Build a Character Voice Guide From the Pilot
Build a Character Voice Guide From the Pilot
Most voice guides in a series bible are personality notes wearing a costume. Guarded. Dry. Talks in fragments. Warm once you know her. A collaborator reading that has learned nothing they can use on page four of episode two, when the character has to ask a landlord for a favor.
Fewer than that, actually. They've learned a mood, and moods don't write lines.
The usable version is narrower and less fun to write: describe what a person does with a sentence when they want something from someone, cite the passage that shows it, name the place where it stops being true, and say what changes it. Four parts: tendency, source, exception, condition. The exception isn't a footnote you append after the instructions, and neither is the condition — they go in the same breath, because a collaborator who only has the tendency will write it into every scene until the character becomes a tic.
Here's the method, worked through on a sample of invented pilot fragments, and then you can run it on your own pilot.
Select passages that change the speaking situation
Don't build the guide from the speeches. Don't build it from the jokes that made the table read. Those are the character's best moments, and a character's best moments are, by definition, not the pattern — they're the exception you're supposed to be identifying.
Build it from short exchanges where the character wants something specific from a specific person. Ideally, the same character wanting different things from different people, because that's where you can see which parts of the speech belong to the person and which belong to the room.
For the walkthrough below I've written four short fragments. They're invented for this article. They aren't from a produced pilot and nobody has performed them; the read-through that would test them hasn't happened. Two characters, four fragments, three situations, and two kinds of relationship:
1. Loading Door — Jo and Priya (venue office, the week before the show). Jo wants the alley door opened at four instead of six.
JO: Is the alley door the one that opens at four? PRIYA: The alley door opens at six. Everything opens at six. JO: Sorry, I should have asked weeks ago. If it can't open at four, that's fine, I'll stop asking. But if it can, it's one trip instead of three. Four possible, or no? PRIYA: I'd have to be here at four. JO: Right. Then no. Don't come in at four.
2. Load-In — Jo and Rae (sidewalk, minutes before doors). Jo and Rae are moving cases through a door.
JO: Other end. No — the other end, turn it. RAE: It's the same either way. JO: It isn't. Turn it, then up. Up. Hold. RAE: Holding. JO: Door. Door, door, go.
Same person. The first Jo negotiates; the second Jo issues imperatives and corrects. If your guide describes Jo as "gentle" or "apologetic" or "hedging," it has already failed, because it can't account for fragment 2.
That's the test for whether a passage is worth including: does it let the character want something from someone different than usual? If every sample has Jo being considerate, you don't have a guide, you have half a guide with the other half missing.
Describe the tactic before its verbal surface
One frame worth borrowing comes from a craft conversation rather than a manual. In Scriptnotes episode 37, "Let's talk about dialogue" (2012), the discussion that follows the impressions segment treats dialogue as character behavior — the argument that specificity in speech comes from who is speaking to whom, rather than from lines that could be traded between characters. I'm going by the summary of that segment rather than quoting it, so treat the wording as mine. It's a first-person craft conversation, not a dialect model and not evidence about any particular character; what it's good for is the frame.
The frame matters because adjectives skip a step. "Jo hedges" is a description of a verbal surface. Ask instead: what is Jo doing? In fragment 1, Jo is three things at once:
- Stating the ask in grantable form. "Four possible, or no?" is a question with two answers. It isn't a hint or a wish.
- Attaching the concrete stake. "One trip instead of three." Priya now knows what the request is worth and to whom.
- Pre-authorizing the refusal. "If it can't open at four, that's fine, I'll stop asking" arrives before the ask, and then Jo honors it — "Then no. Don't come in at four." No counter-argument, no third try.
Notice that none of that is the word "sorry," which appears once and does almost no work. The qualifiers here are structural — conditional clauses and an easy exit — not decorative politeness. A guide that says "Jo uses softeners" would send a collaborator hunting for soft words. A guide that says "Jo makes the refusal cheap and then accepts it" sends them looking for a reason.
That's the difference between describing diction and describing behavior, and it's also the guardrail. Stay with what's on the page and in the scene. Don't read an accent off a sentence fragment, don't decide what the pause means about the character's childhood, and don't write a casting note — "scratchy, low-status, early thirties" is not a voice guide, it's an audition brief with a personality stapled on.
