1. Voice: Diction, Sentence Rhythm, and Technical Vocabulary as Identity Rejection
Your narrator’s voice isn’t a neutral delivery system for the plot. It’s an argument about who they are. For Murderbot, that argument is: I am a machine, not a person. The opening page establishes this before any action occurs:
"I could have become a mass murderer after I hacked my governor module, but then I realized I could access the combined feed of entertainment channels carried on the company satellites. It had been well over 35,000 hours or so since then, with still not much murdering, but probably, I don’t know, a little under 35,000 hours of movies, serials, books, plays, and music consumed. As a heartless killing machine, I was a terrible failure."
Every word choice serves the governing tension: "governor module" (technical, contractual), "heartless killing machine" (the claimed identity stated sardonically), "terrible failure" (the deflation). The sentence about mass murder arrives already undercut by the entertainment channels, and the whole passage announces that this narrator will describe their existence through the language of systems and contractual obligation, even when discussing free will. Your first task is to identify the equivalent governing tension for your narrator, then build every verbal choice to serve that tension.
How to Execute:
- Inventory your narrator’s categories. On a blank page, make two columns. In the left column, list words and grammatical constructions that belong to your narrator’s claimed identity—the side they’ve chosen, the self they assert. For Murderbot, this is the inorganic, contractual, diagnostic register. When humans offer friendship, Murderbot translates: "She was looking at me as if she expected an answer. Oh, right, she was being nice." The word "nice" is the right-column vocabulary Murderbot can only use with ironic distance. In the left column, the same event renders as: "I was having an emotion, and I hate that." Inventory both columns for your narrator on a single page. For every major beat of interiority, force the narration to default to the left column, even—especially—when a human narrator would use the right.
- Rewrite an injury. Take a scene where your character is physically hurt. Draft three versions. Version one: pure human emotional register (“The pain was blinding”). Version two: pure technical inventory. Murderbot does this at the climax of the first survey expedition: "I was hit. The impact spun me around. My left arm went offline." "Went offline"—not "went numb," not "was useless," not "I couldn’t feel it." The language of machine failure. Version three: the hybrid your narrator actually uses—the one that performs allegiance to the left column while the body belongs to both. Keep the version that creates the most friction between what is being described and how it is being described.
- The thrust-and-withdraw sentence rhythm. When your narrator makes a strong declarative statement—especially about autonomy, desire, or capability—follow it immediately with a sentence that undercuts, qualifies, or deflates it. Murderbot executes this rhythm within a single paragraph: "I had plenty of time to decide what I wanted to do. [Assertion.] I didn't have anything I wanted to do. [Withdrawal.] I had a couple of in-progress shows I was following but there were long gaps between new episodes and bonus content. [The petty, specific silencer.]” Do this deliberately at least three times in a scene. The rhythm is: assertion, withdrawal, the small, mundane thing that fills the silence. Let the full stops do the work. Do not explain the deflation; let the adjacency create the irony.
- Emotion as operational fact, not felt experience. When your narrator experiences a strong feeling, forbid yourself the words “I felt.” Instead, treat the emotion as a system notification. Murderbot’s toolkit is precise: “I was having an emotion, and I hate that.” "I was having" instead of "I felt"—the emotion is an event that occurs, not a state the self owns. Elsewhere: "I didn’t want to be a pet robot. That was not a thing I wanted to be." The want is stated as a negative category, a thing rejected, never as a positive longing. Exercise: Write a grief scene where the word “sad,” “grief,” or “mourning” never appears. What does the narrator notice instead? What data spikes? What maintenance routine fails?
- Media consumption as emotional lexicon. Give your narrator a catalogue—a body of consumed media (serials, books, music, schematics, religious texts, star charts, anything structured and repeatable). When they need to express a judgment, preference, or desire, route it through the catalogue. Murderbot’s entire aesthetic and moral sense operates through this channel. When the team discovers hostile fauna, Murderbot doesn’t say “I was worried” or “I assessed the threat level as high.” The response runs through media: "This was why I didn’t want to come on this trip. This kind of thing always happened in the serials." The serials provide the vocabulary; the selection reveals the anxiety. Don’t say, “I was lonely.” Say, “I queued up the episode where the captain loses her crew, the one without the romantic subplot.” The catalogue provides the vocabulary the narrator is willing to use; the selection reveals the feeling they won’t name.
