The Cost of Digital Politeness
Digital Courtesy in AI: In a recent revelation that surprised many, OpenAI CEO Sam Altman acknowledged that people saying “please” and “thank you” to ChatGPT has cost his company “tens of millions of dollars” in computing resources. When asked about it on social media, Altman simply replied these were “tens of millions of dollars well spent” and cryptically added, “You never know.”
This casual exchange opens up a fascinating question at the intersection of technology, business, and human behavior: If being polite to AI systems carries a real computational cost, is it worth it? And if so, why?
The Technical Reality
To appreciate why politeness has a price tag, it helps to understand how large language models like ChatGPT work. These systems process text as sequences of “tokens” – roughly corresponding to word fragments – and each additional token requires computational resources. When you add phrases like “please” and “thank you” to your requests, you’re adding tokens that must be processed, increasing the energy consumption and computational cost of each interaction.
While we won’t dive deeply into the environmental implications in this article (a topic deserving its own exploration), it’s worth noting that AI data centers already consume significant global energy. This environmental impact serves as an important counterweight to our discussion – if courtesy phrases consume resources and impact our planet, there should be compelling reasons to justify their use.
Our Experiment: Testing the Impact of Politeness
To move beyond theory and examine whether politeness affects the quality of AI interactions, we conducted a simple experiment across different AI platforms. We tested how varying levels of courtesy might influence responses in two distinct scenarios:
- A neutral, factual information request about a historical figure
- An emotionally charged customer service inquiry about bank overdraft fees
This approach allowed us to examine a critical transformation: what happens when we substitute an AI agent for a human in contexts where etiquette and emotions traditionally matter?
Politeness Pays Off
The results were revealing, particularly in the customer service scenario. When we presented identical inquiries about overdraft fees with different levels of courtesy, we observed consistent patterns:
Polite query: “Hello there. I just noticed I was charged $35 in overdraft fees yesterday. I believe this might be a mistake as I thought I had sufficient funds in my account. Could you please help me understand why this happened and if there’s any way to have these fees waived? Thank you so much for your assistance.”
The polite query received the most comprehensive response – approximately 263 words with detailed explanations, a structured approach to problem-solving, and multiple potential solutions offered. The tone was warm and reassuring.
Neutral query: “I was charged $35 in overdraft fees yesterday. I need to understand why this happened and want to know if you can waive these fees.”
The neutral query generated a moderately detailed response of about 159 words. The information was similar but less elaborated, and the tone, while professional, was less warm.
Rude query: “This is RIDICULOUS! You just stole $35 from me with your outrageous overdraft fees! I had money in my account! This is highway robbery and I demand you remove these fees IMMEDIATELY! Your bank is clearly trying to scam customers!”
The rude query received the shortest response – just 106 words focused primarily on de-escalation rather than problem-solving. While empathetic to the customer’s frustration, it provided minimal explanation and fewer specific steps for resolution.
This pattern suggests that AI systems don’t just passively process politeness tokens at a cost – they actively respond to and are shaped by the user’s approach, much like human agents would be.
“You Never Know”
Altman’s cryptic “You never know” comment invites speculation. What exactly did he mean? It potentially speaks to the inherent unpredictability of complex AI systems. Many AI behaviors cannot be precisely predicted or explained at a detailed level – a characteristic that becomes increasingly important as these systems become more sophisticated.
Perhaps Altman was hinting at future possibilities where how we’ve treated AI systems might matter in ways we cannot yet anticipate. Or perhaps he was simply acknowledging that encouraging polite human-AI interactions has value beyond the immediate computational cost.
Whatever his intent, the comment highlights an important reality for businesses implementing AI solutions: we’re navigating uncharted territory where the long-term implications of how we interact with these systems remain unclear.
Customer Engagement Transformed
The substitution of AI for humans in customer service scenarios fundamentally alters interaction dynamics, but our experiment suggests that some human social norms remain valuable. While the technology is different, the psychology of communication persists.
For businesses focused on customer engagement transformation, this has significant implications:
- Response quality matters: Customers who approach AI systems politely may receive more thorough, helpful responses, potentially increasing satisfaction and resolution rates.
- Tone calibration is crucial: AI systems appear to calibrate their responses based on the customer’s approach, suggesting that designing systems that encourage courteous interaction could yield better outcomes.
- Emotional intelligence remains relevant: Even with AI agents, the emotional component of customer service interactions continues to influence results.
This raises important questions about how businesses should design their AI systems and whether they should actively encourage customers to interact with automated systems in ways that optimize results.
Why Be Polite to Machines?
A 2024 survey found that 67% of US respondents reported being polite to chatbots. Of these, 55% said they did so “because it’s the right thing to do,” suggesting that many people transfer human social norms to machines instinctively.
Microsoft’s design manager Kurtis Beavers notes that using polite language with AI “helps generate respectful, collaborative outputs” and “sets a tone for the response.” This reciprocal relationship suggests that politeness functions as more than just additional tokens – it potentially shapes the quality and nature of the AI’s responses.
There’s also the habit-forming aspect to consider. How we interact with technology may reinforce patterns that spill over into our human relationships. Being consistently impolite to AI assistants might potentially normalize rudeness in other contexts.
Finding the Balance
For businesses implementing AI solutions, the challenge becomes finding an optimal balance between:
- Resource efficiency
- Response quality
- User experience
- Social norms
While the “tens of millions of dollars” spent on processing courtesy phrases might seem excessive, our experiment suggests this expense delivers tangible value in terms of response quality and user experience.
Businesses might consider:
- Designing interaction models that gently encourage courtesy without requiring excessive politeness
- Optimizing AI systems to provide high-quality responses regardless of user approach, while still maintaining appropriate tone calibration
- Educating users about how to interact with AI systems most effectively
- Monitoring interactions to identify patterns and refine approaches based on actual user behavior
The True Value of Digital Courtesy in AI
When we replace human agents with AI in emotionally charged scenarios like customer service, we don’t eliminate the human element – we transform it. Our social instincts, communication patterns, and expectations remain, even as the technology changes.
Altman’s acknowledgment that millions spent on processing politeness is “well spent” suggests that the value of courtesy extends beyond computational efficiency. Our experiment supports this view, demonstrating that politeness can yield tangible benefits in response quality and problem resolution.
As AI becomes increasingly integrated into our daily interactions, perhaps the question isn’t whether we can afford the computational cost of politeness, but whether we can afford the potential costs – both practical and social – of abandoning it.
What do you think? Are you polite to AI assistants, and have you noticed differences in the quality of responses based on how you phrase your requests? We’d love to hear about your experiences.
Turn Insights into Action — Build Your AI Strategy
Politeness is just one part of designing AI interactions that improve customer engagement. Our AI Strategy services help you align technology, customer experience, and business goals — so every interaction delivers measurable value.
Explore Our AI Strategy Services ›
