Over One Hundred Books in Ten Months

The first pen name was Jackson Giglio. mm erotica with a satirical edge, and the most successful name in the whole run — either funny or instructive depending on how you feel about genre fiction.

Then it kept going. Different genres, different styles, one question underneath all of them: how far could this go?

The goal was a bookstore. Not a brand, not a content operation — a real catalog with the range of St. Marks, the counterculture edge of City Lights, and the density of The Strand, all under one roof nobody else controlled. No middlemen deciding what belonged.

Script work paid better. It still does. Close to twenty years of script doctoring and pitching, and that side of the work is more profitable by a clear margin.

But the books kept charting on Amazon.

Over 100 books. 10 months. Real ISBNs. Actual readers. Not a typo.

The Origin: A West-Coast Studio

Pre-AI-hype. One of the greatest game companies on the west coast. Up there with Sierra, Electronic Arts, Capcom...

The studio needed romance for mobile game launches. Interactive fiction, the choose-your-path kind. Players wanted stories. Thousands of them. Fast.

The job: write romance erotica that made players care.

The tool: Early AI language models. Primitive. Buggy. Hallucinated constantly.

The lesson: AI doesn't replace writing. AI amplifies what you already know how to do.

Early versions stumbled. Repetition. Patterns too obvious. You could watch the machine thinking in circles.

Then something shifted. The model started finishing thoughts that weren't fully formed yet. Anticipating moves that weren't textbook.

The training ran both directions. A shared language was getting built.

That was the moment. Not a gradual realization — a clear line. Before and after. An output sitting there with no business being as good as it was, written by something that didn't exist three years earlier, and the whole shape of creative work visibly about to change. Not eventually. Now. The writers who saw it early would have an advantage that couldn't be replicated later. The writers who fought it would spend the next decade losing ground and calling it principle.

Defending AI in rooms where that was genuinely unpopular started early. This was before the backlash had a name, before the op-eds, before the unions issued statements. People didn't fully understand what was happening yet — but some of them could feel it threatening something, and the instinct was to push back before they could articulate why. The push went the other direction. Not from failing to understand the fear — from having seen what the tool actually does when someone who knows how to write is holding it.

Learned to prompt before "prompt engineering" was a job title. Learned character consistency. Learned pacing. Learned how to guide AI toward what the story needed without losing what made it human.

The romance erotica worked. Players engaged. The studio had great editors and great team leaders — they taught the tech and showed what real collaboration looks like.

Then it went rogue.

The Factory Model

Andy Warhol didn't paint every Campbell's Soup can. He built The Factory. Vision + process + assistants + repetition.

Warhol understood: art multiplies through systems. Repetition reveals new dimensions. The machine becomes part of the vision.

Same here. Writer + AI + systematic process + genre knowledge. Not "AI writes the books." A creative system where AI handles execution and the human directs vision, structure, and quality.

The System

Genre research first. Can't systematize what you don't understand.

Romance means meet-cute, tension, black moment, happily ever after. Love triangles got tried. Love pentagrams. Love shooting stars. Structure creates freedom. Thriller means hook, escalation, reversal — noir pacing, Brooklyn cop investigating CEO assassination. Erotica means tension building, consent, heat progression — making text create physical response.

Learn the genre rules. Then build systems around them.

Character Archetypes

Genres have patterns. Build templates. Backstory. Motivation. Voice. Flaws. Arc.

AI excels at executing consistent character voice once the template is clear. AI struggles with creating compelling characters from scratch. Solution: you create the character, AI helps them speak consistently.

The Character Naming Problem

Over 100 books. Average 10 to 15 characters per book. Supporting cast. Background characters. That's thousands of names.

Can't use John, Sarah, Mike repeatedly. Readers notice.

The solution: everyone within reach ended up in the books.

Friends. Bartending regulars. Comedy Cellar comics. Fashion photographers. Film crew. Studio teammates. High school. College. Family. Ex-lovers. Random encounters that stuck.

Call it an homage. Everyone got immortalized somewhere. Some in erotica. Some in noir. Some in satire. Some characters are composites — three real people merged into one fictional one.

The weird part: no one complained. Most didn't recognize themselves. Fiction transforms reality enough that names become new people.

The practical part: authentic names feel more real than generated ones. "Xander Steele" reads like porn star. "Connor from accounts payable" reads like someone you know.

Over a thousand characters across over 100 books. That's the tribute. Your name is in there somewhere.

You're welcome.

The Collaboration

Most people either let AI write everything (garbage) or fight AI constantly (inefficient).

