How to Make Characters Sound Different in Dialogue

To make characters sound different, change what drives their speech, not only its surface. Give each speaker their own vocabulary, sentence length, subjects they return to, subjects they avoid, and way of handling conflict. Accents and catchphrases help a little; the bigger differences come from what a character notices, what they want in the scene and how they protect themselves.

Then test it. Cover the dialogue tags on a page and ask whether you can still tell who is talking. If you cannot, pick the two speakers who blur most and pull them apart on two or three levers at once, starting with the table below.

StoryWith AI is our own product, so weigh what we say about it accordingly.

Key takeaways

• Characters blur because they all come from one head. The cure is deliberate contrast, planned before you revise.
• Voice works on two layers: how a character speaks and why they speak that way. Fixing only the first gives you costume, not character.
• The strongest levers are vocabulary and register, rhythm, what they notice, what they avoid and how they handle conflict.
• Tics and dialect are seasoning. Overused, they become the only thing a reader remembers about a character.
• The cover-the-tags test is quick and honest: if lines could be swapped between speakers, the voices have collided.
• Fix collisions by pair, not by character. Two speakers who sound alike need to move apart from each other.

Why do all my characters sound the same?

Because they share an author. Randy Ingermanson put it directly on Advanced Fiction Writing (post dated 12 December 2013): "all your characters inherit quite a bit of their personality from you." His suggested remedy is to borrow from real people, the more unlike you the better: "I mean people who are different from you. Wildly different."

Jacquelin Cangro makes the same diagnosis in her newsletter post "Why All Your Characters Sound Alike" (dated 16 December 2025): "If you're worried that all of your characters sound like you, they probably do."

Three habits make it worse in revision:

• Writing for information. When a scene's job is to deliver facts, every speaker becomes a delivery vehicle, and vehicles sound alike.
• Polishing every line to your own taste. Each pass sands speakers toward the author's preferred rhythm and vocabulary.
• Leaning on tags. If every line has a name attached, you never notice that the lines themselves carry no identity. (On tags versus action beats, see our guide to said bookisms.)

A large cast adds a fourth cause. If two characters have the same job in the story, they tend to end up with the same voice too. Our guide to handling too many characters covers when two such characters should become one.

What actually makes a character's voice distinct?

A post on the Rocky Mountain Fiction Writers site (undated, read on 30 October 2026) splits voice into two parts: "There are two layers behind character voice: how they speak, and why they speak that way."

The "how" is what readers hear. The "why" is history, wants and fears, and it is what keeps the "how" consistent across 300 pages. Cangro's version: "Desire creates friction, and friction creates voice." A character trying to impress talks differently from one trying to hide something, even when both are the same age from the same town.

So before changing a single line, write two sentences per main speaker: where their way of talking comes from, and what they usually want from a conversation.

What are the levers of character voice?

Each row is one dimension you can set differently for each speaker. The examples are invented for this guide.

Cangro's post gives a neat background illustration: "a scientist might reach for precision. A bartender might speak in anecdotes." Notice that neither example needs a single accent marker.

You do not need ten contrasts per character. Two speakers who differ clearly on three levers (say rhythm, conflict style and what they avoid) will usually pass the cover-the-tags test.

How much dialect and how many verbal tics should I use?

Less than feels necessary. The RMFW post warns that "crafting distinct voices doesn't mean slathering on the dialect or slang," and of habits: "verbal tics and pet phrases will add depth to your dialogue, but be careful not to overuse them."

Practical limits that keep dialect readable:

• Use word choice and grammar, not phonetic spelling. Ingermanson's advice for a speaker whose grammar differs from standard: "spell the words correctly but misuse the grammar the way the speaker actually says it."
• Set the voice early, then ease off. A few clear markers in a character's first scene teach the reader the voice; after that, a lighter touch keeps it.
• Check for stereotype. If a regional or cultural marker is doing all the characterising, the character has become a label. Cangro's third tip is titled "Let background shape language, not stereotype."
• Ration tics. A catchphrase in every scene stops being character and becomes a countdown. Once a scene, at most, is a sensible ceiling for most tics.

What is the cover-the-tags test?

Take a page of dialogue with three or more speakers. Cover or delete every tag and action beat, so only the spoken lines remain. Then read and try to name each speaker.

Cangro describes the goal the same way: "A strong character voice allows readers to recognize who's speaking even without dialogue tags."

Score it honestly:

• You name every line instantly. The voices are working.
• You can name them, but only from context (who would know that fact). The scene is carrying the voices, not the speech. That is fragile.
• Two speakers swap without loss. You have a collision. Note the pair.

The RMFW post offers a whole-book version on paper: "Print out your manuscript and skim through the whole thing, highlighting each character's dialogue in a different color," then "go back to page one, and read through only one color of dialogue." Read one colour at a time and a voice that drifts, or one that sounds like a neighbour, becomes much easier to hear.

What does fixing a three-person exchange look like?

A hypothetical example, invented for this guide. Three people find the till in a small bookshop short of money on Monday morning. Ruth owns the shop and spent 30 years teaching maths. Kai is nineteen and works weekends. Sol is Ruth's retired brother-in-law, who helps with the accounts.

Before (tags removed):

"I think someone has taken money from the till." "Are you sure? Maybe it was counted wrong." "I counted it three times. It's two hundred and twenty short." "Well, I didn't take it." "Nobody is saying you did." "I think we should check the card machine."

Any line could belong to any of them. All three use the same register, the same medium sentence length and the same neutral politeness. Nobody notices anything only they would notice.