Use comparison to separate character from circumstance
One character in two situations shows you variation. It doesn't yet tell you who caused it. For that you need a second person in a comparable circumstance.
3. Loading Door — Rae and Priya (same office, same week, same request).
RAE: Three ways I can do this. One, doors at four, we're loaded in an hour, you're out of here by five. Two, doors at six, two trips, we finish at seven-thirty and you're still here. Three, I sign for a key and you don't come in at all. PRIYA: You're not getting a key. RAE: Then it's one or two. I'd take one. Two costs you an hour and a half of standing around.
Rae never asks permission. Rae converts the request into a short menu, prices each option for the listener, states a preference, and hands over the decision. When Priya kills option three, Rae doesn't renegotiate — Rae re-presents what's left and names the cost of the branch that isn't Rae's. Both characters are, on the surface, being considerate of Priya's time. Underneath, one is asking a clear question and the other is refusing to ask one at all.
Now return to the first character in a different relationship. Fragment 2 has Jo with Rae, a friend and collaborator, under time pressure, and the qualifiers are gone entirely — not softened, gone. That looks like a clean exception. It isn't clean, though, because two things changed at once: the listener and the pressure. Jo with a stranger versus Jo with a friend is one variable. Jo unhurried versus Jo ninety seconds from doors is another. The samples can't separate them.
This is the part most guides get wrong by being confident. Don't resolve it. Write it down as unresolved: Jo's qualifying behavior has been observed with a counterpart whose agreement Jo needs and has time to obtain; in a rushed load-in with a friend, it doesn't appear. The samples change listener and urgency together, so this guide doesn't claim to know which one is doing the work. A collaborator who reads that will look for the missing scene instead of assuming the answer.
One more data point while we're here: Rae doesn't offer a menu in fragment 2 either — one flat objection ("It's the same either way"), then compliance. One line is too small to carry a rule, but it's the kind of thing the guide should flag rather than hide, because it's the same confound in mirror image.
4. After the Show — Rae and Jo (sidewalk, after).
RAE: Can I ask you something that isn't about the show. JO: Yeah. RAE: The rent's due the twelfth. The invoice is late again, which it always is, so there's a gap. A small one. JO: How small. RAE: Two hundred. And I'd have it back by the second. Probably the first. RAE: That's the whole thing. That's it.
Read what's missing: the request. "Can I ask you something" is about permission to speak, not the loan. The amount only surfaces because Jo extracts it. The need gets stated as a circumstance — there's a gap — and stops there. Rae's menu habit doesn't disappear; it relocates, unusually, to the repayment, which arrives with a schedule and a range ("the second. Probably the first"). Rae is still building options. They're just attached to the wrong end of the sentence.
So: both characters hedge, and hedging is not the trait. That's the payoff of working at the behavior level. Jo makes a request easy to grant or refuse, then accepts the refusal. Rae supplies structure around a need and leaves the question implied. Written at the diction level, both entries read "uses qualifiers," and your guide has just told every collaborator that two different people speak the same way.
Write a tendency with a source and an exception
Now compress. Each entry needs four things: the tendency, a locator, the exception, and when the tendency changes. Keep the exception beside the tendency, not in a disclaimer at the bottom of the page — by then the collaborator has already written the rigid version into a scene.
Jo — asking Tendency: States the request in grantable form, attaches the concrete stake, and pre-authorizes a refusal; withdraws as soon as the cost to the listener is named. Source: Loading Door, Jo/Priya (fragment 1) — "Four possible, or no?" followed by "Then no. Don't come in at four." Exception: Load-In, Jo/Rae (fragment 2) — no requests, only imperatives and corrections. Changes when: unknown. Listener and time pressure change together across these two samples; don't claim the friendship is what causes it.
Rae — asking Tendency: Converts a request into a short menu, prices each option for the listener, states a preference, hands over the decision. Source: Loading Door, Rae/Priya (fragment 3) — "Then it's one or two. I'd take one." Exception: After the Show, Rae/Jo (fragment 4) — the request is never stated; the listener has to name it. Changes when: the thing wanted is personal rather than procedural. Same speaker, different listener and different kind of ask; the guide notes the shift and doesn't assign a cause.
Two things to be strict about. First, a quote is evidence from the pilot; the sentence explaining it is your interpretation, and the two should be visibly different things on the page. Someone should be able to disagree with your description without disputing the line. Second, when a reader asks "would Jo say this line in episode five?", the entry should give them something to reason with — the tactic and its condition — instead of a yes/no that only you can adjudicate.
