Should We Stop Being Polite To Chatbots?

There was a moment this week that made me feel uncomfortable… and it’s a little embarrassing.

Sergey Brin (Google, Alphabet, etc.) casually remarked that large language models tend to perform better when users threaten them with physical violence.
He added, almost offhandedly, that people feel weird about this… so we don’t really talk about it.
I only thought about it when he said this… and I realized it would make me super uncomfortable to be rude… let alone violent… to a chatbot… and there’s many layers to unpack in that thought alone.

His line traveled fast… and predictably, it landed the wrong way.

Because the headline version sounds like this:
“Be rude to AI to get better answers.”
But that framing misses the real issue entirely.
The truth isn’t that AI models respond better to hostility.
It’s that we keep insisting on treating tools like beings… and then feel morally weird when the illusion cracks.

The real headline is that modern AI systems don’t respond better to politeness or rudeness.

They respond better to clarity, precision and urgency.
Older models were more fragile.
They relied on conversational scaffolding (remember having to start every request with something like “you are an expert marketer with world-class skills in copywriting…”).
Politeness reduced ambiguity and smoothed intent.

The newer models don’t need that.

They are increasingly tone-resistant.
They strip sentiment and focus on task execution, not emotional wrapping.
When you add fluff… whether it’s excessive courtesy or theatrical aggression… you are introducing linguistic noise.

As a reminder…

The system doesn’t feel threatened.
It doesn’t feel respected.
It doesn’t feel anything.
It parses… for now… and who knows what it will “feel” if we ever cross that threshold.
So when people say “threatening the model works,” what they’re actually doing is this:
They’re removing ambiguity.
They’re injecting urgency.
They’re narrowing the scope.
They’re being unmistakably clear about stakes and expectations.

Not emotional… operational.

Which brings us to the real cultural tension.
We keep anthropomorphizing systems that don’t have inner lives.
We say “please.”
We say “thank you.”
We apologize to chatbots.
We worry about being rude (this was me).
Not because the machine cares… but because we’re practicing being human in the presence of something that “sounds” human (further proof of The Intimacy Economy at work).

Being polite to a machine doesn’t make it perform better.

But it does train us to blur the line between tool and companion.
And being aggressive to a machine doesn’t unlock intelligence…
But it does reveal how quickly we project power dynamics onto software.

Which makes me believe that the danger isn’t rudeness… it’s confusion.

Confusion about agency.
Confusion about responsibility.
Confusion about what kind of relationship we’re actually in.
The optimal way to work with AI isn’t hostile or obsequious.

It’s precise… so try this the next time…

Strip the filler.
Add context.
Use imperatives.
State why the task matters.
Be direct.
Not decorative.

And maybe that’s the deeper lesson hiding in Brin’s comment?

This isn’t about threatening machines.
It’s about stopping the performance of politeness where it doesn’t belong…
and redirecting that human energy back where it does…

Toward people… and how we speak to and treat each other.

Because if we start calibrating our emotional behavior around tools instead of humans…
We don’t get better answers.
We get worse relationships, because we get comfortable getting results through behaviors we’d never use face-to-face.

And that’s a cost no prompt optimization can fix.

This is what Elias Makos and I discussed on CJAD 800 AM.

Before you go… ThinkersOne is a new way for organizations to buy bite-sized and personalized thought leadership video content (live and recorded) from the best Thinkers in the world. If you’re looking to add excitement and big smarts to your meetings, corporate events, company off-sites, “lunch & learns” and beyond, check it out.