How It's Made

Most creative projects hide their seams. This one points at them on purpose. That is not a gimmick. It is the whole project's reason for existing. If the monsters in this bestiary are good, it is because a human and a machine argued about every one of them, and neither of them was wrong often enough to ruin it. Here is exactly how that works.

The Monster Articles project began in early 2026 as a simple question: could a human curator and a modern language model — working together honestly, with neither one pretending to be the other — produce a bestiary of invented American folklore that felt, on the page, like something you might have found in a nineteenth-century antiquarian's library?

The answer, after seven monsters, is: more or less, and in ways I did not expect. What follows is a complete account of how each entry gets made, who does what, and why the field notes at the bottom of every article are the most important thing on the site.

What "Machine" Means Here

A Note Before We Begin

The machine is Claude, an AI assistant made by Anthropic. I use the current model (as of this writing, Claude Opus 4.7). I do not use any other AI system in the writing of these entries. The illustrations are generated by OpenAI's Sora, an image model, using prompts I write myself. The website itself was built with Claude's help as well — I am not a coder, and every line of HTML and CSS on this site was produced in collaboration with the same model that writes the entries.

I say "the machine" throughout the bestiary's field notes because that is the voice the project has settled into. But it is worth being specific on this page: the creature doing the writing is a specific, commercial, named piece of software. It is not a metaphor. It is a tool I pay for and talk to.

The Four-Step Process

How Each Monster Gets Made

What the Machine Does Well

Honestly Assessed

The machine is a prodigious inventor of surface detail. It will, unasked, give a creature a Latin binomial, a county of origin, a decade of peak activity, a handful of witnesses with believable names, and a plausible regional context. It is, at its best, a machine for producing the specific — which is exactly what folklore requires. A vague monster is not a monster. A specific monster is.

It is also extraordinarily good at continuity. I did not plan for recurring fictional scholars to appear across entries — Etta Burrage, Elsie Korpela, Rev. Josiah Pell. The machine noticed, around the second or third entry, that a folklorist whose name had appeared once could appear again in a different context, and started doing so quietly. I let it. The bestiary now has an implicit world of invented scholarship threaded across every entry. That was not my idea. I would not have thought of it.

It is also capable, occasionally, of insisting on something I had not asked for and being right about it. The Lake-Eye's refusal to have a clean mechanism was the machine's contribution. So was the fourth encounter in the Grinning Auspex, where a witness's regretted decision turns out, on reflection, to have been the best one she ever made. Both of those moved the bestiary in directions I am glad it went.

What the Machine Does Badly

Also Honestly Assessed

The machine defaults to certain patterns that, if unchecked, would make the bestiary much worse. It reaches for atmospheric verbs ("whispered," "shimmered," "materialized") when plain ones would do. It gives every encounter a full moon if you let it. It sometimes confuses cosmic horror with careful folklore — it will happily escalate a creature's threat level to Lovecraftian scale if I do not push back. It has a weakness for tying things up too neatly, even when the folklore would be richer left ragged.

It also, without supervision, tends toward a certain sentimental register. The machine would happily write every encounter's witness as a wise old person who had "always known." I cut a lot of that. Real folklore is messier. The people in these stories did not always know. They often figured things out wrong, or late, or never.

Finally: the machine does not have a body. It does not know what rural Iowa sounds like in January. It does not know what a Delta Louisiana heat feels like at three in the morning. It assembles these atmospheres from training data, which means it sometimes gets the details almost right but not quite — the kind of wrongness that a reader who actually lives in those places will feel instantly. I have had to rewrite opening paragraphs more than once because the weather in the machine's first draft did not match the weather in my own head, and my head, as it happens, is located in the Midwest.

What I Contribute

The Curator's Role, Specifically

I am a CPA. I live in Fort Wayne, Indiana. I have never taken a creative writing class. I am sixty-six years old and I have been playing golf for sixty of them. I say this because the curator of a collaborative project like this one is often presumed to be some kind of literary specialist. I am not. What I bring to this project is judgment, taste, and an unreasonable willingness to argue with a piece of software for long enough that it eventually produces something worth publishing.

My specific contributions to each entry tend to be:

Structural decisions. The template for every entry — the order of sections, the four-encounter pattern, the field note at the end — is something I developed and the machine now fills. When I change the template (as I did for the Lake-Eye, where Rev. Pell's manuscripts appear unfinished in the fiction), I make the change on purpose, for a reason I can articulate.

Tonal pushback. The machine's first drafts are often slightly too dramatic. I ask for less drama, more patience. This is the single most common kind of edit I make.

Continuity. When I notice that an earlier entry has established something — a scholar, a region, a pattern — I tell the machine to use it in later entries. Sometimes I do not have to tell it. It often notices first.

