Eavesdropping: Where Real Dialogue Comes From

Nobody speaks the way screenplays think they speak. Read a bad script aloud and you will hear it instantly, that stiff, complete, grammatical way characters explain themselves in full sentences that no living person has ever produced under pressure. The cure is not a dialogue textbook. It is a train compartment out of Ahmedabad, a chai stall at seven in the morning, and the discipline to shut up and listen to how human beings actually talk.

This is the ear half of the eye discipline I laid out in seeing like a filmmaker. Observation is not only visual. The way people speak is a behaviour like any other, and eavesdropping, done as craft and not gossip, is how you learn to write a line that sounds like it came out of a mouth and not a manual.

Real speech is broken, and that is the point

Listen to a real conversation and it is a mess. People interrupt themselves. They abandon a sentence halfway and start another. They repeat, they trail off, they answer a question you did not hear with a shrug and the word "anyway." They talk past each other, both waiting to speak, neither quite listening. That mess is not a flaw to be cleaned up. It is the sound of truth, and it is precisely what wooden dialogue lacks. New writers polish their dialogue until every line is a neat little speech, and in doing so they sand off the exact roughness that makes it sound alive.

If it sounds like writing, I rewrite it.Elmore Leonard

That is the whole test. If a line sounds composed, like something a writer would type, it is dead. If it sounds dropped, half-finished, a little wrong, the way real people speak, it lives. And the only reliable way to learn the difference is to sit among real talkers with your mouth closed and your ear open.

What to actually listen for

Do not try to transcribe whole conversations, you cannot and you should not. Listen instead for the specific music of it. The filler words a particular kind of person leans on. The way two colleagues on a train say almost nothing and communicate everything through pauses. The exact phrase a chai vendor uses that you would never have invented. The thing left unsaid that hangs in the air heavier than anything spoken. Real dialogue is mostly subtext, people talking about the tea while meaning something about the marriage, which is the whole idea behind show, don't tell applied to speech.

People rarely say what they mean. Write the gap, not the meaning.

Catch the strange true fragment and put it straight in the notebook. Not the whole exchange, the one line that could not have come from anywhere but a real throat. Those fragments are the seasoning. A single overheard line can unlock a character's entire voice, because you suddenly hear how they would say everything else.

Do not transcribe, distil

Here is the mistake that follows a good ear. You catch a brilliant real conversation, drop it into your scene word for word, and it falls flat. Why? Because real speech is ninety percent dull. In life we tolerate the dead stretches because we are living inside them. On the page there is no room. So the craft is not transcription, it is distillation. You take the rhythm, the fillers, the interruptions, the strangeness of real talk, and you compress it, keeping the flavour of chaos while cutting the actual chaos down to something a scene can carry. It should feel overheard. It cannot actually be overheard, unedited, or it drowns the story.

This is close cousin to how you steal any detail from the world, which I wrote about in stealing from real life ethically. Take the sound, reshape the words, make it serve the character rather than the stranger who first said it. The craft archives at No Film School keep hammering the same lesson: dialogue that works is real speech that has been carved, not copied.

The ear the machine does not have

There is a reason AI-written dialogue so often sounds slightly off, like a very fluent non-native speaker who has read every phrasebook and heard no real conversations. The machine learned speech from text, from written dialogue that was already once removed from how people actually talk. It has never sat in a train compartment. It has no ear, only a memory of other people's transcriptions. So it produces lines that are grammatically perfect and humanly wrong, complete where real speech breaks, tidy where real speech spills. That gap is exactly the territory I mapped in whether AI replaces screenwriters, and dialogue is where the seam shows most.

Your advantage is the chai stall the machine can never visit. The more you listen to real, broken, contradictory human speech, the further your dialogue drifts from the generated average, toward something that sounds like a specific person and not like the internet's idea of a person. That specificity is also empathy in disguise, because to catch how someone truly speaks you have to have listened closely enough to feel who they are, which loops straight back to empathy on the page.

Sharpen the ear this week

Give yourself one assignment. On your next train, bus, or wait at a stall, do not scroll. Just listen for one line that is so real you could never have written it, and get it into your notebook before the stop. Do that a dozen times and your inner ear recalibrates. You will start hearing the fakeness in your own drafts, the lines that sound like writing, and you will know exactly how to break them until they sound like people. The wider skill of catching what others miss is its own discipline, which I wrote about in noticing what everyone else ignores.