The Hostile Witness Method: Interview Your Protagonist Like a Prosecutor

A character development exercise for the moment when you know exactly what your protagonist does — and have no idea why they do it.

By Johnson Liu

The problem your outline can't solve

There is a specific kind of stuck that no amount of outlining fixes.

You know what your character does. You've watched them do it for two hundred pages. What you don't know is why — and every time you write the scene where the why should surface, you get behavior instead of motive. The character performs. The scene is technically fine. And you close the laptop with the feeling that you've just transcribed a stranger.

I spent two years like this with a character named Maya Chen, a Chicago defense attorney who only takes cases she knows she'll lose. That was her defining trait — the thing that made the book worth writing. I could describe the pattern in one sentence. I could not, for two years, tell you what the pattern was for. My outline said "fear of failure." My outline was wrong, and some part of me knew it, which is why the middle of the book kept dying on me.

What finally worked was not a better outline. It was an interrogation. One session rewrote my understanding of three hundred pages, and it followed a structure I've since repeated enough times to trust. I call it the Hostile Witness Method, and you can run it tonight with nothing but a notebook and a willingness to be rude to someone you love.

An old exercise with the safety off

The character interview is not a new technique. Actors call it hot-seating: one performer sits in a chair, in character, and the room fires questions at them. It's a rehearsal staple because it works — the actor discovers things about the role that aren't in the script, because the questions aren't in the script either.

Fiction writers have their own version of this, and most of it is too polite to be useful. Character questionnaires — the descendants of the Proust Questionnaire — will tell you your protagonist's favorite meal and childhood pet, which is trivia, not motive. And the standard "interview your character" journaling exercise fails for a subtler reason: you're interviewing a cooperative witness. When you sit your character down and ask warm, open-ended questions, they tell you what you already believe. Of course they do. You're writing both parts.

E.M. Forster argued that the test of a round character is whether it can surprise you in a convincing way. Here is the uncomfortable corollary: if your protagonist has never surprised you, you may not actually know them yet — and a friendly interview will never expose that, because friendly witnesses stick to their statements.

Trial lawyers solved this exact problem a long time ago. When a witness is aligned against you — or too rehearsed to be trusted — you don't ask open questions and hope. You cross-examine. You lead, you accuse, you refuse answers, you circle back. Not out of cruelty. Because pressure is the only reliable way to get past the story someone has practiced telling, down to the story underneath it.

Your protagonist has a practiced story too. You wrote it for them. The point of this exercise is to get underneath it.

The Hostile Witness Method: five steps

You need three things: your character, twenty to thirty minutes, and a way to keep the answers from coming out of your own assumptions — more on that partner problem below. The steps are the same regardless.

1. Open with an accusation, not a question

"Why do you keep taking losing cases?" invites the prepared answer — which is just your outline talking back to you in a character voice. Skip it. Open with a charge: You're hiding. You never forgave him. You wanted her to see it.

An accusation forces a defense, and people reveal what they value by what they bother to defend. Even a wrong accusation works, maybe better — the way a character corrects you shows you the true shape of the thing. A witness who says "I wasn't there at nine, I was there at eight" has told you more than one who says "no comment."

2. Refuse the first answer

First answers are alibis. They're polished because the character has given them before — to colleagues, to family, to themselves in the mirror. Every cross-examiner knows the move here: don't accept it. Say "that's not what I asked." Repeat the accusation flatly. Narrow the question and ask again.

The second answer is where the performance cracks. Not because the character was lying, exactly, but because nobody keeps two rehearsed answers for the same question. The second one has to be improvised, and improvisation runs on whatever is actually down there.

3. Ask about the case you haven't written

This is the step that changes manuscripts. Ask about events outside your book: the client before chapter one, the summer you never wrote, the sibling who's never once come up. Inside the manuscript, your character has a script — yours. Outside it, there is nothing to fall back on except who they are. Whatever comes out is pure character logic, unpolluted by your plot.

Be warned that this step generates canon. Characters under pressure will hand you whole scenes you now have to write. That's not a side effect. That's the yield.

4. Watch for the tells

Transcribe everything, because what a character says matters less than how they say it. What are their hands doing? What question makes them suddenly interested in logistics? What do they deflect with a joke, and what do they refuse to touch at all?

Pay special attention to answers that arrive in the wrong category. A witness who responds to a question about winning with a fact about sleep is not dodging — that's a confession with the serial numbers filed off. Your job is to notice, not to interrupt. Save the analysis for after.