After (tags still removed):

"Two hundred and twenty. Three counts. I don't miscount." "Okay but the card reader was being weird all Saturday, so." "Before anybody accuses anybody, may I point out that the Friday float was never written in the book? Not once. I've raised it every week since March." "Nobody's accusing." "You're kind of looking at me, though." "She looks at everyone like that. Thirty years of marking homework."

What changed, lever by lever:

The core facts did not change: the scene still delivers a missing sum and a suspicion. What changed is that each line could only come from one person. Notice too that none of the fixes relies on spelling an accent.

How is POV narration voice different from dialogue voice?

Dialogue voice is how a character talks aloud to other people. Narration voice, in close third or first person, is how that character thinks: the words the narrative uses when it sits behind their eyes.

The two should be related but not identical. People speak more guardedly than they think. A character who is all politeness aloud may narrate with real contempt, and that gap is often where the reader learns who they are.

Three practical points:

• In close POV, the narration takes on the viewpoint character's voice. A scene told from Kai's head should not suddenly use Sol's vocabulary in the description.
• When the viewpoint changes, the narration's vocabulary should change with it. If two POV chapters read the same once the names are covered, the narration has collided even if the dialogue has not.
• Keep one viewpoint per scene unless you mean to switch. Unplanned slips between heads blur voice faster than anything else; our guide to head-hopping covers how to spot them.

One sentence on a related problem: a line can be in perfect voice and still say too plainly what the character feels. That is on-the-nose dialogue, a separate fault with its own fixes.

How do you find voice problems across a whole manuscript?

A routine that works on a full draft:

1. List your speaking cast. For each, write the two "why" sentences from earlier and pick three levers that set them apart.
2. Find the pairs most at risk. Characters of similar age, job and role in the plot, and any two who share many scenes.
3. Sample, do not read everything. For each at-risk pair, pull three dialogue-heavy scenes from the beginning, middle and end.
4. Run the cover-the-tags test on each sample. Note every line you could not attribute.
5. Check for drift. A voice that is clear in chapter 2 and generic by chapter 20 usually faded during fast drafting. Compare the samples side by side.
6. Fix by pair. For each collision, move one speaker on one lever and the other on a different lever, so they separate in both directions.
7. Re-test with a reader. Ask a beta reader which two characters they mixed up. You can add that question to a beta reader questionnaire you already use.

You do not have to get every voice right in the first draft. Voice is a revision task as much as a drafting one.

Where StoryWith AI fits

StoryWith AI reads the manuscript you wrote and returns findings. It does not write the book, and it does not rewrite dialogue.

Its Dialogue analysis runs on one scene at a time, on fiction books only (the Dialogue page is hidden for non-fiction books). Each run uses one AI text action and reads up to about 14,000 characters of the scene; the scene needs at least 80 characters. It returns:

• each speaker with their line count, average line length and voice traits
• a 0 to 100 voice distinctiveness score, meaning how recognisable each speaker is by their lines alone
• pacing rhythm and subtext scores
• voice collisions: pairs or groups of speakers whose voices feel interchangeable, each with a short example excerpt and a note on what is missing
• up to 5 on-the-nose lines
• 2 to 5 recommendations phrased as options, not commands, plus a short summary

The voice collisions list is the automated version of the cover-the-tags test. The dialogue report lives on its own per-scene Dialogue page.

On plans (storywith.ai/pricing, 30 October 2026): Author costs $19 a month or $149 a year with 500 AI text actions a month; Pro costs $39 a month or $299 a year with unlimited AI text actions.

Paid plans include reader feedback sessions with readers you invite. StoryWith AI does not supply the readers. Its AI reader personas are a simulated first pass, not evidence of how real readers respond.

It runs on the web and Android, not iOS, and manuscripts are not used to train AI models.

It is the wrong tool if you want lines rewritten for you, or if your book is non-fiction. A distinctiveness score is a prompt to listen again, not a verdict, and it does not replace an editor's ear for whether a voice is right for the book.

Frequently Asked Questions (FAQ)

How do I make my characters sound different from each other?

Pick the two characters who blur most and set them apart on three levers at once, such as sentence rhythm, how they handle conflict and what they avoid talking about. Ground each choice in why the character talks that way. Then cover the dialogue tags on a page and check that every line can still be attributed.

Why do all my characters sound like me?

Every character comes out of the author's head, so they inherit the author's vocabulary, rhythm and opinions unless you plan otherwise. Randy Ingermanson's advice (2013) is to model voices on real people who are very different from you. Revision passes that polish lines to your own taste make the problem worse.

Should I use accents and dialect to make characters distinct?

Sparingly. Show a regional or cultural voice through word choice and grammar rather than phonetic spelling, establish it early and then lighten it. Heavy dialect is hard to read, and if it is the only thing distinguishing a character, it can tip into stereotype.

What is the cover-the-tags test for dialogue?

Remove or cover every dialogue tag and action beat on a page, then try to name each speaker from the spoken lines alone. If two speakers could swap lines without anything feeling wrong, their voices have collided and need separating.

Should a POV character's narration sound like their dialogue?

Related, but not identical. Narration in close POV reflects how the character thinks, and most people think more frankly than they speak. The gap between a guarded spoken voice and an honest inner voice is a useful way to show character.

Can software tell me if my characters sound the same?

Software can flag speakers whose lines look interchangeable and show you the excerpts where it happens. StoryWith AI's Dialogue analysis, for example, lists voice collisions per scene with a note on what is missing. It cannot judge whether a sameness is deliberate, such as twins or a chorus, so treat any flag as a prompt to reread the scene yourself.