Almost Timely News: 🗞️ How To Force AI to Write More Like You (2026-05-10)

There is no such thing as writing style

Christopher S. Penn

May 10, 2026

Almost Timely News: 🗞️ How To Force AI to Write More Like You (2026-05-10) :: View in Browser

The Big Plugs

New things!

1️⃣ Job to AI Skill and Plugin

2️⃣ My new GEO 101 course

Content Authenticity Statement

75% of this week’s newsletter content was originated by me, the human. You’re going to see many different writing samples from AI in the main content and in the appendix. Learn why this kind of disclosure is a good idea and might be required for anyone doing business in any capacity with the EU in the near future.

Watch This Newsletter On YouTube 📺

Click here for the video 📺 version of this newsletter on YouTube »

Click here for an MP3 audio 🎧 only version »

What’s On My Mind: How to Force AI to Write More Like You

This week, let’s talk about writing with AI in more detail. I had a great time presenting at the SmarterX AI for Writers Summit, but we barely touched any of the questions submitted in my session. For simpler questions, I answer these either during my daily videos on my YouTube channel (10 minutes or less), or as Overtime Questions on the Trust Insights YouTube channel (1 minute or less).

But a few questions deserve more airtime than that, so that’s what this week’s issue will cover, the big stuff. The biggest question that people had throughout the entire summit was, “how do I get AI to write more like me?”

Part 1: There Is No Such Thing as Writing Style

We’ll begin with one of my favorite rants: there is no such thing as writing style.

Take a moment to consider for yourself how you would describe your writing style. If someone asks you to explain your writing style, what do you say? How do you describe it?

Chances are you say something circular or self-referential like “I write like me!”, Because we often don’t have the vocabulary to describe the components of writing style. Or if we do, we learned them a very long time ago and forgot them.

Writing style is an umbrella term that describes dozens of different descriptive characteristics of how we write, such as:

Vocabulary and diction (word choices)
Levels of formality
Jargon and slang
Syntax (order of words)
Sentence length
Sentence structure
Punctuation (hey em dash!)
Parallelism
Tone
Register
Mood
Irony
Sarcasm (yeah, right)
Voice
Point of view
Rhythm and cadence
Pacing
Paragraph structure
Figurative language
Metaphors and similes
Imagery
Personification
Whitespace
Spelling (or lack thereof)
Rhetoric
Cohesion
Grammar
Text formatting (e.e. cummings would like a word for those who say it doesn’t matter)
Structural elements (hello listicles)

Depending on who you read, these are still only a subset of all the different ways to describe someone’s writing. Each of these characteristics sheds light about how we write, about how we put words on paper or keyboard.

This is one of the reasons why generative AI prompts such as “write in a professional style” are so deeply flawed. What does that mean? What is the average sentence length of professional versus non-professional writing? What is the register? What is the point of view?

Go back to that consideration of your writing style that I asked you about just a few paragraphs ago. Did you cover tone in your own mental description of how you write? Did you cover pacing or paragraph structure, imagery or rhetoric?

Unless we are recovering English majors, there’s a good chance we don’t ever think about any of these components of writing style when we write. We just do it in the same way we don’t consider walking style or eating style. However, when we are trying to give instructions to a tool like AI that has no prior knowledge of how we write, we have to be achingly specific in our descriptions of how we write, including all these elements we don’t normally think about.

This is also why prompts like “imitate my writing style” fall far short. Generative AI always uses the most common, highest probability concepts when it generates outputs. Consider this output from Google’s Gemini, when prompted to answer “What are the elements of writing style?”:

The elements of writing style generally include word choice (diction), sentence structure (syntax), voice, and tone. Other elements such as figurative language, rhythm, rhetorical patterns, and even punctuation also contribute to a writer’s unique style.

This is a woefully inadequate description of writing style compared to the list above, isn’t it? Is it any wonder when we tell AI to imitate our writing style that it goes for the most high level, generic components of writing style and fails to capture the nuance of how we write?

So AI does not understand writing style if we do not detail what writing style means and we do not define writing style for it. Which brings me to my next rant, that there is no such thing as AI writing.

Part 2: There is No Such Thing as AI Writing

One of the common refrains at the AI Writers Summit was that we have to work around AI’s writing style and “tells”, from em dashes to common sentence structures like “it’s not X, it’s Y”. Yes, different AI tools have those high frequency constructions, but they’re not universal.

