Back to the table

30 Intermediate Cv Character Voice Character

Character Voice

Each character sounds like nobody else.

What it is

Character voice is the sound of one person. The sound of your whole book is narrative voice, the personality of the telling, and the craft of writing a conversation is dialogue. This page has a narrower and harder job: making sure Character A's lines could never pass for Character B's, and keeping that true for nine books. Narrative voice hands individual people over to this page. Dialogue covers tags, beats, compression and the whole argument about how to write accents.

Readers run a simple test without knowing they're doing it. Cover the dialogue tags, and see if you can still tell who's talking. Most drafts fail that test, and they fail it in a particular way. You can tell the characters apart by content (this one keeps mentioning the money, that one keeps mentioning the children) but not by manner. If you swap their lines, nothing sounds wrong. When that happens, the whole cast is speaking the author's own idiolect, meaning the author's personal way of talking, and only their opinions differ.

Voice sounds mystical until you look at one line of speech, and then it turns into a short list of decisions you're already making, usually by accident.

Here's the cover-the-tags test done properly. Strip every tag and beat out of a scene, so all that's left is the bare lines, one after another. Then come back to it tomorrow. If you can't tell whose line is whose, the differences are in your head and not on the page.

It's the best check you have, and it has three limits, and they matter because writers overcorrect against it. It rewards oddity, because the fastest way to pass is to give somebody a strange habit. A cast of five people with five strange habits reads like a novelty act, and passing by way of tics is worse than failing without them. Some characters really do sound alike, and forcing them apart hurts the book. Think of two junior officers who joined in the same year, two sisters raised in the same house, or two consultants in the same specialty. These people should share a register, meaning the same level of formality and the same way of putting things. Telling them apart is thinner, subtler work than separating a duchess from a dockworker. The page has other tools for it: who speaks first, who gets called by name, whose hands are busy, and which one the viewpoint character is watching. What you need is clarity, and a distinct voice is only one way to get it. And it only tests dialogue, so a book can pass the tags test and still have four narrators who think in identical prose.

The whole page rests on one claim. A voice you can't justify from the character's life is a costume, and readers feel the seams even when they can't name the problem. Here's the test. For every distinctive feature of a voice, answer where did that come from in one sentence, using that character's own history. The nurse's flat euphemisms came from twenty years of telling families things. The politeness that shows up when the landlord is angry came from a childhood of managing somebody. If your answer is "because I needed him to sound different from his brother", you've got a costume, and the character will slip back the moment you stop concentrating. So backstory, wound and flaw come before voice rather than sitting beside it. You don't choose a voice and then invent a past to support it. You know the past, and the voice falls out of it.

For any viewpoint character, voice shapes the narration as well as the speech. That goes for first person, close third, or one of the viewpoints in a multi-POV novel (a book told from several points of view). Voice decides the rhythm of the sentences, which details get reported, what gets named and what gets stepped around, and where the narration makes a joke. The two elements on either side of this page overlap here, so it helps to draw the line once. In close modes, the book's narrative voice passes through the character, and the two partly merge. What belongs on this page is a warning. A book with four viewpoint characters needs four narrating voices, not four sets of opinions written in one prose style. When readers complain that a multi-POV novel "blurred together", they're almost never complaining about the plot. The page on point of view covers how far voice and narration merge, and at what distance.

Why it matters

Voice is how readers keep a large cast in their heads. A reader tracking twenty named characters isn't holding twenty life stories in mind. They're holding twenty sounds. Get it right and you can leave a character offstage for a hundred pages, then bring them back without introducing them again. Get it wrong and readers flip back to check who somebody is, which is the most expensive thing they can do.

Voice saves you narration you'd otherwise have to write. You can tell a reader that a man is evasive, or you can let him answer two questions with questions. The second way takes fewer words, can't be skimmed, and isn't telling.

A returning character is made of voice. By book six, much of the reader's pleasure in a series regular is recognition. They already know her, and what they're waiting for is to hear her sound like herself. Voice is a series asset and a series liability at the same time. It also decides how closely an audiobook performance matches what you meant, because a narrator will make those choices whether you supplied them or not.

Voice helps you produce books. Writers who turn out several books a year usually type no faster than anyone else. They have fewer decisions left to make. Write a voice down (three words this person uses, two they refuse, their usual sentence shape, what they do when cornered) and it takes hundreds of tiny decisions out of every chapter, so the first draft moves.

Drift bites hardest when you write a lot of books. Book seven is written by a different person from book one. You're three years older and more skilled, you're bored with your own habits, and you can no longer hear what the early books sounded like. Each new book sounds only slightly unlike the last, so you never notice. Readers going through the series back to back notice right away, and they say the character has been ruined, without seeing it as a craft problem. The fix is dull and reliable. Write the voice down while you can still hear it, and reread it before you start the next book.