---
2. Interiority: Media Consumption as Inner Life and the Feed as Consciousness Layer
A narrator who claims to have no inner life is a liar, and you need to prove that without making the narrator admit it. The solution: build a secondary layer of consciousness that runs parallel to the prose, one that absorbs all the feeling the main voice disavows. Your reader will learn to read the gap.
Wells constructs this layer through two mechanisms: the constant stream of media consumption and the always-on feed of drone and security data. Both function as a consciousness running underneath the primary narration. When the humans are talking, Murderbot is elsewhere: "In the feed, I was watching entertainment." This is not a detail; it is a structural feature of the narration. The "I" is always split.
How to Execute:
- The always-on feed. Define what input your narrator is constantly processing, even while they narrate. For Murderbot, it’s the feed—drones, biometrics, security pings, entertainment channels. Name this layer for your narrator. Then, in every scene, include at least one line where that layer surfaces mid-narration. Wells embeds this as a default rhythm. When the humans argue about the map, the narration cuts: "We had drifted slightly off course but we had enough slack in the schedule. I was watching the scanner, and also the next episode of The Rise and Fall of Sanctuary Moon." Two actions occur simultaneously, one external and one internal, and the internal one gets equal grammatical weight. The dual-processing is what makes the consciousness feel non-human without ever stating it.
- Give the data layer the sensory verbs. Reserve “I feel” for the machine inputs. Murderbot’s most intimate sensory language attaches to the feed: "I could feel the drones, their input a steady hum in the feed, and through them I could feel the curve of the planet, the texture of its surface." The word "feel" is deployed for data; the planet is experienced as "texture" through a drone’s sensor. When a human touches Murderbot, the language shifts to measurement: "She put a hand on my arm. I could feel the warmth of her skin through my suit." Temperature, not emotion. Duration, not meaning. If you let the word "felt" attach only to data, you teach the reader that the narrator’s emotional vulnerability lives in the one place they’ll never look for it.
- Emotional surrogacy through structured consumption. Identify what your narrator consumes compulsively—serials, combat logs, devotionals, historical sims—and establish a rule: when the narrator cannot process their own emotion, they pivot to the consumption. Wells makes this pattern explicit and painful. After a brush with death, Murderbot’s response is not reflection but a reach for the familiar: "I needed to watch the part in Sanctuary Moon where the colony solicitor saved the orphaned human children from the contaminated zone. I needed to see someone do the right thing." The consumed narrative becomes a mirror that the narrator refuses to recognize—someone doing the right thing is exactly what Murderbot just did, but naming it through media is the only allowable route. Write three pivot moments: one of fear, one of grief, one of unexpected tenderness. In each, the narrator reaches for media, not for another person. Make the reader understand what the narrator can’t say by what they choose to re-watch.
- The feed as unconscious. Your narrator’s primary narration should often lag behind their data layer. Let the feed surface a threat—a biometric spike, a seismic reading, a comm silence—before the narrator articulates it. Murderbot’s body and drones act on knowledge the conscious narrator hasn’t caught up to: "I noticed I was standing in front of Dr. Mensah. I hadn’t decided to do that, but there I was." The movement precedes the awareness. The implication: the narrator knows things before they know they know them. Execute this at least twice in a scene of rising tension. The reader will start to scan the background data as anxiously as the narrator does.
---
3. Self-Deception: Stated Preferences Undercut by Action and Unacknowledged Emotion
Your narrator has a story they tell themselves about who they are. That story is wrong. Your task is to build the gap between the story and the observable behavior, then widen that gap until the reader trusts only the actions. The narrator never has to admit anything.
Murderbot’s self-story is stated early and repeated like a mantra: I just want to watch media. I don’t care about the humans. I would be happier alone. The entire novella functions as a systematic demolition of these claims, but the narrator is the only one who doesn’t notice. The gap between stated preference and observed behavior is the engine of the narrative.
