Dialogue

Every line of dialogue should do at least two things from this list: advance plot, reveal character, develop a relationship, establish tone, or carry subtext. A line that does only one thing is probably weak. A line that advances plot and reveals character in the same breath is doing its job. When you audit dialogue and find lines that exist only to convey information, those are the lines to cut or rewrite.

This rule cuts both ways. Dialogue that’s all subtext and atmosphere, with nothing happening, stalls the story. Dialogue that’s all plot delivery, with no character texture, reads like a screenplay outline. The two-function test is a minimum, not a ceiling. The best dialogue scenes — Pinter, Mamet, McCarthy — are doing five or six things at once, and the density of work is why they’re unrepeatable.

The test is simple to apply: underline each line of dialogue and write what it’s doing. If you can only write one answer, the line probably needs revision.

The eye skips "said" and "asked." That invisibility is a feature, not a limitation — it means you can use attribution without interrupting the reader’s attention. The moment you swap "said" for a said-bookism — "she exclaimed," "he opined," "she breathed," "he hissed" — the word becomes visible. The reader notices the word instead of the line.

Elmore Leonard’s rule: never use a verb other than "said" to carry dialogue. Leonard’s fiction is exhibit A for how much work that constraint doesn’t prevent — his dialogue scenes are among the most alive in American crime fiction, and they’re almost entirely "said."

Adverbs on attribution have the same problem. "'I hate you,' she said angrily" is the writer not trusting the dialogue to carry its own weight. If "I hate you" needs the adverb "angrily" to clarify that it’s delivered with anger, the line itself isn’t working. Anger in dialogue comes from word choice, rhythm, and context, not from attribution descriptors. Strip the adverbs from attribution and see whether the dialogue still works. If it doesn’t, the problem is the dialogue, not the missing adverb.

The exceptions are real but narrow: "whispered" and "shouted" occasionally work because they specify volume in a way that the word choice alone may not convey. Use them sparingly enough that they retain their function.

This is where dialogue earns most of its power. Harold Pinter built a career entirely on the gap between what characters say and what they mean. In real conversations, people almost never say what they mean directly — especially about things that actually matter to them. They approach, deflect, imply, test, retreat.

Two characters talking about whether they should sell the house are often talking about whether their marriage is over. The house is the text. The marriage is the subtext. The writer’s job is to engineer the surface conversation so that the underlying one is completely legible without being stated. When it works, the reader simultaneously processes both conversations — and the gap between them is where the emotional charge lives.

The technique is architectural: the subtext conversation has to be coherent on its own terms, not just a vague atmosphere of significance. If you stripped the surface conversation away, could you reconstruct the subtext from the scene alone? If not, the subtext isn’t anchored enough to carry the scene.

Pinter’s Betrayal (1978) reverses the temporal order of a love affair and unfolds across conversations where everything that matters is in the negative space — what isn’t said, what isn’t asked, what neither party will name. The play is about the lies people maintain with each other and themselves, and almost none of that is stated. The craft is in constructing dialogue that sounds like small talk and reads as self-betrayal.

See Subtext for the full treatment of how this operates at the thematic level. For the scene-level mechanics of conversations where practical talk carries emotional weight, see Displacement Activity Intimacy.

Put "he said" or "she asked" after the first sentence of a speech, not after the third or fourth. Readers want to know who’s talking as soon as possible. Making them read three sentences before orienting them is a small tax on attention that compounds over a long scene.

In rapid back-and-forth exchanges between two characters, attribution can disappear entirely once the rhythm is established. But re-anchor periodically — readers lose track faster than you expect, and a reader who has to stop and recount speeches to establish who said what has been pulled out of the scene completely.

Action beats placed between speeches function as attribution: "He poured another glass. / 'I’m not going to apologize.'" The beat identifies the speaker without needing a tag. The action beat has the added advantage of adding character information — what the person does between speeches tells us something the speech alone may not. See Blocking and Physical Choreography in Prose for the full treatment of action beats in dialogue scenes.

Representing dialect phonetically — "Ah wuz goin' down to the stoah" — almost always creates problems. It slows reading. It risks condescension. It calls constant attention to itself. The better approach is to suggest dialect and regional voice through vocabulary, rhythm, idiom, and syntax.

The character who says "I don’t care for that none" or "She was after asking me to leave" is signaling dialect through grammar and idiom rather than phonetic respelling. The reader hears it without having to decode it. Irvine Welsh’s Trainspotting (1993) is the most prominent exception — the phonetic transcription of Edinburgh working-class dialect is the point of the novel, not incidental to it — but Welsh earns the difficulty through sustained commitment and because the voice itself is the argument the novel is making. Most dialect writing is not doing this work and shouldn’t attempt the full Welsh treatment.

The reliable principle: make dialect legible to the reader without making the reader work to decode it. Vocabulary and rhythm carry the identity; phonetic spelling carries the difficulty. Choose the parts of dialect that create recognition rather than obstacle.

Dialogue scenes breathe. Generous paragraph breaks — new paragraph for each speaker, action beats treated as their own paragraphs — create visual pace that pulls readers forward. Dense blocks of dialogue without action beats or white space feel breathless and hard to track.

This is a pacing tool as much as a readability tool. Short exchanges, short paragraphs, minimal action beats — fast pace. Longer speeches, embedded action, interiority between exchanges — slower, more deliberate. The writer controls the reader’s experienced tempo through the visual rhythm of the page as much as through the content.

Action beats — a character pours coffee, looks away, picks up a pen — serve double duty. They break up the visual density of dialogue and do character work. What a character does while talking is often more revealing than what they say. The character who straightens papers while discussing infidelity is saying something. The character who stops moving is saying something different. Physical behavior during dialogue is never neutral; the writer’s job is to make it non-neutral intentionally rather than accidentally.

Every character should sound different in dialogue. Not as a technical exercise — readers don’t need to be able to cover the attribution and identify the speaker — but because characters who speak in the same rhythms and vocabularies are, at some level, the same person.

Character Voice is the full article, but the practical principle for dialogue: before writing a scene, know what each character’s conversational signatures are. Their characteristic sentence length. Their preferred vocabulary register. Whether they deflect or press. Whether they answer questions directly or with other questions. What they say when they mean the opposite. What they never say, even when it’s what they mean.

These signatures should emerge from character, not from the writer’s desire for variety. A character who grew up in a technical profession speaks technically; a character who processes emotionally speaks emotionally. The differentiation is character-true before it’s dialogue-crafted.

Read dialogue aloud. If it sounds like writing — if you’d never hear it said — rewrite it. The test is whether it sounds spoken while remaining purposeful. Those two conditions together are harder than either alone.

Hemingway’s dialogue in The Sun Also Rises (1926) and Hills Like White Elephants (1927) remains the benchmark for this: characters who sound entirely like people while carrying enormous weight in the gaps. The iceberg principle applies: what’s said is the surface fraction. The rest is below, legible to the careful reader without being stated. "The hills look like white elephants." "I’ve never seen one." "No, you wouldn’t have." The subtext of that entire story is encoded in how people discuss a landscape.

The practical test: does every line pass the spoken-aloud test? Not theatrical delivery, just natural speech. Lines that stick in the mouth when spoken are lines that stick in the reader’s comprehension when read. Smooth dialogue reading aloud is the minimum; the maximum is dialogue that can only exist in fiction — compressed, weighted, subtext-laden — while still sounding like something a person would actually say.