5. End when they ask YOU a question

Every good session ends the same way: the witness turns it around. At some point the character stops answering and asks you something — and that flip is the signal that they've stopped being a puppet on the stand and become a person with their own theory of you.

You will not get anything better after that moment. Don't push for one more revelation. Thank the witness, close the session, and write down everything while it's hot.

The Maya transcript

Here's what the method looked like the night it earned its name. No greeting, no warm-up. I opened with the accusation (step one):

Me: You're hiding.

Maya: Excuse me? (she doesn't look up from the file she's not actually reading) ...That's not hiding. That's the opposite of hiding. Hiding is what people do in the suburbs. Who let you in, anyway? Denise leaves at six.

A textbook first answer: redefine the terms, challenge my standing, change the subject to office logistics. Note that the tells (step four) start immediately — she's holding a file she isn't actually reading. The prop matters more than the objection.

So I refused the answer (step two). Same accusation, flat, no elaboration. And the deflection dropped:

Maya: I'm surviving. There's a difference. One of them requires honesty about how badly I want to be loved for doing something impossible. I'm not there yet.

I'm not there yet. Two years of outlining never produced that sentence — a person locating herself on a timeline of her own avoidance, and admitting the timeline exists.

Then step three. I asked about a case that appears nowhere in my manuscript: the one she could have won. Long pause. And then she told me about the Darnell case — a case I had never written, never planned, that did not exist until she said it. A client she believed in. Evidence that was actually going her way. And then this:

Maya: I could feel the acquittal. Like weather coming. And I couldn't sleep.

Read that as a cross-examiner (step four). I asked about winning. She answered with insomnia. Losing, she can discuss in professional vocabulary all day — losing is her habitat. Nearly winning comes back as weather and lost sleep. The category error is the confession: for Maya, losing is controlled, and winning is exposure.

She ended the session herself, exactly on schedule (step five):

Maya: You want to tell me what that was?

I closed the notebook.

The unlock, once I laid the transcript out: Maya doesn't lose because she's incompetent, and she doesn't lose from fear of failure. She loses because winning feels more dangerous than losing. The doomed cases are the only place where nothing she does can disappoint anyone — all the hours, all the care, it's surplus; it's hers. Everything in her file said what she did. The interrogation said what it was for. I restructured three hundred pages around that sentence, and the Darnell case — which I never planned — is now a chapter.

Why this beats freewriting

You might reasonably think: I can just freewrite this. Journaling in your character's voice has real uses. But for excavating motive, it has a structural flaw you can't write your way around: you're playing both sides with one brain. Your questions carry your assumptions; your answers are generated by the same head that made the assumptions; and given ten minutes, the answers regress to the outline. This is why "interview your character" exercises so often produce pages that feel productive and change nothing. The witness and the prosecutor are colluding.

Cross-examination works in court because it's adversarial — two people who want different things. To understand your protagonist, you need the same asymmetry: a scene partner who doesn't share your assumptions and has no stake in your outline surviving the night.

There are a few ways to get one:

• A writing partner or critique-group friend. Hand them your character notes and one instruction: "You are not helping me. You have things you don't want me to know. You're allowed to deflect, stonewall, and lie." Timebox it to twenty minutes.
• An actor friend. This is hot-seating's home turf, and actors are frighteningly good at step four — they'll invent tells you'll steal for the page.
• An AI, set up to be the character rather than assist you. The same instruction applies and matters even more, because you have to switch the helpfulness off. A character who cheerfully explains their own psychology is worthless. The friction is the entire point.

Whichever partner you choose, the rule is identical: they must be allowed to refuse you. A witness who can't say no can't tell you anything you don't already know.

Try it tonight

Pick the character who's been stonewalling you — you know which one. Write a single accusation: the sentence a hostile prosecutor who had read your whole manuscript would open with. Don't prepare answers. Don't outline the interview; that defeats the purpose of the exercise. Find your scene partner, set a timer for twenty minutes, and run the five steps. Keep a transcript.

One of two things will happen. Either the session goes nowhere — which usually means your accusation was too polite, so sharpen it and go again tomorrow. Or, somewhere around the second refused answer, your character will say a sentence you have never heard before, and you'll feel the particular vertigo of discovering that someone you invented knows something you don't.

That vertigo is the why you've been outlining in circles to find. Go get it.