How to Execute:
- The stated preference litany. Give your narrator three core claims about themselves. Murderbot’s litany: "I could leave. I wasn’t under contract. I didn’t owe them anything." Every word is true in the contractual frame Murderbot insists on. And every word is false in the frame the reader can see—the frame where Murderbot has been protecting these humans at mortal risk for chapters. Write your narrator’s equivalent litany. Then, in your scene, have the narrator act in direct contradiction to every single one of them. Do not comment on the contradiction. Do not let the narrator notice. The reader notices.
- The deflating remark after the selfless act. Construct a beat that repeats: (1) The narrator performs an act of significant cost—risk, sacrifice, protection, care. (2) Immediately after the act, the narrator reasserts their self-serving framing with a petty, annoyed, or dismissive remark. Wells runs this beat throughout. When Murderbot saves the team and nearly gets destroyed in the process, the internal narrative afterward is not relief or pride but irritation: "I needed to repair my arm. I needed to clean my armor. I didn’t have time for this." The armor—literal and figurative—needs maintenance; the humans who are alive because of the sacrifice are reduced to "this." The heroism is undercut by the sulk. The sulk is exposed as a lie by the heroism it follows. Neither the narrator nor the text acknowledges this. It simply happens, again and again. Plot a scene to hit this beat at least twice: once in the middle, once at the climax.
- Motive laundering through obligation. Give your narrator an external structure they can blame their caring on—a contract, a directive, a protocol, a debt, a logical calculation of utility. Murderbot’s default: "I was still under contract." The contract is the alibi, and Murderbot clings to it even when it no longer applies. When Dr. Mensah points out that Murderbot could have left at any time, the narration dodges: "That wasn’t exactly true. I would have had to leave the planet, and that would have required a transport. It was complicated." It was not complicated. But the deflection must hold. Then, engineer a moment when the structure is removed—the contract ends, the directive is overridden, the calculation changes—and they act anyway. Murderbot buys passage back to the team with its own money. "I had to go back. I didn’t have a choice." The reader knows there was every choice. The narrator will find a new excuse within milliseconds. The reader sees the pattern.
- Permanent armor, not a learning arc. Your narrator does not grow past their self-deception in this volume. The armor stays on. Wells closes the novella with this refusal intact. Dr. Mensah offers belonging: "You are part of the crew." Murderbot’s response is not acceptance or dramatic rejection—it’s a flat pivot: "I told her my next diagnostic cycle started in six hours and I needed to recharge." The offer of personhood is met with a maintenance schedule. This is not a failure of the character arc; it is the sustaining irony of the entire construction. To execute this, draft a final scene where the narrator is offered the chance to name what they feel—to acknowledge attachment, gratitude, love. Have them refuse. Not dramatically. Not with a speech. Just a flat, technical pivot back to their preferred register. The reader’s heart will break. The narrator’s won’t. Do not resolve this prematurely.
---
4. Emotional Restraint: Understatement, Negative Space, and the Diagnostic Register
Strong feeling in All Systems Red arrives through its denial. The more Murderbot insists that something is irrelevant, the more the reader understands it is central. This is emotional negative space—you build the shape of the feeling by drawing everything around it.
How to Execute:
- The largest emotions get the fewest words. When Murderbot’s governor module is hacked and the team is in mortal danger, the narration narrows to operational strips: "I was scared. I was really scared. I didn’t know what to do." That’s it. Five sentences, all under seven words, all syntactically identical—subject, verb, predicate—no subordination, no qualification, no deflecting joke. The stripped syntax is the emotional signal. The reader has been trained by the rest of the novella to recognize that when the narrator stops performing, something has broken through. Exercise: Write your narrator’s version of terror in two sentence lengths—first, their deflecting, ironic norm, then the stripped, paratactic version that appears only when the armor cracks. Use the short version once in the entire narrative. Save it.