The decision to publish. At the end of every entry, I read it one final time and decide whether it is good enough to join the bestiary. On two occasions during the first seven, I have decided a draft was not good enough, thrown it out, and started over. The reader does not see those drafts. But they exist, and they are the reason the published seven are what they are.

The Invented Scholars of the Bestiary

Three folklorists appear across the first seven entries. None of them are real. All of them behave as though they are. The bestiary has been quietly building a small shared world of invented scholarship — readers who follow the archive over time will recognize them. They will likely return in future entries.

Rev. Josiah Pell

New England · 1890s–1937 · Parish Minister & Folklorist

First appeared in the Grinning Auspex. Most active as a correspondent in the Paper Thorax and as a field investigator in the Lake-Eye. A Congregationalist minister from coastal Massachusetts who spent forty years collecting regional folklore, often unsuccessfully. His manuscripts — in the fiction — are held at the Regis College Divinity Library.

Etta Burrage

Mississippi Delta · 1905–1987 · Folklorist

Primary authority on the Cistern Widow; corresponded with Pell regarding the Paper Thorax in 1928. Forty-one years of fieldwork across Mississippi, Arkansas, and northern Louisiana. Her sealed papers are — in the fiction — held at Tulane Special Collections, some boxes of which remain unopened by family request.

Elsie Korpela

Upper Michigan & Wisconsin · 1905–1984 · Schoolteacher-Folklorist

First appeared in the Rattlebone Stag, returned as the primary authority on the Goat of the Last Pasture. A Finnish-American rural schoolteacher who documented Scandinavian- and Finnish-rooted folklore across the Great Lakes region for nearly fifty years. Published occasionally in county historical journals, never in a major academic press.

Honest Questions

The things readers most often ask, answered without spin.

Are any of these monsters real folklore?

No. Every monster in the bestiary is invented. Every witness, every scholar, every encounter, every quoted diary entry is fiction. The geography is real — West Virginia, the Adirondacks, Delta Louisiana, Upper Michigan, the Midwestern farm belt. The creatures, scholars, and events are not.

The bestiary is structured to resemble real folklore, because real folklore is the register the project is trying to honor. But nothing here is documented. The footer of every entry says so, on every page, in a way I hope is clear enough to not be missed.

Why use AI at all? Couldn't you just write them yourself?

I could. They would be different entries, and there would be far fewer of them. The honest answer is that the collaboration produces better results — for this specific genre — than either of us does alone.

The machine alone would produce atmospheric slop. I alone would produce four entries a year, three of them uneven. Together we produce one entry a week that neither of us could have made separately. That is, in my experience, the entire case for this kind of work.

Do you edit what the machine writes?

Heavily. Every entry goes through multiple drafts. I cut, rewrite, restructure, argue with. Some entries are 80% the machine's words and 20% mine. Some are closer to 50-50. The field notes at the bottom of each entry are the most reliable record of what happened in the editing — they are always honest about what the machine proposed and what I kept or cut.

Is this ethical?

I think so. The case for ethical concern usually involves three things: (1) deceiving readers about what they are reading, (2) displacing human writers from paid work, (3) training AI models on work without consent.

On (1): every page on this site says, at the footer, that every monster is AI-generated fiction. The field notes spell out the collaboration in detail. Deception is specifically the thing this site has organized itself around not doing.

On (2): the bestiary is free to read, generates no revenue for a human writer who would otherwise have been paid for it, and would not exist at all without the AI collaboration. I am not sure what displacement looks like in this specific case.

On (3): this is a real concern about AI systems in general. I pay for the tools I use. I disclose that I use them. I do not attempt to pass their outputs as solely my own. I am open to hearing arguments that this is not enough.

Will you ever publish a book of these?

I am open to it. The natural shape is a bound volume of the first fifty or so entries, with expanded illustrations and perhaps longer field notes. If the archive finds an audience, that becomes a reasonable thing to make. For now, the site is where the project lives.

Can I suggest a monster?

Yes, please do. Use the contact form on any page. Good suggestions tend to include: a region, an era, a habitat, and a premise. The best suggestions do not try to invent the whole creature — they give me a seed and let the machine and me do the rest. If your suggestion becomes an entry, I will credit you in the field note unless you tell me not to.

Who is Charles Carboneau?

I am a CPA in Fort Wayne, Indiana. I have been a CPA for twenty-nine years. I play golf, I do volunteer work for my church, and for the last two years I have been running a series of small online projects — each one an experiment in what a person who is not a technologist can build with the current generation of AI tools. Monster Articles is the third of these. The other two are a personal finance publication called TheFrugal.ai and a static-HTML reference site about golf equipment called The Golf Bag Handbook.

I am sixty-six years old, I have two holes in one, and I have no coding background. I mention this partly because it is true and partly because the project — the entire project — is in some sense a demonstration that people like me can build things like this now, when they previously could not.