Skip to main content
Film & TelevisionScreenplay Audit272 lines

ai-dialogue-detector

Detects AI-generated dialogue patterns specific to screenplays: on-the-nose dialogue where characters say exactly what they feel, exposition dumps disguised as conversation, every character being equally articulate, missing subtext, dialogue that reads as written rather than spoken, therapy-speak, characters narrating the plot to each other, and the AI tendency to resolve conflict through a single perfect speech. Produces a per-scene dialogue quality score with specific rewrite direction. Use when screenplay dialogue feels flat, over-written, or like characters are reading essays to each other.

Quick Summary28 lines
Identifies AI-generated dialogue patterns that kill screenplays — the ways machine-written dialogue fails to sound like real people talking under pressure in cinematic situations.

## Key Points

- Dialogue feels "written" rather than "spoken"
- Every character sounds equally smart and articulate
- Scenes feel like therapy sessions or TED talks
- Conflict resolves through monologues instead of action
- Dialogue is doing all the work because the AI can't write visually
- Coverage readers flag dialogue as the primary weakness
- "As you know..."
- One character explaining something the other would already know
- Dialogue that exists solely to inform the audience
- Characters asking questions they'd know the answer to
- Monologues summarizing backstory, lore, or world rules
- Every character speaks in complete sentences

## Quick Example

```
DETECTIVE CHEN
    Same block as the Holloway case.

          DETECTIVE RUSSO
    Don't start.
```
skilldb get screenplay-audit-skills/ai-dialogue-detectorFull skill: 272 lines
Paste into your CLAUDE.md or agent config

AI Dialogue Detector (Screenplay)

Identifies AI-generated dialogue patterns that kill screenplays — the ways machine-written dialogue fails to sound like real people talking under pressure in cinematic situations.

When to Use This Skill

  • Dialogue feels "written" rather than "spoken"
  • Every character sounds equally smart and articulate
  • Scenes feel like therapy sessions or TED talks
  • Conflict resolves through monologues instead of action
  • Dialogue is doing all the work because the AI can't write visually
  • Coverage readers flag dialogue as the primary weakness

The AI Screenplay Dialogue Tells

Tell 1 — On-the-Nose Dialogue

Characters say exactly what they think and feel. No gap between intention and expression. This is the #1 AI dialogue problem in screenplays because film dialogue lives in the gap between what characters say and what they mean.

AI writes:

          SARAH
    I'm angry at you because you
    lied to me and I feel betrayed.

          MARK
    I know I hurt you. I was afraid
    of losing you, so I made a
    terrible choice. I'm sorry.

A screenwriter writes:

          SARAH
    How was work?

          MARK
    Fine.

          SARAH
    Good. That's good.
    (beat)
    Your mother called.

          MARK
    What did you tell her?

          SARAH
    What do you think I told her?

The anger, betrayal, and lie are all present. Neither character names them. The audience feels more because they have to lean in.

Detection method: For each line of dialogue, ask: is the character saying what they mean, or something adjacent to what they mean? If more than 40% of dialogue is characters directly stating their feelings, motivations, or assessments, the script is on-the-nose.

Tell 2 — Exposition Dumps ("As You Know, Bob")

Characters tell each other information they both already know, purely for the audience's benefit.

AI writes:

          DETECTIVE CHEN
    As you know, the victim was found
    at 3 AM in the warehouse district.
    The same warehouse district where
    those three murders happened last
    year, which we never solved.

A screenwriter writes:

          DETECTIVE CHEN
    Same block as the Holloway case.

          DETECTIVE RUSSO
    Don't start.

Two lines. The audience gets that there's history, unresolved cases, and tension between the detectives. The AI version is a news report.

Detection method: Search for:

  • "As you know..."
  • One character explaining something the other would already know
  • Dialogue that exists solely to inform the audience
  • Characters asking questions they'd know the answer to
  • Monologues summarizing backstory, lore, or world rules

Tell 3 — Universal Eloquence

Every character speaks at the same vocabulary and emotional intelligence level. The CEO and the janitor both construct perfect sentences. Nobody stumbles, interrupts, or says something stupid.

