Identifies when the narrative voice shifts inconsistently throughout a novel — the telltale sign of a manuscript assembled from multiple AI sessions without voice unification.

## Key Points

- Different chapters feel like they were written by different authors
- The narration shifts between literary, casual, and flat without pattern
- Multi-POV novel where all POV characters sound like the same narrator
- The prose style in Act 3 doesn't match Act 1
- After a Dialogue Voice Audit, to check the NARRATOR's voice (not just characters)
- Each generation session starts fresh, finding a slightly different voice
- The AI's "default voice" (neutral, slightly literary, generically competent) creeps in when the prompt doesn't strongly enforce a specific voice
- Style imitates whatever was most recently prompted rather than maintaining manuscript consistency
- Emotional scenes shift to a more "literary" register; action scenes shift to a more "commercial" register
- Sentence length average shifts by 20%+ between chapter clusters
- Vocabulary complexity changes (suddenly more/fewer Latinate words)
- The narrator's personality changes (wry in chapters 1-5, earnest in 6-10, detached in 11-15)

## Quick Example

```
READABILITY BY CHAPTER:
Ch 1: 62  Ch 2: 65  Ch 3: 61  |  Ch 4: 78  Ch 5: 74  Ch 6: 76  |  Ch 7: 58  Ch 8: 55
                                |                                  |
              Session A         |         Session B                |     Session C
         (grade 8-9 level)     |    (grade 6 level — simpler)     | (grade 10 — more complex)
```

Voice Drift Detector

Identifies when the narrative voice shifts inconsistently throughout a novel — the telltale sign of a manuscript assembled from multiple AI sessions without voice unification.

When to Use This Skill

• Different chapters feel like they were written by different authors
• The narration shifts between literary, casual, and flat without pattern
• Multi-POV novel where all POV characters sound like the same narrator
• The prose style in Act 3 doesn't match Act 1
• After a Dialogue Voice Audit, to check the NARRATOR's voice (not just characters)

---

What Is Voice Drift?

Every novel has a narrative voice — the "personality" of the narration itself. In first person, it's the character's voice. In close third, it's the character's voice filtered through the narrator. In omniscient, it's the narrator's own voice.

Voice drift occurs when this voice changes unintentionally. AI causes drift because:

• Each generation session starts fresh, finding a slightly different voice
• The AI's "default voice" (neutral, slightly literary, generically competent) creeps in when the prompt doesn't strongly enforce a specific voice
• Style imitates whatever was most recently prompted rather than maintaining manuscript consistency
• Emotional scenes shift to a more "literary" register; action scenes shift to a more "commercial" register

---

The Five Types of Voice Drift

Type 1 — Session Drift

The most common. Chapters generated in different sessions have detectably different voices.

Indicators:

• Sentence length average shifts by 20%+ between chapter clusters
• Vocabulary complexity changes (suddenly more/fewer Latinate words)
• The narrator's personality changes (wry in chapters 1-5, earnest in 6-10, detached in 11-15)
• Specific stylistic choices appear and vanish (em dashes used heavily in some chapters, absent in others; semicolons abundant then gone)

Diagnostic: Compute a readability score (Flesch-Kincaid or similar) for each chapter. Plot the scores. Clusters of similar scores that shift abruptly indicate session boundaries.

READABILITY BY CHAPTER:
Ch 1: 62  Ch 2: 65  Ch 3: 61  |  Ch 4: 78  Ch 5: 74  Ch 6: 76  |  Ch 7: 58  Ch 8: 55
                                |                                  |
              Session A         |         Session B                |     Session C
         (grade 8-9 level)     |    (grade 6 level — simpler)     | (grade 10 — more complex)

Type 2 — Register Drift

The formality level of the narration shifts based on scene content rather than maintaining a consistent register.

What AI does:

• Action scenes → short sentences, punchy, commercial style
• Emotional scenes → longer sentences, metaphor-heavy, literary style
• Dialogue scenes → flat, minimal narration, screenplay-adjacent
• Exposition → textbook-like, authoritative, distant

What a consistent voice does:

• Maintains the same fundamental personality across all scene types
• The narrator sounds like themselves whether describing a fight or a funeral
• Sentence length varies for rhythm but the vocabulary and attitude stay consistent

Type 3 — POV Bleed (Multi-POV Novels)

In multi-POV novels, each POV character should filter the narration through their unique perspective. AI often writes all POV chapters in the same narrator voice with the character's name swapped in.

Diagnostic questions:

• If you removed the POV character's name, could you tell whose chapter this is from the narration alone?
• Does the narration notice different things depending on who's observing? (A chef notices food. A soldier notices exits. A child notices adults' shoes.)
• Does the vocabulary of the narration shift with the POV character? (A teenager's close-third should not use the same words as a professor's.)
• Does the narration's emotional temperature change with the POV character?

The swap test: Take a paragraph from Character A's chapter and a paragraph from Character B's chapter. Swap them. If neither feels wrong in the other character's chapter, there's POV bleed.

Type 4 — Tonal Whiplash

The novel's overall tone shifts between serious and light, dark and warm, without the shifts being earned by story events.

