word reduces clinical performance, so be clear, be direct, and let the AI do what it does best.

Loading summary

Transcript1 lines

[00:01]
A
Welcome to the Health AI Brief. Breaking down the AI Shaping Our World One Concept at a Time we're trained to be polite in a high pressure, multidisciplinary team meeting or on the ward. Please and thank you are the social glue that keeps the team functioning. But when we transition to using large language models, that social habit becomes a technical hindrance potentially. Does being nice to your AI actually make it less effective at processing clinical data? The short answer is yes. It comes down to token optimization. Every word that you include, including please, if you don't mind, and thank you, is converted into a token. As we've discussed, large language models have a fixed context window, a limited amount of space they can keep in the working memory. When you fill that space with social pleasantries, you're paying a token tax. You're literally crowding out the space available for the patient's actual clinical details. Think of it like a radio transmission to the trauma team during a pre alert. You don't say hello, hope you're having a lovely morning. Would you mind awfully preparing for a patient? You say, trauma alert, 30 year old male penetrating chest wound. ETA five minutes. The AI is a mathematical engine, not a person. It follows instructions best when they're framed as direct imperative commands. So takeaways for your next prompt are to use imperatives, start with summarize, extract, or analyze. Direct verbs signal the core task that the AI's attention mechanism should focus on. Two is cut the fluff, remove all social filler. Every token saved on politeness is a token earned for clinical accuracy, and three is be precise, not nice. The AI won't be offended if you're blunt. In fact, it will follow your instructions more accurately because the signal to noise ratio is higher with clear direct instructions. So that's politeness versus performance in a nutshell. If you found that useful, don't forget to hit subscribe so that you don't miss future episodes on this Similar topics.