What AI does:

  • Every character speaks in complete sentences
  • Every character can articulate their emotions precisely
  • No one uses filler words (um, uh, like, I mean, well, look)
  • No one interrupts or talks over each other
  • Every character's dialogue could be swapped with any other character's

What real screenplay dialogue does:

  • Characters have different vocabularies, rhythms, and intelligence levels
  • Some characters are inarticulate — they can't find the words, and that's the drama
  • Interruptions convey power dynamics (who gets to finish their sentence?)
  • Characters misunderstand each other because they're not really listening
  • One character speaks in run-on sentences while another speaks in fragments

The swap test: Take a line from Character A and attribute it to Character B. Does it still sound right? If yes for more than 50% of lines, all characters have the same voice.

Tell 4 — Conflict Resolution by Monologue

AI resolves dramatic conflict through one character delivering a perfect speech that changes the other person's mind.

AI writes:

          JAMES
    You know what your problem is?
    You're so afraid of being hurt
    that you push everyone away. But
    that's not strength — it's fear
    dressed up as independence. And
    someday you're going to realize
    that the people you pushed away
    were the only ones who ever really
    saw you.

Sarah's eyes fill with tears. She nods slowly.

          SARAH
    You're right. I've been so afraid.

A screenwriter writes:

James says something clumsy and wrong. Sarah walks out. James follows. They argue in the parking lot. Neither changes the other's mind. Three scenes later, Sarah does something that shows she heard him — but she'd never admit it.

Detection method: Search for monologues (4+ consecutive lines from one character) that:

  • Diagnose another character's psychological problem
  • End with the other character agreeing or being visibly moved
  • Resolve a conflict that's been building for multiple scenes
  • Sound like they were written, rehearsed, and delivered perfectly

Tell 5 — Therapy-Speak

Characters talk like therapists or therapy patients — using clinical emotional language that real people don't use outside a therapist's office.

AI dialogue red flags:

  • "I feel like you're not hearing me"
  • "I need you to validate my experience"
  • "That's your trauma talking"
  • "I'm setting a boundary"
  • "You're projecting"
  • "I need to sit with that"
  • "That triggers me"
  • "I hear you, and I appreciate your vulnerability"

Some characters (actual therapists, overly online millennials) might talk this way. When EVERY character does, it's AI default.

Tell 6 — The Summarizing Scene

AI writes scenes where characters discuss what happened in previous scenes instead of showing new events.

          ALEX
    I can't believe what happened
    at the restaurant last night.

          JORDAN
    I know. When she said that thing
    about your father, I thought you
    were going to lose it.

          ALEX
    I almost did. But then I thought
    about what Dr. Walsh said.

This scene is a podcast recap. The restaurant scene should have been ON SCREEN. If it was, this scene is redundant. If it wasn't, it should have been instead of this one.

Detection method: Count scenes where characters discuss events vs. scenes where events happen. If the ratio exceeds 30% discussion, the script is summarizing instead of dramatizing.

Tell 7 — Question-Answer Tennis

AI defaults to dialogue structured as Q&A — one character asks questions, the other answers. This creates the rhythm of an interview, not a conversation.

          EMMA
    Where were you last night?

          RYAN
    I was at Jake's.

          EMMA
    Until what time?

          RYAN
    Around midnight.

          EMMA
    And you didn't think to call?

          RYAN
    I didn't think you'd care.

Real conversations are messier. People answer questions with questions. They deflect. They answer a different question than the one asked. They change the subject. They attack.