What AI does:

• Chapter 7 ends on a devastating note → Chapter 8 opens with lighthearted banter as if nothing happened
• The narrator cracks jokes during a scene that should be somber
• The novel is darkly comic for three chapters, then completely earnest for three chapters, then darkly comic again

What healthy tonal shifts look like:

• Tone follows story logic (things are lighter when things are going well, darker when they're not)
• Shifts are gradual or deliberately jarring for effect
• The underlying voice personality remains consistent even as tone shifts (a wry narrator is wry about both happy and sad things)

Type 5 — Default Voice Intrusion

The AI's "house style" — neutral, competent, slightly literary, vaguely contemporary — intrudes over the intended voice.

AI default voice characteristics:

• Moderate sentence length (12-18 words average)
• Occasional sophisticated vocabulary but mostly mid-register
• Tendency toward nature metaphors
• Balances showing and telling evenly
• No strong opinions or personality in the narration
• Inoffensive, slightly warm, generically "good writing"

Detection: Read five random paragraphs aloud. Do they sound like a specific person, or like "good AI writing"? A strong voice has personality. The AI default voice is personality-neutral.

---

Scanning Process

Step 1 — Voice Baseline Extraction

From the first 3 chapters (or whichever section best represents the intended voice), extract:

Quantitative markers:

• Average sentence length
• Sentence length variance (standard deviation)
• Paragraph length average
• Vocabulary complexity score
• Contraction frequency
• Punctuation patterns (em dash, semicolon, ellipsis, exclamation usage)
• Simile/metaphor density

Qualitative markers:

• Narrator attitude (wry, earnest, detached, warm, acerbic, playful)
• Narrator relationship to the reader (intimate, distant, conspiratorial, authoritative)
• Characteristic constructions (does the narrator use fragments? Rhetorical questions? Lists? Parentheticals?)
• What the narrator notices first (sensory priority: visual, auditory, tactile, olfactory)
• What the narrator finds funny, beautiful, or important

Step 2 — Chapter-by-Chapter Comparison

Measure each chapter against the baseline on all markers. Flag any chapter that deviates significantly.

Step 3 — Drift Map

VOICE DRIFT MAP:

Chapter:  1   2   3   4   5   6   7   8   9  10  11  12  13  14
Baseline: ██  ██  ██  ░░  ░░  ██  ██  ░░  ░░  ░░  ██  ██  ░░  ██

██ = Within baseline range
░░ = Drifted from baseline

DRIFT ZONES:
  Zone 1: Chapters 4-5 — register shifts to more literary, longer sentences
  Zone 2: Chapters 8-10 — register shifts to simpler, shorter, more commercial
  Zone 3: Chapter 13 — single chapter of very different voice (possible different session)

Step 4 — POV Distinction Matrix (Multi-POV Only)

Marker	Character A	Character B	Character C
Avg sentence length	14	15	13
Vocabulary	Mid	Mid	Mid
Sensory priority	Visual	Visual	Visual
Narrator attitude	Warm	Warm	Slightly warm
Unique constructions	None	None	None

DIAGNOSIS: All POVs are functionally identical. Each should have at least 3 markers that differ significantly.

---

Recalibration Guidance

For each drift zone, provide:

1. What the voice should sound like (based on the established baseline)
2. What it actually sounds like in the drift zone
3. Specific sentences to revise with before/after examples
4. The likely cause (session change, scene-type register shift, or default voice intrusion)

DRIFT ZONE 2 (Chapters 8-10):
  BASELINE VOICE: "She tasted the wine and found it thin, the kind of Beaujolais
  that restaurants pour when they're hoping you won't notice."
  
  DRIFT VOICE: "She took a sip of her wine. It wasn't great. She put the glass down
  and looked around the restaurant."
  
  CAUSE: Likely new session. Voice lost its specificity and personality.
  
  FIX: Restore the narrator's opinionated, sensory-specific quality.
  Rewrite flat sentences to match the baseline's characteristic attitude.

---

Output Format

# Voice Drift Report
**Title**: [Novel title]

## Voice Baseline (extracted from chapters 1-3)
[Quantitative and qualitative markers]

## Drift Map
[Visual chapter-by-chapter consistency chart]

## Drift Zones
[Per-zone analysis with cause and fix guidance]

## POV Distinction Matrix (if multi-POV)
[Per-character voice comparison]

## Session Boundary Estimates
[Where the AI sessions likely changed, based on voice shifts]

## Priority Revisions
[Ordered list of chapters/passages most in need of voice recalibration]

---

Anti-Patterns

• Demanding robotic consistency. A real human narrator's voice naturally evolves slightly over a novel. Some drift is organic. Flag only drift that feels unintentional or jarring.
• Confusing intentional register shifts with drift. A narrator who shifts to short, punchy sentences during an action scene is making a craft choice. Drift is when the shift has no narrative justification.
• Imposing a "correct" voice. The baseline comes from the manuscript, not from the auditor's preference. If the author's intended voice is simple and direct, don't flag it as "too plain."
• Treating multi-POV sameness as automatically bad. In omniscient narration, one narrator voice across all characters is correct. POV bleed only applies to close-third or first-person multi-POV.
• Over-prescribing stylistic quirks. Adding a character's verbal tic to every paragraph is not voice — it's a gimmick. Real voice is subtler.