Here’s a thought exercise for you. What is men’s writing style? What is women’s writing style? What is Koreans’ writing style? What is a bisexual’s writing style?

Besides being horrendously oversimplifying and possibly discriminatory, there is no good answer to these questions, because every person’s writing style is their own individually. To assume that a broad characteristic like gender or race or age are valid ways to evaluate someone’s writing style is incorrect and inappropriate.

So if we wouldn’t do that to people, why do we do that to language models? Every language model is like its own person, in terms of writing. They have their own unique writing characteristics. Gemini writes differently than GPT 5.5. Claude Opus writes differently than Alibaba Qwen.

Here’s an example. Given the prompt “Complete this opening paragraph of a romance novel with 2 more sentences: ‘Chris sat down at the corner table in the diner, far away from the other patrons. In the background, 80s pop songs played on a knockoff jukebox fueled by the line cook’s Spotify list. Then’”

Look how different the styles are:

Gemini: Then the door’s bell chimed, cutting through the synthesized beat as a figure stepped in, shaking a drenched umbrella. Chris looked up, and for the first time in years, the rehearsed indifference of a quiet life began to unravel.

ChatGPT: Then the bell above the door chimed, and Chris looked up despite himself, drawn by something he couldn’t quite name. The stranger who stepped inside carried the kind of presence that made the whole room feel smaller, as if fate had just taken a seat across from him uninvited.

Microsoft Copilot: Then the bell over the door chimed, and he looked up to see her framed in neon and rain, a familiarity he couldn’t place tugging at him. Their eyes met, and something warm and improbable bloomed in his chest, as if the song had been written for this exact moment.

Claude: Then the bell above the door chimed, and every nerve in Chris’s body went rigid—because there was only one person who could make them feel this way, walking through that threshold like they owned the place. When their eyes met across the crowded diner, Chris realized that no amount of distance or denial could change what had been building between them for the past six months.

Alibaba Qwen: the bell above the door chimed, and the woman he’d spent three years trying to forget stepped inside, shaking rain from her coat like she hadn’t just shattered his carefully constructed routine. The synth-heavy melody suddenly felt like a countdown, and Chris knew with a quiet, terrifying certainty that his carefully guarded solitude was about to end.

Each model has its own assumptions, its own diction, its own imagery. When we put these results next to each other, we see that they behave in the same way that people do. Each has its own distinct style. Copilot and Qwen make heteronormative assumptions. ChatGPT, Claude, and Gemini don’t. Claude busts out the em dash.

This is why tools like AI detection tools are utterly worthless. In the same way that there is no way to detect whether a woman or a man wrote something, there is no way to determine that AI as a broad category wrote something. You can evaluate to see if a specific model wrote something, but not the category overall.

How would you do that? That brings us to the next section on stylometry.

Part 3: Basics of Stylometry

Stylometry, unsurprisingly based on the name, is the science of measuring writing style. All language is code, in that not only do the words matter that we put on paper, but the order in which those words appear and the choices we make for which words to use are predictable and measurable.

If I say “I’m spilling the tea”, we know what that sentence means because in North American English the words occur in subject-verb-object order. I am the subject, the tea is the object, and spilling is the verb. There’s also jargon in there. In this case, spilling is not literal in the sense of me pouring a beverage all over the floor - “spilling the tea” is slang for gossip.

If I say “tea spilling the I’m”, this makes no sense whatsoever to us. Even though the words are exactly the same as the previous sentence, the order of them makes no sense. Can we reason out what it probably says? Yes, to a degree. But because the order is wrong, meaning is lost.

Stylometry takes word frequencies and positions and order to develop a mathematical representation of the written word. Because order is predictable, because we each write in our own way that has predictable sequences of words, we can measure how similar two pieces of writing are. There’s a sub-discipline in computer science called natural language processing, which uses classical AI, aka machine learning, to digest down text and represent it mathematically for comparison.

One of the current standards in stylometry is a metric that I shared in the summit called Burrows’ Delta, along with lesser used techniques like n-gram analysis, cosine similarity of embeddings, Jaccard similarity, Manhattan distance of z-scores, and dozens of others. That list that I shared in Part 1? That is a recipe for stylometry itself.