Real-world examples

David Milch's Deadwood (2004 to 2006): a cast separated by register

Deadwood ran for three seasons on HBO, from March 21, 2004 to August 27, 2006. David Milch created it. Before he was a showrunner (the person in charge of a TV series), he was an English lecturer at Yale, and he had worked with Robert Penn Warren and Cleanth Brooks on literature textbooks. The show is the most useful worked example there is of a large cast told apart by register rather than by accent. Register here means how formal and how fancy a person's speech is.

Matt Feeney reviewed the first season for Slate on May 21, 2004. He described characters uttering long, serpentine sentences in diction that, depending on the speaker, could ascend to courtly abstraction or sink to the ripest vulgarity. That "depending on the speaker" is the whole craft point. The newspaperman Merrick is courtly to the edge of absurdity, and Doc Cochran overstates as a professional tactic. Al Swearengen, the saloon owner, plays an elaborate word game in which the swearing and the formality do the same work at once. None of these people is marked by a dropped consonant. What marks each of them is how much structure they build around a sentence before they get to the point.

Milch has explained where both layers came from. In his memoir Life's Work (2022), he describes the extreme language as one of the few alternatives to law in a camp that had none. He also notes that characters with book learning (the Bible, Shakespeare, Dickens) spoke in an almost Elizabethan way, which set them against the less educated. And he describes deciding against period-accurate profanity, because to a modern ear nineteenth-century blasphemy sounds quaint, and it doesn't shock.

So what can you take from this? Register (how formal, how ornate, how far a sentence winds before it reaches the verb) tells a cast apart more reliably than phonetic spelling, and it costs the reader nothing. Each register here is paid for, because it comes from what that character has read, who they need to impress, and what they're protecting.

Barbara Kingsolver's The Poisonwood Bible (1998): five narrators from one house

Kingsolver's novel came out from Harper in 1998, and it was a Pulitzer finalist and an Oprah's Book Club pick. It's narrated in the first person by five members of one family: Orleanna Price and her four daughters, who arrive in the Belgian Congo in 1959. Rachel is fifteen, the twins Leah and Adah are fourteen, and Ruth May is five.

Kingsolver took on the hardest version of the problem. The girls have the same parents, the same Georgia upbringing, the same house and the same church, and three of them are within a year of each other. Yet every chapter has to announce whose it is within a sentence or two, with no tags available at all.

She solves it with the parts listed earlier, not with quirks. Rachel's voice is built on malapropism, which means using a wrong word that sounds like the right one. She reaches for a grand word and gets a neighboring one, and that tells you her ambition and her limits at once. Adah has been partly paralyzed down one side since birth (hemiplegic), and for most of the novel she's silent by choice. She reads and writes palindromes and keeps up an ironic commentary nobody in the book can hear. Leah is earnest, certain, and speaks in plain statements. Ruth May's sentences are short, and her attention goes where a five-year-old's would. Orleanna narrates from decades later, which gives her a tense and a weariness nobody else has.

Adah is the most useful character on this page, because she barely speaks aloud and has the strongest voice in the book. Whatever character voice is, it isn't dialogue. It's the shape of a mind on the page.

Her method is on record, and that's what you can use. Kingsolver's own FAQ, quoted by the Open College of the Arts, describes spending nearly a year "getting the hang of the Price girls". She chose one practice scene and wrote it in every voice in turn, the same scene over and over, until she had each daughter's rhythm and verbal instincts. That year went on the voices alone, before the novel itself could work.

Jane Austen's Pride and Prejudice (1813): two sisters who ought to sound alike

Pride and Prejudice was published on January 28, 1813 by T. Egerton. It passes the cover-the-tags test with a cast of a dozen, even though every speaker comes from a narrow slice of one society and nobody is marked by regional speech. It's the best free demonstration there is, and the full text is a click away.

Start with the hard pair: Jane and Elizabeth Bennet, the two eldest sisters. They were raised in the same house by the same parents, they're close in age and equally well spoken, and they confide in each other. They should sound the same, and they don't, because the difference is a refusal. Jane won't think badly of anybody, so her sentences hedge, qualify and look for the kindest reading. When she can't avoid a judgment, she reaches for a softer word. Elizabeth judges instantly, in balanced sentences with the punch at the end, and she enjoys doing it. Neither has a catchphrase. The whole difference is in what each one lets herself say.