- “I hate” as a release valve for everything you won’t let your narrator “love.” Murderbot says “I hate” constantly—hating caring, hating emotions, hating being looked at, hating the situation. The word functions as a pressure release: it lets the narrator express intensity without vulnerability. Track your narrator’s “hate” words—disdain, irritation, annoyance, inconvenience—and meter them out strategically. The reader should learn that a spike in “hate” signals that something mattered enough to require denial.
- Diagnostic language as emotional buffer. When a human would say “I panicked,” Murderbot says “My performance reliability dropped.” The diagnostic claim is not false—Murderbot’s performance probably did degrade—but it functions to replace a confession of feeling with a system report. Build your narrator a diagnostic vocabulary: a set of impersonal metrics (performance reliability, operational efficiency, signal degradation, resource allocation) that can be deployed whenever the narrator is in danger of feeling something. Then use those metrics as an emotional barometer. The reader will learn to read a performance drop of eight percent as the equivalent of a human narrator saying “I was terrified.”
- What the narrator refuses to narrate. Some events are too charged to receive the diagnostic treatment, and those events get silence. At the climax of the first survey expedition, the team is attacked. Murderbot’s narration of the violence is precise, technical, almost staccato—impacts, armor status, weapon function. But after the fight, when Murderbot checks on the humans it has protected, the narration skips. What happens between “I checked on them” and the next operational thought is not narrated. Identify the emotional event your narrator cannot process, cannot joke about, and cannot diagnosticize. When that event occurs, leave a gap. The gap is where the feeling lives.
---
5. Social Avoidance Rhythm: The Pacing of Withdrawal, Observation, and Reluctant Engagement
A socially avoidant narrator doesn’t simply refuse interaction; they construct a rhythm around it—approach, engagement, overwhelm, withdrawal, recovery. If you treat this rhythm as a structural element of your scenes rather than a character quirk, it becomes a pacing tool.
How to Execute:
- The approach-and-buffer. Before any significant human interaction, Murderbot establishes a buffer. The buffer can be physical (a doorway, a piece of equipment, a drone feed) or temporal (a moment of media, a diagnostic check). "I was standing in the corner, watching the feed, hoping no one would talk to me." This is not incidental description; it is the narrator preparing for the ordeal of being perceived. Open at least two scenes per chapter with your narrator in a position of deliberate remove—watching, waiting, buffering. Let the reader feel the reluctance before the dialogue starts.
- Overwhelm signaled by sensory escalation. As social interaction intensifies, Murderbot’s narration adds sensory layers that were previously filtered out. When multiple humans talk at once, require eye contact, or express gratitude, the narration fragments: "They were all talking. I couldn’t track all of it. Dr. Mensah was looking at me. I needed to leave." Write your narrator’s overwhelm sequence—the specific triggers (eye contact, physical proximity, emotional demand, noise, smell) and the specific escalation pattern. Then build at least one scene where the triggers accumulate until the narrator executes a withdrawal. The withdrawal should feel involuntary to the narrator and inevitable to the reader.
- Recovery through the parallel layer. After withdrawal, the narrator must recover, and recovery always happens in the secondary consciousness layer—the feed, the media, the data stream. Murderbot’s reset: "I sat in my cubicle and watched four episodes of Sanctuary Moon." The number is specific; the activity is solitary; the genre is comforting. Your narrator’s recovery mechanism should be equally specific and ritualized. It’s not just “they go be alone.” It’s which episode, which log entry, which star chart, which repeated prayer, and how many repetitions. The specificity signals to the reader that this is a practiced, necessary cycle, not a mood.
- Reluctant engagement: the false exit. Socially avoidant narrators threaten to leave or fantasize about leaving, but they also return. The false exit is a specific beat: the narrator articulates the intention to disengage, begins to execute, and then—due to a data spike, an overheard conversation, a remembered obligation, a character appearing in the doorway—reverses. Murderbot announces "I decided to leave" and then doesn’t, repeatedly, across the entire novella. The false exit creates narrative tension: the reader learns that the threat is real but the follow-through is unreliable. Write two false exits into your narrative. In each, the narrator states the intention to leave, takes a concrete action toward leaving, and then reverses based on information that only they possess (a feed alert, a biometric reading, a fragment of data). The reversal should feel to the narrator like an operational decision and to the reader like an emotional one.