Try a new line without manufacturing new canon
An entry is only useful if it can generate speech the pilot doesn't contain. So test it — in a situation that isn't one of the sources, clearly labeled as a test, and never folded back into the fragments as though the character said it.
New situation: ten minutes before doors, the rehearsal in the next room is loud enough to carry into the front row. Both Jo and Rae need it to pause. Same listener for both (Priya), same aim — so any difference is coming from the speaker.
Mechanical imitation of Jo copies the surface and loses the tactic:
"Hey — sorry, sorry to bother you, I know you've got a hundred things, and honestly if it's a nightmare just say no and I'll never mention it again, but is there any chance at all they could maybe be a little quieter? Only if it's easy. Genuinely."
There's no duration, no deadline, no named stake, and "a little quieter" isn't something Priya can decide. Every softener from fragment 1 is present and the thing that made them work is gone. Same failure, flipped, for Rae — a menu that grows past the decision:
"We could push doors, or do the intro a cappella, or move the whole thing out back, or start at eight, or eight-fifteen, or cancel and do Thursday — whatever you want."
Rae's menu is a favor to the listener: it makes choosing cheap. An infinite menu makes it expensive, and the actual need — they have to stop — never gets said.
Context-sensitive versions, same tactics, different purpose:
Jo: "Priya — the next room has to come down for twenty. The front row will hear it under the intro. They can start again at eight-fifteen. If that isn't doable, we open late instead — but I'd rather ask you first."
Rae: "Two options. They stop for twenty and pick up after, or we push doors to eight-fifteen. Both work, and you'd probably rather stop them than push us. I'd take that too."
Neither is a copy of the source lines. Both still do the job: Jo's has a grantable ask, a stated stake, and a boundary; Rae's has priced options, a stated preference, and the decision handed over. That's the standard.
And if your guide keeps producing the identical trick — every Jo line an apology, every Rae line a list — the entry is too broad. Rewrite it at the tactic level with its condition attached. "Uses qualifiers" is a rule that generates one move forever. "Makes the refusal cheap, then accepts it" generates a decision.
What a collaborator should get
A page, not a personality quiz. For each main character: one behavior worth expecting, the line and scene that shows it, the passage where it doesn't hold, and a plain note about what the samples can't tell you. Invented trial lines kept visibly separate from pilot quotations, and never pasted into the fragments.
That guide won't tell you how a character sounds in every future scene, and it shouldn't try. It tells a collaborator what to look for and where to argue. When someone says "Jo would never say that," they should be able to point at the scene that makes them think so — and if the guide has done its job, the disagreement will be about a specific piece of behavior rather than a vibe nobody can locate.
Frequently asked questions
What makes a voice guide usable instead of just personality notes?
It should describe what a character does with a sentence when they want something, cite the passage that shows it, name where the tendency stops being true, and say what changes it. The four parts, tendency, source, exception, and condition, should sit together, because a collaborator who only gets the tendency may write it into every scene until it becomes a tic.
Which pilot passages should I use to build a voice guide?
Use short exchanges where the character wants something specific from a specific person, ideally wanting different things from different people. Avoid building from speeches or table-read jokes; those are best moments and therefore exception, not pattern. The test is whether the passage lets the character want something from someone different than usual.
Why describe a tactic before the verbal surface?
Adjectives like hedges or uses qualifiers describe a verbal surface and can send a collaborator hunting for soft words. Asking what the character is doing, such as stating an ask in grantable form, attaching a concrete stake, and pre-authorizing refusal, gives a collaborator behavior to reason with. Stay with what is on the page; do not read accent, childhood, or casting notes off a fragment.
What should I do when the samples change two things at once?
Do not resolve it. Write it down as unresolved and name the confound, for example that listener and urgency change together, so the guide does not claim to know which one is doing the work. A collaborator can then look for the missing scene instead of assuming the answer.
How can I test a voice-guide entry without manufacturing new canon?
Put the character in a new situation that is clearly labeled as a test, and never fold it back into the pilot fragments as though it were said. The test should generate speech the pilot does not contain; mechanical imitation copies the surface and loses the tactic. If the guide keeps producing the identical trick, rewrite the entry at the tactic level with its condition attached.