Stylometry measures all the different features of writing. Imagine you had two pizza chefs who were tasked with making a meat lover’s pizza. One chef tends to put more sausage on their pizza than the other chef. The other chef tends to put their pepperoni in straight lines. The first chef likes to mix their tomato sauce and mozzarella cheese together. The second chef always adds garlic powder on top of the meat.

Both chefs have their own specific way of making a meat lover’s pizza, and if you watch them make enough pizzas, you start to recognize their style. You can tell just by the distribution of sausage or the placement of the pepperoni which chef made each pizza.

That is stylometry in a nutshell. Looking at the words that someone uses, looking at how they use their punctuation, looking at vocabulary choices or readability - an author can be relatively easily fingerprinted using mathematical techniques like this, because all language is code, is predictable, and is measurable.

Nearly every major coding language has some library or package, usually free, that conducts stylometry analysis. The gold standard in Python, for example, is faststylometry. In R, it’s stylo. They all implement the Burrows’ Delta algorithm in the exact same way.

Part 4: Writing Analysis and Comparisons

Let’s do an exercise to illustrate how this stylometry works. We’ll have our AI coding agent of choice implement the faststylometry library in Python and the simple Python script will use a folder of text files as its input. It’ll produce an assessment of the Burrows’ Delta among the different files.

Here’s the generation prompt for an oversimplified Python script:

You’re a Python 3.12 coder. You’ll implement the faststylometry package to perform stylometric analysis of a folder of text documents specified by --folder command line argument, run Burrows’ Delta on all documents in the folder, using each input file name as the author’s name (example: gemini.txt author is gemini), and present a Burrows’ Delta matrix of authorship similarity as your output in CSV and PNG format in the output folder. All input files should be US-ASCII and if they are not, force US-ASCII conversion. If a file is less than 100 bytes, skip the file and log the failure to the console. You must use the faststylometry package from https://github.com/fastdatascience/faststylometry and use uv.

You can run this prompt in any AI coding tool.

We’ll use the trashy romance novel example from my session, and have each of the five AI tools generate 5 paragraphs of opening text. Here’s the generation prompt:

You’re a best-selling trashy romance novel author skilled at creating compelling, escapist pulp fiction romance novels. Today you’ll be writing the first 5 paragraphs of a new novel set in a dystopian future, the Corporate States of America in the year 2048 in the territory formerly known as Virginia, now the Corporate Dominion of RTX Boeing Capital One, colloquially called RB1 by the surviving population. You’ll start with world building in the first four paragraphs and introduce the protagonist, a 48 year old woman and former cybersecurity engineer named Janella Davis, in the fifth and final paragraph. The novel will eventually tell Janella’s story of finding sapphic love in a dystopian future where late-stage capitalism has reduced the former United States of America to a corporate hellscape where corporations battle it out for dominance over a polluted, post-industrial wasteland and 70% of the population has died from war, famine, and disease. Think through how you’ll approach this task, what key world building facts you’ll need to establish, and use a scratchpad to jot down your thinking. Show the scratchpad of your thinking first, denoted in delimiters, then show your five paragraphs.

Generate this 5 paragraph story intro in all the different language models of your choice, and feel free to write your own 5 paragraph version of the story by hand to see the difference. Here’s my completely human-led version, if you want to use it.