The technique is easier to see in the rest of the cast. Mr. Bennet speaks in short, dry ironies and stops. Mrs. Bennet exclaims, repeats herself and comes back to her nerves. Lydia runs on, and Mary produces improving generalities, little bits of moral advice. Mr. Collins, the clergyman cousin, has the heaviest voice in the novel. He's pompous and roundabout, and he can't make a short statement. He's also the clearest case of a voice fully paid for by a life. He's socially insecure, he depends on a patroness, and his job requires him to be agreeable, so every sentence he speaks is that situation in miniature. The voice is what he had to become, much more than a joke Austen is telling about him.

Here's something you can use in your own work. Austen matches how loud a voice is to how important the character is. The two women the book is really about have the subtlest voices in it, and the broad ones belong to minor characters.

Sue Grafton's Kinsey Millhone novels (1982 to 2017): one voice held for twenty-five books

Sue Grafton published "A" Is for Alibi in 1982 and "Y" Is for Yesterday in 2017: twenty-five novels narrated in the first person by the Santa Teresa private investigator Kinsey Millhone. Grafton died on December 28, 2017 without writing Z, and her family said plainly that the alphabet would end at Y, with no ghostwriter. That's thirty-five years and one voice, with no help.

Most of the work came from a pair of decisions about how the series was built, and you can copy both.

The first was freezing the world. Every book is set in the 1980s, each one only weeks or months after the last, so Kinsey ages just a few years across twenty-five novels. Grafton was clear about why. There were no cellphones and no internet. As she put it to CrimeReads in August 2017, the joy of it from her point of view is that Kinsey "has to do her sleuthing the old-fashioned way". What that does for the voice is the interesting part. A narrator whose tools, town, wardrobe and routines never change has almost nothing to drift into. Voice drift usually sneaks in inside world drift (new technology, a new city, a new job), and Grafton took away the way in.

The second was a written rule she could check. In the same interview, she describes deciding to frame each novel as a written report from Kinsey, after reading a real report that closed with "respectfully submitted". Each book ends with that signature, and she says readers complain when she leaves it out, which makes her straighten up her act. The voice rule is written down in a form that's impossible to forget or fudge, and an audience enforces it.

You won't have readers policing your voice in book two. You can do the rest: a fixed frame, a short written list of what this narrator does and doesn't do, and a handful of sentences copied out of book one to use as a tuning fork. That belongs in the character bible, and it's the one practice most likely to make book seven sound like book one.

Dickens's David Copperfield (1849 to 1850): the verbal tic, what Dickens gets away with, and what you probably cannot

The most common wrong turn here is using a catchphrase in place of a voice, and writers keep making it because they've seen it work in Dickens. In David Copperfield (1849 to 1850), Uriah Heep keeps insisting that he's "umble", Wilkins Micawber gets to his point by way of "in short", and Mrs. Micawber declares that she will never desert Mr. Micawber. All three are unforgettable.

They're also the textbook case of a flat character. In Aspects of the Novel (1927), E.M. Forster used Mrs. Micawber as his example. A really flat character, he said, can be expressed in a single sentence, and hers is that she will never desert Mr. Micawber. She says she won't, she doesn't, and there she is. Forster wasn't sneering. Flat characters are useful, memorable and easy for a reader to carry. The trouble is that a catchphrase produces one, so if you give your protagonist a catchphrase, you flatten your protagonist.

A couple of distinctions will save you a lot of grief. A marker isn't a tic. Across Terry Pratchett's Discworld novels, from 1983 to 2015, Death's speech is set in small capitals with no quotation marks. He's a skeleton with no vocal cords, so his words arrive in your head without going through your ears. The capitals make a fact about what he is visible on the page, which is very different from a habit standing in for a character. Markers that encode something true about the speaker last through decades of books. Habits bolted on just so readers can identify someone don't. And scale it to importance, as Austen does. A doorman who appears twice can have a catchphrase and be better for it, because the reader files him away instantly. Your viewpoint character can't, because the reader will spend three hundred pages inside a repetition they start counting.

An everyday scenario: covering the tags on a four-hander

Here's the test done properly. Four people are in a hospital relatives' room, waiting for news of the same woman, and every tag and beat has been stripped out.

"There's a machine on the second floor does a proper tea. This one only does the powdered." "Before anyone agrees to anything, I'd like to be clear on what we're being told and what we're being asked to sign. Those are two separate conversations." "Is she awake? Can I go in?" "Right, well, if she does go, I'm having the good chair." "Tomás." "Sorry. Sorry. I'll get the teas." "There's no point going in if she's not awake." "I'll ask someone. I'll just ask."

Four people, no tags, and you can still tell whose line is whose. The mother's sister turns everything into catering and won't say the word. The son is a lawyer and can't stop splitting a question into its parts. The daughter asks short, direct questions and wants to be doing something. The son-in-law says the unsayable, hears his name said back to him flatly, and apologizes twice.