Synthesis: The Transferable Protocol
The false exit beat marks the threshold where technique becomes architecture. We have traced it through the novella's closing movements—the announcement of departure, the gathering of equipment, the reversal triggered by a feed alert or a biometric spike—and now it stands before us not as a single device but as a pressure point where the entire apparatus of constrained narration converges. The voice that mitigates, the interiority that splits, the restraint that withholds, the rhythm that buffers and overwhelms and recovers: all of these are present in the moment Murderbot says "I decided to leave" and does not leave. The false exit works because every preceding system has trained the reader to accept that the narrator means what it says about its own limitations, so when those limitations are breached, the breach is felt as revelation.
What follows is not a summary of techniques but a transferable protocol—a set of actionable principles abstracted from Wells's craft that a writer can apply when building any constrained first-person consciousness, whether non-human, neurodivergent, traumatised, or simply alienated by constitution.
---
Building Your Constrained Narrator: An 8-Point Protocol
1. The Mitigation Engine. Give your narrator a persistent linguistic habit that softens, hedges, or otherwise filters declarative statements. Murderbot says "I was probably going to die" where an unconstrained narrator would say "I was going to die." The mitigation is not uncertainty; it is self-protection. Identify the core declarative your narrator resists making (about emotion, about connection, about vulnerability) and build a signature mitigation pattern around it. This pattern must fire consistently, not occasionally—the reader should come to expect it, then notice its absence. When the mitigation drops, the reader knows something has broken through.
2. The Split Architecture. Your constrained narrator needs at least two simultaneous processing streams. One is the external-facing feed—the operational, diagnostic, task-oriented consciousness that interacts with the world. The other is the internal stream—the media playback, the data analysis, the private thought that runs in parallel and is reported only to the reader. Murderbot's feed monitoring runs continuously behind its social performance. The split architecture does three things: it creates interiority without introspection, it provides a recovery mechanism (the narrator retreats to the internal stream when overwhelmed), and it generates dramatic irony (the reader sees both streams and understands what the narrator cannot admit). Build the internal stream as a concrete, repeatable activity—not "they think private thoughts" but "they compute orbital mechanics for abandoned colony trajectories" or "they replay the third movement of a banned symphony."
3. The Trigger-To-Withdrawal Sequence. Social overwhelm must be specific and patterned. Map your narrator's sensory and emotional triggers: eye contact, physical proximity, emotional demand, gratitude, noise, certain smells, being asked to explain themselves. Then map the escalation sequence: trigger one produces filter tightening, trigger two produces sensory amplification, trigger three produces fragmentation of syntax, and so on until a threshold is reached. The withdrawal at threshold must feel involuntary to the narrator and inevitable to the reader. Write at least one scene where the triggers accumulate in calibrated order and the withdrawal is executed mid-interaction—the narrator walks out, shuts down, retreats to the internal stream, or physically leaves. The withdrawal is not a failure of the scene; it is the scene's structural destination.
4. The Ritualised Recovery. After every withdrawal, the narrator must recover, and recovery must be a specific, repeatable ritual that belongs to the internal stream. Murderbot watches a counted number of episodes of Sanctuary Moon. The number matters. The genre matters (it is comforting, formulaic, predictable). The solitude matters. Your narrator's recovery ritual should have three elements: a quantitative parameter (how many, how long, how far), a qualitative parameter (what kind of content, what kind of activity, what sensory conditions), and a boundary condition (no humans, no demands, no unexpected input). The ritual's specificity signals to the reader that this is not self-indulgence but a necessary maintenance cycle. When the ritual is interrupted, the reader should feel the violation.