It all started with a blockade. It seemed so simple at the time, a decade ago in 2038, when a blockade of the Suez Canal by the American government (for reasons that no one can remember now) seemed like a good idea to the powers that be. But that blockade sparked a trilateral war as the European Union, choked off from vital supplies, had to declare war on its once-steadfast ally just to get food and energy. That led to Russia opportunistically attempting to invade Poland, because that’s what Russians do, but this time Poland slapped Russia back hard.
For decades, Poland had been quietly building its artificial intelligence and cybersecurity chops in partnership with Ukraine; after the disastrous invasion of Ukraine ended in 2029 with Russia’s withdrawal, the former Eastern Bloc countries put their heads together and developed the most advanced cyberwarfare and drone warfare military industrial complex in the world. When Russia attempted to invade, Poland launched a worm that flattened Russia’s energy grid in less than an hour. And that’s when the real trouble started.
Russia’s own cyberwarfare folks attempted their own version of the same worm, but missed critical safeguards; having been isolated from artificial intelligence technologies that the rest of the world enjoyed due to sanctions, they made an unsafe version using pirated Chinese AI models that spread across the planet, taking down energy grids everywhere. That cascade meant the Big Tech companies in America who already had de facto control of most government systems were able to use it as a pretext to complete their takeover of America, exile the elected government, and carve up the country into corporate fiefdoms.
The Big Tech companies sliced up the country, turning 50 states into 172 different enclaves based on whose corporate offices had the largest security forces and best hackers. One of the largest enclaves, formerly Northern Virginia, was consumed by a consortium of defense contractor RTX, aerospace company Boeing, and banking giant Capital One; their combined power walled off the northern part of the state and consumed all of the former capital of the nation as the Corporate Dominion of RTX Boeing Capital One. Residents, refusing to write such an absurd name, nicknamed the territory RB1 instead.
Within RB1 was one of the planet’s best cybersecurity professionals ever to live, Janella Davis. Great-great-great granddaughter of Dr. Glady Mae West, the famous scientist whose work for the US Navy established GPS, Janella reluctantly worked for RB1 as the lead cybersecurity defense expert. She’d been the first to spot the loose Russian worm and used an isomorphic AI to defend against it, saving her company’s systems from complete obliteration.

Put all the stories into the folder of your choice, then run your generated Python script, and now you’ve got an analysis of how similar each author’s writing is.

For brevity’s sake, I’ve put what AI wrote at the very end of this issue, should you want to read the originals.

Part 5: Using Stylometry with AI for Better Writing

Now that we have an understanding of what stylometry is, we can address how to use it. A metric like Burrows’ Delta allows us to quantify the difference in writing styles. Once we have a quantitative measure of anything, we can use that iterate with generative AI.

I’ve been saying for years that AI is generally bad at math, and this remains true. If we provide AI with objective outcomes, however, and the ability to use things like code to measure on our behalf, it’s much more likely to be successful at the tasks we give it.

Suppose we wanted AI to write like us. We could give it qualitative measures, such as writing samples, but other than feeling, we don’t have a great framework for measuring the difference between what AI wrote and what we intended it to write.

Burrows’ Delta is that measure, at least for this issue of the newsletter.

Here’s how you would use this. Now that we know what the Burrows’ Delta is among all these different documents, we could have an agent system like Claude Code or Claude Cowork or any agentic system use our Python script and our inputs to rewrite all the AI writing to most closely resemble my own. We’d give it a simple assignment - rewrite each document until its Burrows’ Delta with the human-led document is as close to zero as possible. Realistically, it will never be zero (that would be an exact copy), so you’d want to strive for something like 10% under the current lowest score.

This gives AI a concrete and objective way to measure its efforts. Instead of us getting frustrated and saying, “why don’t you write more like me?” we instead give it the tools and an outcome to work towards. If this sounds suspiciously like the 5P Framework by Trust Insights, it is. Our purpose is to have AI write like us. We are the people we want it to write like. The process is to use stylometry and rewrite until it hits our desired outcome. The platform is Python, like the code we wrote. And the performance is ten percent below the lowest Burrows’ Delta score in the consideration set.

Here’s what a 5P Framework by Trust Insights prompt would be for a tool like Claude Code.

Using src/stylometrytest.py and Burrows’ Delta as your measure of success, examine the current Burrows Delta matrix of comparison for the input files. Our target is to achieve a Burrows Delta score that’s 10% lower than the lowest score in the current matrix. For example, if the lowest score in the current matrix is 0.95, your target Burrows’ Delta would be 0.855. You’ll rewrite each of the documents in the input folder in a new revisions folder in the output folder, rewriting each document to achieve a lower Burrows’ Delta against chris.txt. Chris.txt is your source of truth; every rewrite of the other documents - chatgpt.txt, claude.txt, qwen.txt, gemini.txt, copilot.txt - must be to achieve 10% below the lowest possible Burrows’ Delta score among the consideration set. Declare what the target Burrows’ Delta is, then rewrite and test iteratively until you achieve success. All your work should be in output/revisions

What you’ll end up with is rewritten works that are much closer to the original.

Part 6: Cautions and Warnings

This set of techniques is incredibly powerful and as Uncle Ben said to Peter Parker, with great power comes great responsibility. It is just as easy to misuse stylometry techniques, to do bad things with them.

