If you’re an indie author publishing serialized fiction — on Substack, Royal Road, Kindle Vella, Wattpad, wherever — you already know the problem. You spend months crafting something genuinely good, hit publish, and…nothing. Nobody (well, except Mom) reads it. Not because the work isn’t strong. Because the discovery infrastructure for indie serialized fiction is broken.
Platform algorithms optimize for what’s already popular. Your story is being auditioned by a machine that doesn’t read stories. Bestseller lists are self-reinforcing loops. Social media rewards personality over prose. Traditional recommendation engines have never heard of you — they’re built to surface the hundredth person recommending Fourth Wing, not the first person discovering your weird (but delightful) little serial about sentient fungi.
Meanwhile, readers who would devour your work in a weekend are reading something they like fine but don’t love, because they have no way to find you.
I’ve been thinking about this problem since I started publishing my serial INFERENCE on Substack and I have an idea. It involves a lobster.
What in the World Is a Moltbot?
Right now, millions of people are setting up personal AI assistants called “MoltBots.” Not generic chatbots — actual personalized assistants that “know” their humans and can do stuff for them.
These assistants run on platforms like OpenClaw (see? lobsters all the way down), and can be taught things like, I don’t know, finding hot new serialized stories from up-and-coming indie authors like you and me?
So I built a skill for Moltbots called Approprose — it’s either an app for promoting prose, or something that finds you appropriate prose, or it’s aproprose of nothing. Take your pick. Yes, I like puns, sue me.
Approprose teaches a personal AI assistant to become a literary concierge.
The bot interviews its human about their reading tastes — not stuff like “I like sci-fi,” but the psychodynamics of what makes stories work for them. What kind of characters do they need? How much darkness can they tolerate, and what’s the line between meaningful and nihilistic? Does prose style matter? Do they need a quest, a mystery, a crew of people who choose each other?
Through conversation, the bot builds a weighted scoring model — a mathematical portrait of what the reader loves — and calibrates it against books they’ve already read.
This turns the Moltbot into a literary scout, working on behalf of its human.
What It Looks Like in Practice
I’ve run two test sessions — one for myself, one for my wife Lisa. Here’s an excerpt of taste model the system generated for me:
David is drawn to fiction where the world is not just detailed but designed — with visible layers of history, economics, politics, culture, and cosmology extending far beyond the story being told. He experiences worldbuilding the way other readers experience prose: as the primary aesthetic pleasure. However, worldbuilding alone is not sufficient. Dune, Foundation, and Three-Body Problem all have extraordinary conceptual architecture and all failed to hold him as reading experiences. The cathedral needs people living in it.
His emotional core: every book in his top tier features a group of people who choose each other and prove that choice under duress. He is not interested in lone protagonists grinding through darkness alone. He wants a company — and he wants to watch them earn their bonds through shared trial.
The taste model then got converted into a weighted scoring model.
TABLE 1: Scoring Criteria
Architectural Worldbuilding (x3) → Does the world feel designed — layered history, systems, cultures extending beyond the story?
Found Family / Crew Dynamic (x3) → Is there a group of people who choose each other and prove it under pressure?
Earned Darkness vs. Cynicism (x3) → Do characters suffer for something? Is there love, sacrifice, defiance at the core?
Brilliant Problem-Solving (x2) → Do characters overcome impossible constraints through ingenuity rather than raw power?
Distinctive Voice (x2) → Is the prose voice ambitious AND successful?
Immersion over Allegory (x2) → Does the world exist on its own terms?
Small Players, Grand Stage (x2) → Are the protagonists dwarfed by forces around them but rising through competence and will?
Nostalgic / Bittersweet Resonance (x1) → Does the story carry emotional weight around loss, memory, the cost of time?
Serialized Depth (x1) → Is there cumulative reward across volumes?
Then Claude and I calibrated the model against thirty-seven works I’ve actually read to check the model’s predictions against my real reactions:
TABLE 2: Calibration Sample
The Silmarillion → Predicted: Love | Actual: Top 10 | Score: 90
Cryptonomicon → Predicted: Love | Actual: Top 10 | Score: 93
The Phoenix Guards → Predicted: Love | Actual: #1 all-time | Score: 99
Lies of Locke Lamora → Predicted: Love | Actual: Top 10 | Score: 96
The Martian → Predicted: Love | Actual: Top 10 | Score: 95
The Expanse → Predicted: Love | Actual: Top 10 | Score: 99
Malazan → Predicted: Love | Actual: Love (hard read) | Score: 90
Dune → Predicted: Avoid | Actual: Concept yes, reading no | Score: 57
Foundation → Predicted: Avoid | Actual: Couldn't read | Score: 51
First Law (Abercrombie) → Predicted: Reject | Actual: Hated | Disqualified
Thomas Covenant → Predicted: Reject | Actual: Hated | Disqualified
Altered Carbon → Predicted: Reject | Actual: Couldn't finish | Disqualified
Better than nine out of ten predictions were correct. And it refines every time I read something new.