Notice what's doing the work. There are no accents, no dialect spelling and no catchphrases, and nobody's vocabulary is exotic. The differences are sentence length, grammatical habit, what each person is paying attention to, whether they joke, and what each one refuses to name. Every one can be justified from a life: thirty years in a hospital canteen, a career in contracts, a job that means managing thirty children at once, and a man who has never known what to do with a silence.

Compare that with the version most drafts produce, where all four say some version of I'm worried about Mum in slightly different words, and the scene needs its tags just to make sense.

Try it yourself

Take four characters from your current book, ideally including two who are supposed to be similar, and set aside about two hours.

  1. Run the tags test cold. Copy a scene with several characters into a blank document and delete every tag, beat and name. Leave it overnight, and tomorrow, try to work out whose line is whose. Score yourself truthfully, because this is the only step that tells you whether you have a problem.
  2. Do Kingsolver's drill. Pick one ordinary scene (somebody is late, a bill has arrived, a phone is ringing) and write it from each character in turn, half a page each. Write their sentences, not their opinions. Then do the same scene again the next day. It's the core exercise, and it's where voices actually get built.
  3. Fill in a voice card for each character. Give it five minutes each, using the card below.
  4. Find each one's refusal. For every character, write down one thing they won't say and what they say instead. If you can only do this for your protagonist, you've found your problem.
  5. Justify every feature. Go down each card, and beside each item write where it came from in that person's life. Anything you can't trace in one sentence is a costume, so either find the history that earns it or cut it.
  6. Work the hard pair on purpose. Take the two characters who share a background, and instead of giving either of them a quirk, find one real difference. Maybe one gives in faster, one names feelings while the other names facts, or one asks questions while the other makes statements. Write a page of them arguing, then check whether the page needs its tags.
  7. Read it aloud in four voices. Stand up and do it out loud, and you'll hear right away which two characters your mouth is doing the same way.
  8. Set a drift check. Copy six to ten lines of each series regular's best dialogue into the character bible under their name. Read them before you start the next book. It takes five minutes, and it catches the slide that readers notice and you can't.

The voice card, to paste into your notes, one for each character:

Character voice: [name] Words they own (3): __________ Words they never use (2): __________ Default sentence shape: short / long / qualified / fragmentary Under pressure they get: shorter / longer / quieter / more formal Grammatical habit: __________ What they notice first in a room: __________ Do they joke? Cruel / self-directed / deflecting / never: __________ What they refuse to say, and what they say instead: __________ When they are wrong they: concede / relitigate / change the subject / apologize for something else Where all of the above came from in their life: __________ Tuning-fork lines from book one: __________

Common mistakes and misunderstandings

Using a tic instead of a voice. A catchphrase, a repeated gesture or a habit of saying somebody's name passes the tags test and produces a flat character, as the Dickens example above shows. It's fine for a walk-on part and fatal for a lead.

A cast who all sound like the author. It's the default failure, and you can't see it from the inside, because your own way of talking is the one voice you can't hear. Every character shares your jokes, your sentence rhythm and your vocabulary, and only their opinions differ. The tags test catches it, and the voice cards fix it. See also supporting cast, where it shows up worst, because minor characters get the least attention.

Voice that drifts between books. It usually sneaks in inside something else. The character changes job, city or technology, and their speech follows. Write the voice down in book one, freeze what you can, and reread before book two.

Building a voice you can't justify. If the only reason a character talks that way is that you needed him to sound different from his brother, he'll slip back whenever you're writing quickly.

Reaching for accent and dialect first. It's the most common shortcut, and the one with the worst return. Register, rhythm, attention and refusal do the same work without any of the cost. The dialogue page covers why phonetic respelling (spelling words the way they sound) is a bad bet, including its history.

Forcing apart characters who should sound similar. Think of two sisters, two colleagues or two soldiers from the same unit. Tell them apart in small ways, and carry the clarity with blocking (where people stand and what they're doing), names and viewpoint. A made-up difference makes the relationship less believable, not more.

Only voicing the dialogue. A viewpoint character's narration is their voice too. Four narrators who speak differently and think the same way is the standard flaw of multi-POV drafts, and it's why readers say the strands blurred.

Mistaking a strong voice for a strong character. A vivid manner doesn't give a character agency (the power to act and choose), a want or an arc. A character with a memorable voice who does nothing is just a memorable extra.

Making everyone articulate. Under emotional pressure, real people get worse at speaking, not better. A cast who all express themselves well at the worst moments of their lives sounds like a cast.

Leaving voice until revision. You can polish rhythm late. You can't go back and fit in a voice that comes out of a history you haven't decided yet. That takes a rewrite, more than a line edit, which is why the voice card belongs before the draft.

Related elements