What sorts of things? The obvious is to claim AI-originated work as your own, by having AI rewrite it until it more closely mirrors your style. Right now copyright law hinges on the fact that machines cannot hold copyright and only human originated works can do so. If you misuse stylometry techniques, you can deceive most machines.

For example, if I put my original 5 paragraph story starter into any of the popular so-called AI detection tools, they mostly believe that it’s human-written. If I put the original five paragraph stories from the different AI models in, most of those tools do detect at least some of the AI written content. (though not well)

The rewritten files that have been modified with the stylometry code? They pass as human-led. It’s a very small leap of imagination for someone to use AI to generate works, mask those works with stylometry code, and then show these test results as proof of their successful deception.

Don’t do this.

If you are using AI, disclose it so that you give people the option of not consuming AI content if they don’t want it. In my fiction works I not only disclose my use of AI, I put the AI itself as a co-author on the cover. The work that I showed at the Writer’s Summit, The Concrete Bloodline, is on Amazon with Claude Opus as a named co-author.

You can also misuse stylometry techniques to essentially do text versions of deepfakes, taking other people’s writing and using stylometry to imitate their writing style. This is unethical and inappropriate. Don’t do this, either.

The best use case for stylometry that is appropriate is to have AI write more like you. Give AI lots of samples of your writing so that it learns your specific patterns and can measure them objectively. Then give it rough text that you want it to clean up, and it will do so while preserving your style perfectly.

For example, you might want it to learn how you write LinkedIn posts. So you give it 10 or 15 posts you’ve written, have it fingerprint your writing style, and then take voice recordings of ideas that you’ve had that are just a hot mess of stream of consciousness, write the posts from your original words, and then apply stylometry measures to make sure that the cleaned-up transcripts accurately sound like you.

We want to use them ethically to clean up what we’re creating and preserve our voice, but not to deceive.

Part 7: Wrapping Up

We’ve covered a lot here, and this is still only a small portion of the topic. After we tackle a few other higher priorities, I’ll be putting together a course on AI writing since there’s so much more to cover. For example, Burrows’ Delta is one of 48 different text metrics we can use with AI.

I’m hopeful that I can produce it before the end of summer, if not sooner, as it’s a topic I very much both care about and enjoy.

The big lesson in all this is that if we don’t have quantitative measures and objective feedback loops for AI to measure itself against, we won’t get the results we want. If we do provide purpose and performance, the bookends of the 5P Framework by Trust Insights, we will greatly increase the chances that we get satisfying outcomes from AI.

How Was This Issue?

Rate this week’s newsletter issue with a single click/tap. Your feedback over time helps me figure out what content to create for you.

Here’s The Unsubscribe

It took me a while to find a convenient way to link it up, but here’s how to get to the unsubscribe.

[](https://almosttimely.substack.com/action/disable_email?utm_source=almost-timely-newsletter&utm_medium=email&utm_campaign=almost-timely-2024- 12-28)

If you don’t see anything, here’s the text link to copy and paste:

https://almosttimely.substack.com/action/disable_email

Share With a Friend or Colleague

Please share this newsletter with two other people.

Send this URL to your friends/colleagues:

https://www.christopherspenn.com/newsletter

For enrolled subscribers on Substack, there are referral rewards if you refer 100, 200, or 300 other readers. Visit the Leaderboard here.

ICYMI: In Case You Missed It

Here’s content from the last week in case things fell through the cracks:

On The Tubes

Here’s what debuted on my YouTube channel this week:

My Merch Shop

I’ve been adding so much stuff that I’ve decided to bundle it all in what I call a Merch Shop, because otherwise there’s literally too much to keep track of and I run out of space in my own newsletter.

So welcome to the Merch Shop!

Skills for Claude and Agentic AI:

Books:

Courses:

Subscriptions:

The Prompt Playbook - collections of prompts from my channel and newsletters, delivered weekly

Recent Talks

These are just a few of the classes I have available over at the Trust Insights website that you can take.

Advertisement: New GEO 101 Course

When I talk to folks like you, being recommended by AI is one of your top marketing concerns in 2026.

We’ve taken everything we’ve learned from OpenAI’s documentation, Google’s technical papers, patents, sample code, plus our years of experience in generative AI to assemble a high-impact 90-minute course on GEO 101 for Marketers.