Scoring

Per-Scene Dialogue Score

For each scene with dialogue:

MetricScore 1-51 =5 =
SubtextCharacters say what they meanMeaning lives beneath the words
Voice distinctionAll characters sound the sameEach voice is unmistakable
Exposition handlingInfo dumps in dialogueInformation woven into conflict
EconomyOverwritten, 10 lines where 3 would doEvery line earns its place
SpeakabilityReads like proseSounds like real speech rhythms
ConflictCharacters agree or resolve quicklyDialogue crackles with opposing wants
Behavior gapDialogue carries all meaningAction/behavior contradicts or complicates words

Manuscript-Level Metrics

  • On-the-nose percentage: [N]% of dialogue directly states feelings/motivations
  • Exposition dump count: [N] scenes with dialogue-as-information
  • Monologue resolution count: [N] conflicts resolved by speech
  • Voice distinction score: [swap test results]
  • Summarizing scene count: [N] scenes discussing offscreen events

Rewrite Principles

For every flagged line:

  1. What does the character WANT in this moment? (not feel — want)
  2. What are they ACTUALLY saying to get it? (probably not the truth)
  3. What's the other character doing while they talk? (action reveals more than words)
  4. What would this character NEVER say? (negative space defines voice)
  5. Can this line be cut entirely? (the best dialogue revision is often deletion)

The Silence Test

For every scene: what if you removed all the dialogue and played it silent? Would the audience still understand what's happening from the visuals and behavior alone? If not, the dialogue is doing too much. Film is a visual medium.


Anti-Patterns

  • Making all dialogue oblique. Not every line needs subtext. Sometimes "Pass the salt" means pass the salt. The key scenes need subtext; functional scenes can be direct.
  • Removing all exposition. The audience needs information. The craft is delivering it during conflict, not eliminating it. The best exposition happens when a character is forced to reveal information they'd rather hide.
  • Adding profanity for "realism." Swearing doesn't make dialogue authentic. Specificity, rhythm, and behavioral truth do.
  • Rewriting dialogue without understanding the scene's purpose. Every scene has a function. Dialogue rewrites must serve that function, not just "sound better."
  • Applying novel-dialogue standards. Screenplay dialogue is terser, more interrupted, and more dependent on the actor's delivery. Leave room for performance.

Install this skill directly: skilldb add screenplay-audit-skills

Get CLI access →

Related Skills

Animation Script Checker

Format-specific checker for animated screenplays across kids, adult, film, stop-motion, and CG formats. Verifies visual stageability, voice performance dialogue, gag specificity, age-appropriate content, and correct act structure for episode length.

Screenplay Audit168L

character-flattening-screenplay

Detects AI character flattening in screenplays — where characters lose complexity because AI can only convey interiority through dialogue (not narration like novels). Scans for: characters who speechify their feelings, arc-by-monologue, identical speech patterns across the cast, behavioral predictability, missing character-specific action/business, characters who exist only for exposition or to be wrong so the protagonist can be right, and antagonists who are either cartoons or have their "good point" neutralized by a single speech. Produces character depth scores and rewrite guidance specific to the visual medium. Use when screenplay characters feel thin, functional, or indistinguishable.

Screenplay Audit216L

Cross-Episode Continuity Checker

Specialized for multi-episode works including limited series, ongoing series, and web series. Tracks character knowledge, relationship status, prop and location continuity, and serialized arc beats across all episodes. Builds a continuity matrix and flags contradictions.

Screenplay Audit121L

Dialogue Subtext Analyzer

Analyzes screenplay dialogue for subtext depth. Scores each exchange on a 1-5 subtext scale, identifies exposition dumps and on-the-nose dialogue, and provides rewrite suggestions that add layers of meaning beneath what characters actually say.

Screenplay Audit164L

Documentary Script Checker

Format-specific auditor for documentary screenplays. Checks narration for factual accuracy and attribution, interview setups, B-roll directions, archival footage cues, and adherence to observational, expository, or participatory conventions.

Screenplay Audit138L

originality-logline-scanner

Screenplay originality scanner — detects stock premises, predictable beats, clichéd set pieces, genre-default plot structures, and derivative concepts at the story level. Catalogs 40+ screenplay-specific clichés (the ticking bomb, the training montage, the liar revealed, the slow clap, the mentor's death speech), assesses logline originality, identifies which produced films the script most resembles, and provides specific subversion strategies. Use when the screenplay feels derivative, predictable, or like an assembly of familiar parts from other movies.

Screenplay Audit226L