02of tenWorking Notes on Prompting · What the Tool Is Actually Doing

Politeness Is a Confound

A working note on whether saying "please" to an AI does anything, and what's really going on.

There's a persistent folk belief, held just as firmly in both directions, that being polite to an AI either unlocks better answers or is completely pointless. Both camps are missing the same thing. Politeness isn't a lever, but it isn't nothing either: it's a confound.

The common move

The familiar belief "If you say please and thank you, the model tries harder." — vs. — "Politeness is useless, the model doesn't care."

People defend both with real conviction, often citing a time it seemed to work or seemed to make no difference. The question worth asking: what is the model actually responding to?

What you think it does, and what it actually does

What you think it does
  • Earns effort. "Please" feels like it motivates the model to try harder.
  • Is the cause. When a polite prompt gets a great answer, the politeness gets the credit.
  • Or: does nothing. The other camp concludes manners are pure waste.
What it actually does
  • No special mechanism. The model isn't flattered or motivated. A "please" is just a few more tokens in the context.
  • It's a confound. Polite requests tend to be fuller and more specific. The clarity does the work; the politeness rode along.
  • Not nothing, either. Because it travels with clarity, it correlates with better answers, which is why the myth survives.

Compare "structure this" with "Could you walk me through how to structure this, including the edge cases?" The second gets a better answer, and it's tempting to credit the courtesy. The real work was done by the specificity and context the polite phrasing happened to carry with it.

The more honest move

Aim at the thing that actually helps. The lever isn't tone; it's clarity, context, and specificity. Polite phrasing tends to smuggle those in, so isolate them and ask for them on purpose.

Specificity is the lever Walk me through how to structure this argument. Here's the context: [what it's for, who reads it]. I'm unsure about [the specific problem]. Cover the edge cases I'm likely to miss.

Notice there's no "please" doing the work here, and the request is strong anyway. Say "please" if you want to (there are good human reasons to keep the habit), but don't mistake it for the mechanism. The mechanism is telling the model what you actually want.

Try this

Next time a polite prompt gets a great answer, look again and ask: was it the manners, or the detail the manners carried?

Strip the courtesy in your head and check whether the request is still clear and specific. If it is, that's what worked.

Keep being polite if it suits you. Just spend your real effort on clarity, not tone.

The principle underneath

Politeness doesn't move the model; the clarity that usually travels with it does. Credit the specificity, not the manners, and aim your effort there.