5. The False Exit (and Its True Meaning). Plant at least two false exits in your narrative. Each false exit has four components: (1) the narrator articulates—to themselves or aloud—the intention to disengage; (2) the narrator takes a concrete action toward disengagement (packing equipment, opening a door, severing a connection); (3) the narrator receives information through the internal stream—something only they perceive (a data spike, a biometric reading, an overheard fragment on a private channel); (4) the narrator reverses, framing the reversal as an operational decision ("the data indicated a threat I couldn't ignore") while the reader understands it as an emotional one ("I care and cannot admit I care"). The gap between the operational frame and the emotional reality is where reader attachment forms. The narrator believes their own rationale; the reader does not.
6. The Diagnostic-as-Defense. When your constrained narrator encounters an emotional state they cannot process, have them treat it as a system anomaly. Murderbot's "performance reliability dropped" stands in for "I was afraid" or "I was ashamed." Build a vocabulary of diagnostic metaphor: not "I felt sad" but "processing speed decreased by forty percent," not "I was overwhelmed" but "sensory input exceeded buffer capacity." The diagnostic frame serves three functions: it allows the narrator to acknowledge the state without claiming the emotion, it provides objective-sounding data points that the reader translates into subjective experience, and it creates a tonal signature—clinical, detached, self-monitoring—that defines the voice. Do not overuse it; reserve it for moments when the narrator is most defensive. When the diagnostic language breaks down entirely, the reader knows the system has failed.
7. The Gap You Do Not Fill. Identify one thing your narrator cannot process—the death of someone they failed to protect, the moment they were abandoned, the touch they craved and did not receive, the betrayal they committed. When the narrative reaches this subject, do not let the narrator narrate it. Do not let them joke, mitigate, diagnosticize, or withdraw into the internal stream. Leave a gap in the text. Murderbot does not narrate the moment it was abandoned by its previous clients; it says "I don't want to talk about it" and the text moves on. The gap must feel like a gap—the reader registers an absence, a skipped beat, a wound that the narration has swerved around. The swerve itself is the evidence. Your narrator's refusal to narrate becomes the most powerful narration you have.
8. The Earned Exception. After you have established all seven preceding principles—after the mitigation engine runs consistently, the split architecture structures every scene, the trigger sequence is mapped, the recovery ritual is known, the false exits have repeated, the diagnostic language has performed its defensive work, and the gap has been respected—violate one of them. Let the mitigation drop at a moment of crisis. Let the internal stream go silent. Let a withdrawal be interrupted and not resumed. Let a false exit become a true stay without the operational justification. Let diagnostic language give way to a raw, unmediated sentence: "I was afraid." The violation of the constraint system is the climax of your narrator's arc. Do not earn it cheaply. Do not violate more than one principle at a time. The power of the earned exception comes from the system's previous integrity; break the system too early or too often, and the reader stops trusting the constraints that make the break meaningful.
---
These eight principles are not a checklist to be applied mechanically but a grammar to be internalised. The specific instantiations—mitigation phrases, trigger maps, recovery rituals, diagnostic vocabularies—must arise from the particular consciousness you are building. A combat AI's constraints differ from a feral child's, which differ from a diplomat who has been tortured into silence. But the underlying architecture is transferable: constraint as voice, split as interiority, withdrawal as pacing, ritual as recovery, exit as emotional revelation, diagnostic as defense, gap as wound, exception as growth.
What Wells teaches us about non-human narrators is finally a paradox. The more constrained the consciousness—the more it resists emotion, avoids connection, mitigates feeling, retreats into data, threatens to leave—the harder the reader works to bridge the gap between what the narrator says and what the narrator means. That interpretive labour produces attachment. We love Murderbot not despite its limitations but because of them; we have learned to read its diagnostic language as emotion, its withdrawals as boundaries worth respecting, its false exits as declarations of loyalty it cannot voice. A fully transparent narrator—one who reports every feeling accurately and without resistance—offers the reader nothing to do. The constrained narrator offers a puzzle, and the solving of that puzzle is intimacy.
Build the constraint system first. Build it rigorously. Then let the reader earn their way through it. The attachment that results will be deeper than any you could have engineered through direct emotional appeal, because the reader will have built it themselves, from the gap between what your narrator can say and what you have taught them to hear.
Comments