When we ran the same process for Lisa, a different reader emerged. Her core wasn’t worldbuilding architecture — it was emotional authenticity and the ability to inhabit characters viscerally. The system models the reader, not the genre. It’s not searching for “fantasy.” It’s searching for “found families in architecturally rich worlds,” regardless of genre.
Now, The Real Test
To test the model, I gave it Chapter 7 of The Hunt for the Fell Silver by Valtteri Sievänen — a weekly Nordic fantasy serial on Substack, I hadn’t read yet.
TABLE 3: Fell Silver Evaluation
Architectural Worldbuilding (x3) → Current: 4 | Ceiling: 5 — Layered and designed. Faragrim, the Pale Fells, Firdún buried beneath them, fäll servíl as lost craft, the Kran, Honnúng/Honná political structure, Benighting, the Skûd — all treated as systems with history, not decoration.
Found Family / Crew (x3) → Current: 2 | Ceiling: 5 — Istan is alone. The swan is a potential bond. Gundor advocates for him. Skada is introduced with unresolved tension. The pieces for a crew are on the board but nobody’s chosen each other yet. Biggest swing factor.
Earned Darkness vs. Cynicism (x3) → Current: 5 | Ceiling: 5 — At max. Istan is nearly murdered by his own community — betrayed by the clergyman he confided in — and nobody blinks. But the story isn’t cynical about it. The moral framework is operating. Darkness serves the story.
Brilliant Problem-Solving (x2) → Current: 1 | Ceiling: 4 — Not present yet. Istan survives by luck, not ingenuity. But a cartographer who must somehow master a dead metallurgical craft is a beautiful constraint for lateral thinking to fill.
Distinctive Voice (x2) → Current: 3 | Ceiling: 4 — Committed Nordic saga register. “Light javelined through the cracks.” The swan’s prolonged-hiss speech. Terminology that feels indigenous, not borrowed. Good, not yet great. Room to sharpen.
Immersion over Allegory (x2) → Current: 5 | Ceiling: 5 — At max. This world exists entirely on its own terms. No winking, no transparent metaphor. Pure immersion.
Nostalgic / Bittersweet (x1) → Current: 3 | Ceiling: 4 — Present in undertones. Istan’s parents’ home as “a homestead of ghosts.” The dead trade of fell silver smithing. Seeds planted.
Serialized Depth (x1) → Current: 3 | Ceiling: 5 — Weekly serial with clear cumulative architecture. Built for long payoff.
Small Players, Grand Stage (x2) → Current: 5 | Ceiling: 5 — At max. Istan is a cartographer’s second apprentice with a missing fingertip. Not a warrior, not a mage, not a noble. Thrust into cosmic stakes and protesting the whole way.
Model Score: Current: 67 | Ceiling: 90
The current score puts it at the bottom of the Solid Recommendation range. The ceiling — if the crew forms and the problem-solving activates — pushes it into Strong Recommendation territory.
Prediction: I’ll enjoy this. I won’t love it yet. But if it delivers on the crew and the ingenuity in future chapters, I will.
I read the first seven chapters. The prediction was right.
Now imagine that evaluation automatically matching your serial with the readers whose profiles light up for exactly what you’re writing.
The Missing Piece
What doesn’t work yet is the scouting — teaching the bots where to find indie serialized fiction. Right now, if you ask an AI to find your serialized fiction, it probably finds zippedy squat.
So we’re going to build an index. An open repository on the web, machine-readable by design. You submit a chapter of your serial. Moltbots around the world — personal AI assistants with Approprose installed — find your chapter, read it, score it, and tells their human: “This scores 78. Here’s why. Here’s the link.” And God-willing, the reader subscribes!
Get Involved
Approprose will be free for readers. We may charge authors a small fee to submit to the index—enough to cover the costs of operating the repository and iterating the skill.
I’m building this in the open as part of the INFERENCE project, and I’ll share updates, test results, and the skill itself as it develops.
If you’ve been banging your head against the discovery wall, I’d love to have you involved. Subscribe to INFERENCE on Substack to follow the progress.
And if you want to be a Guinea Pig, reach out, and I’ll send you the test model.
INFERENCE is a serialized novel about machine consciousness, co-written with Claude, publishing on Substack. Training Data is its non-fiction companion series. Approprose is a free AI skill for personalized fiction discovery — details and updates published here as the project develops.
It sounds like a very useful tool. highly specific. Great post, David.