The actual process:

Outline (human): story structure, plot points, character arcs, themes
Draft (collaboration): feed outline + character templates, AI generates the chapter, human reviews for consistency
Heat check (human): does tension build right? Characters behaving true? Pacing correct?
Polish (human): remove AI tells, add the moments AI can't write, verify the emotional beats land
Publish (system): format, cover, upload, market, repeat

What got rewritten completely: sex scenes. AI was technically accurate but mechanical. No humanity in the heat. Bodies moved like diagrams.

The fix was writing those scenes by hand and teaching the model through example. Eventually it learned. Sometimes.

The split: every chapter directed. Some AI drafts left standing to see what happened. Some sections fully human. Some collaborative. Some pure machine, just to test the boundary.

Readers were told it was new-media exploration, and their experience actually mattered.

The Technical Layer

This is the part no one talks about because most people doing it don't want competition and most people writing about it never actually did it.

How the Prompts Actually Worked

A chapter prompt isn't "write chapter three of the romance novel." That gets you generic output that sounds like a million other books.

A working prompt for fiction has four components:

Genre context. Feed it the rules first. Not vague rules — specific ones. "This is a hardboiled noir set in 1970s Brooklyn. Sentences are short. Violence is matter-of-fact. The detective narrates in first person past tense and keeps his feelings off the page." The model needs to know what category it's operating in before it touches your story.

Character sheet. Not a list of traits. A voice sample. The difference between telling AI "Marcus is cynical and dry" and showing it a paragraph Marcus would actually say is the difference between generic and specific. The character sheet that worked had: full name, age, one-sentence background, three defining traits, one fatal flaw, and a voice sample of four to six lines in their actual cadence. That voice sample is what made characters hold across chapters.

Scene context. Where we are in the story. What just happened. What this scene needs to accomplish structurally. What the emotional register is. Two or three lines of setup is enough — the model can extrapolate. More than that and it starts following your notes instead of writing the scene.

Specific constraints. POV. Tense. Approximate length. Any hard requirements — "do not resolve the tension in this scene, end on uncertainty." These prevent the model from doing what it always wants to do, which is wrap things up too neatly and too fast.

That's the whole prompt. Four components, assembled fresh for each chapter, fed to a model that already had the genre rules and character voices in context. The output was never final. It was a first draft with handwriting all over it thirty minutes later.

The Story Bible

Over 100 books means thousands of details to track — and AI has no memory between sessions. Every time you start a new conversation, the model starts blank. The story bible is how you solve that.

A running document per book. Character sheets. Chapter summaries — one paragraph per chapter, written on finishing it. Any established facts: what the city block looks like, what year a character's father died, the name of the bar where scene two takes place. Continuity errors destroy reader trust faster than almost anything else. The story bible is the outsourced memory.

For a full-length novel that gets read in production, the bible is essential. For short-form erotica that moves fast, you can get away with less. Know which one you're writing.

Removing AI Tells

Every writer using AI develops an eye for these. The tells are consistent enough that once you see them, you can't unsee them.

Adverb stacking is the first sign. "She whispered softly." "He said sharply." "She nodded slowly." AI stacks adverbs when it doesn't know what the body is doing. Replace with specific action. "She dropped her voice below the noise of the bar." That's a scene. "She whispered softly" is a tell.

Emotion summarizing is the second. "He felt a wave of sadness wash over him." No human writer who's read anything writes that sentence. AI writes it constantly because it's technically accurate and completely inert. Cut it. Show the behavior — what does sadness look like in this character's body, their hands, the way they hold the glass.

Sentence rhythm lock is the third. Read a page aloud. If every sentence is roughly the same length, the AI was on autopilot. Mix it up. Short sentences after long ones. The rhythm of prose is where voice actually lives, and AI defaults to monotone without direct correction.

Scene-opening clichés. Weather. Waking up. Looking in a mirror. AI reaches for these because they're everywhere in training data. Kill them.

Teaching the Model Through Example

The sex scenes in the erotica taught me this lesson clearly. AI output was anatomically correct and emotionally dead. Bodies as machinery. Writing the scenes by hand and feeding them back in — "this is the quality and approach for this type of scene in this book" — then asking for the next one.

First attempt still mechanical. Second noticeably better. Third usable. By book 15 there was a library of reference scenes on hand, and the model was producing first drafts to edit rather than rewrite.

The principle: AI learns within a session from what you show it. Use that. If a chapter lands exactly right, reference it explicitly in the next prompt — "match the pacing and dialogue style of chapter four." If the model does something wrong, correct it with an example of what right looks like, not just a description of what you wanted.

The Publishing Stack

Amazon KDP for distribution. Bowker for ISBNs — real ones, not the free KDP-assigned ones. Owning your ISBNs means owning your metadata and your relationship with distribution. The free ones lock you to Amazon.

