A listener cannot go back

One difference explains all the others

A reader controls the pace. They can stop, re-read the clause they lost, look at the previous sentence, skim ahead. A listener can do none of these things without an act of will: finding the scrub bar, dragging it back, missing the next bit while they do it. In practice they lose the sentence and keep going, slightly less attached than before.

Everything below follows from that one asymmetry. Writing for the ear is writing that can be parsed in one pass, in order, with no backtracking.

The nested clause is the main offender

“The lathe, which had been sitting under a tarpaulin in the back of the shop since the previous owner retired in 1998, still runs.”

On paper this is fine. Spoken, the listener holds “the lathe” in memory for nineteen words while a subordinate clause runs, and by the time “still runs” arrives they have lost the subject and have to reconstruct it. They do reconstruct it, mostly, and it costs them the first half of your next sentence.

Unpacked:

“The lathe still runs. It had been under a tarpaulin at the back of the shop since 1998, when the previous owner retired.”

Same facts, same length, no held-open structure. The rule that generates this: finish each grammatical thought before starting the next one. Subject, verb, object, stop.

Numbers have to be rounded to be heard

“Thirty-one point four per cent” is five distinct pieces of information arriving in two seconds. Almost nobody retains it. “Nearly a third” is one piece, retained by everybody, and in a spoken piece it is more accurate in the only sense that matters: it is what the listener will end up believing.

Put the precise figure on screen if precision matters. The ear gets the shape, the eye gets the decimal.

The same applies to lists. Three items is the practical limit in speech. Beyond that a listener stops tracking and waits for the end.

Read it aloud, and time it

This is the whole quality-control process and it is not optional. Read the script aloud, standing up, at the pace you would actually deliver it. Three things happen:

1. You run out of breath somewhere, and that is where the sentence is too long. Mark it and break it.
2. You stumble over a phrase, and a phrase you stumble over while reading your own writing is one a listener will stumble over too. Almost always it is a consonant cluster or an accidental rhyme.
3. You find out the actual duration. Clear narration runs at roughly 140 to 160 words a minute, slower than conversation, because a listener has no facial cues and no ability to ask. A ninety-second piece is therefore about 210 to 240 words of narration, and that is far less than people expect. If your script is 400 words you are not making a ninety-second film, you are making a two-and-a-half-minute one, and you should decide that deliberately.

Write the contractions in

People say “it’s”, “we’d”, “didn’t”. A script that says “it is”, “we would”, “did not” will either be read stiffly or silently contracted by the reader, and the second means your timing is wrong. Write what will be said.

Similarly, write the pauses. A full stop is a pause; a paragraph break is a longer one. If you need a specific beat, put it on its own line. Narrators follow layout more than punctuation.

Homophones and the things only the ear hears

Spoken text has failure modes that never occur in print. “Their”, “there” and “they’re” are identical. So are “our” and “hour” in many accents. “A grade A” is ambiguous aloud. Accidental rhymes and alliteration that look inert on the page become conspicuous when spoken, and a listener who notices the sound has stopped following the sense.

The free test for all of this is a text-to-speech read-back: every phone and operating system has one built in, and hearing your script in a flat synthetic voice is unusually good at exposing the sentences that only work because you know what they mean.

The limitation: ear-writing is not always better writing

This style (short sentences, front-loaded subjects, rounded numbers) is worse for some purposes. It cannot hold a qualified argument, because qualification is exactly what subordinate clauses are for. A legal position, a technical caveat, a claim that is true only under three conditions: these become misleading when flattened, because the flattening removes the conditions.

When you need precision that will not survive the ear, the answer is not to write a more complicated script. It is to put the precise version on screen or in the description and let the narration carry the shape. Two channels, each doing what it is good at, rather than one channel doing both badly.