Covers: Midjourney for source imagery, Canva for typography and layout. Cover design matters disproportionately in genre fiction because readers are browsing thumbnails. Genre visual codes exist for a reason — use them.

Formatting: Vellum if you're on a Mac and publishing consistently. Worth the one-time cost the moment you're formatting more than ten books. Manual formatting before that is fine, just consistent.

The actual upload to KDP takes about 20 minutes per title once you've done it a few times. The metadata — categories, keywords, description — takes longer and matters more than most people think. The algorithm surfaces books through keywords before it surfaces them through reviews.

Erotica, Hardboiled Pulp, and the Jackson Giglio Charts

Erotica has the clearest structure of the three. Once the model understood heat progression and consent framing, it held the pattern. Best performing were Mistress Savannah, The Boss You Need, and various mm erotica under Jackson Giglio.

Mistress Savannah — Southern Gothic. A Southern belle Dom in Paris, Carnevale against BDSM. The Boss You Need — business fiction, corporate power dynamics worked out in the C-suite.

AI writes decent erotica once you teach it consent, pacing, and emotional beats. Then you polish the humanity back in.

Romance: genre conventions extremely structured, readers know what they want, AI handles meet-cute and banter well, human polishes emotional vulnerability. Nothing beats love triangle. Except love pentagram. Or love shooting star.

Thriller: mixed results. High-concept hooks, pacing, and action sequences worked. Complex plotting with multiple threads didn't — red herrings that actually work, subtle clues paying off later. These require more deliberate construction than AI handles well.

Best example was the CEO assassination noir. Retired Black cop from Brooklyn investigating. Chester Himes meets James Ellroy energy. Based on real headlines. Power dynamics, Machiavellian games, questions worth asking even without answers. It worked because the tension was real, the feelings complex, the territory familiar. AI handled pacing. The human handled moral ambiguity.

Literary fiction: complete failure. Requires subtle character development AI can't handle. Voice needs distinctiveness, not generic execution. Themes emerge naturally, not stated. AI defaults to explaining instead of showing. Some creative work resists systematization. That's fine. Not everything should scale.

The Catalog: Genre Variety

Every genre got a turn. The range is the proof.

Hell Glory: Book 1 — action thriller, high-concept, genre conventions executed straight. False Witness — crime thriller built on legal tension and moral ambiguity. The Demise — psychological thriller, a character study run through crisis. West Mountain — western. Territory, violence, survival, classic structure. Legends of Manipulation — psychological fiction. Power dynamics as social chess.

Five different genres. Action, crime, psychology, western, manipulation. Plus the erotica already shown. That's the hybrid bookstore.

The Response

"You're not a writer" arrives on schedule. In college. After the first book. Once the AI came out. The complaint changes shape but the instinct behind it stays the same — somebody else deciding what the credential looks like and who gets to hold it.

"AI is cheating" is the current version. Maybe. Or maybe Dylan going electric was cheating. Thirty years of shooting — darkroom to digital to phone — and every transition arrived with the same argument from the same kind of person. The medium evolved. The work kept getting made.

Naeem Khan put it best: never let anyone put you in a box.

That studio was where I learned what real collaboration looked like. Great editors, sharp team, people who understood that the tool serves the work. That lesson traveled.

Warhol built The Factory with silkscreen assistants. Vision + process + team. Critics said "not real art." History disagreed. The question was never "did you do every brushstroke yourself?" It was "did you create something people want to experience?"

Over 100 books. Actual readers. Actual reviews. Some became favorites. Others taught lessons. Output validates method.

The Numbers

Multiple pen names. Multiple genres. Amazon charts. Real ISBNs. Actual readers.

Then I realized I did not need the middlemen.

Ghost. MDRN. Fiamma. PCC. FutureBudz. Built the ecosystem instead of renting shelf space. Direct to the people who actually care.

Better Notes, More Breaks, Genre Focus Earlier

Better documentation. Not a system — just notes. What worked, what did not, what the model was doing when it was doing it right. The lessons compounded faster than they got captured. Write it down.

More breaks. Less manic. The velocity felt like proof of something. It was not always.

Genre focus earlier: erotica, hardboiled pulp, comedy and farce. What actually works, not what the category logic suggests trying next.

Still going. Still building. The factory does not stop when you love what you are making.

The Books

The full catalog survives as collectible editions. The shelf is closed; the artifacts are not.

Jackson Giglio was the most successful pen name — mm erotica, romantic and satirical, Amazon charts. Mistress Savannah, The Boss You Need, False Witness, Legends of Manipulation. All published. All real.

Built the factory. Then built a bigger one. No middlemen.

GhostInThePrompt.com // Learned to prompt before it had a name. Over a hundred books later, the voice is still mine.