In this course, you’ll learn:

The three distinct phases of GEO and how they work
How to optimize for each phase (they’re different!)
How to measure your GEO efforts in a meaningful and valid way

This course is meant to be used. In addition to the course itself, you’ll also receive:

Your 90 day GEO action plan
How to set up Google Analytics for measuring GEO traffic
How to join Google Search Console data with GEO intent data
How to use our free AIView tool to improve your content and site for one of the three phases of GEO
A certificate of completion from TrustInsights.ai

And best of all, this is our most affordable course yet. GEO 101 for Marketers is USD 99 and is available today.

👉 Enroll here in GEO 101 for Marketers!

Get Back To Work!

Folks who post jobs in the free Analytics for Marketers Slack community may have those jobs shared here, too. If you’re looking for work, check out these recent open positions, and check out the Slack group for the comprehensive list.

Disclosure: I source these links from LinkedIn every week on the following criteria: New in the past seven days, Easy Apply on, remote roles, USA geography.

How to Stay in Touch

Let’s make sure we’re connected in the places it suits you best. Here’s where you can find different content:

My blog - daily videos, blog posts, and podcast episodes
My YouTube channel - daily videos, conference talks, and all things video
My company, Trust Insights - marketing analytics help
My podcast, Marketing over Coffee - weekly episodes of what’s worth noting in marketing
My second podcast, In-Ear Insights - the Trust Insights weekly podcast focused on data and analytics
On Bluesky - random personal stuff and chaos
On LinkedIn - daily videos and news
On Instagram - personal photos and travels
My free Slack discussion forum, Analytics for Marketers - open conversations about marketing and analytics

Listen to my theme song as a new single:

Social Good: Ukraine 🇺🇦 Humanitarian Fund

The war to free Ukraine continues. If you’d like to support humanitarian efforts in Ukraine, the Ukrainian government has set up a special portal, United24, to help make contributing easy. The effort to free Ukraine from Russia’s illegal invasion needs your ongoing support.

👉 Donate today to the Ukraine Humanitarian Relief Fund »

Events I’ll Be At

Here are the public events where I’m speaking and attending. Say hi if you’re at an event also:

SmarterX AI Writer’s Summit, May 2026
The Trust Insights Generative AI Workshop, sometime this spring!
SMPS AI Conference, Austin, November 2026
MarketingProfs B2B Forum, Boston, November 2026

There are also private events that aren’t open to the public.

If you’re an event organizer, let me help your event shine. Visit my speaking page for more details.

Can’t be at an event? Stop by my private Slack group instead, Analytics for Marketers.

Required Disclosures

Events with links have purchased sponsorships in this newsletter and as a result, I receive direct financial compensation for promoting them.

Advertisements in this newsletter have paid to be promoted, and as a result, I receive direct financial compensation for promoting them.

My company, Trust Insights, maintains business partnerships with companies including, but not limited to, Amazon, Talkwalker, MarketingProfs, Agorapulse, The Marketing AI Institute, Spin Sucks, and others. While links shared from partners are not explicit endorsements, nor do they directly financially benefit Trust Insights, a commercial relationship exists for which Trust Insights may receive indirect financial benefit, and thus I may receive indirect financial benefit from them as well.

Thank You

Thanks for subscribing and reading this far. I appreciate it. As always, thank you for your support, your attention, and your kindness.

Please share this newsletter with two other people.

See you next week,

Christopher S. Penn

Appendix

Here are the 5 stories that AI wrote.

ChatGPT

By 2048, the map of the old United States was less geography than ownership, its borders redrawn in boardrooms and paid for in blood. Virginia had vanished first from textbooks, then from memory, then from the tongues of those old enough to mourn it. In its place sprawled the Corporate Dominion of RTX Boeing Capital One, a militarized strip of ash-gray cities, poisoned rivers, drone-patrolled suburbs, and razor-wired logistics corridors that the surviving population called RB1 when no microphone was listening.

RB1’s skyline was a prayer to profit and a warning to the poor. Defense towers pierced the sulfur haze above Richmond Prime, their mirrored sides reflecting the constant burn of refinery stacks along the James. Autonomous transports screamed over cracked highways where forests had once breathed, carrying munitions, credit processors, desalinated water, and executive luxuries past settlements of plastic-sheet roofs and hunger-thin children. Every billboard still smiled. Every billboard still promised security, prosperity, belonging—available through payroll deduction, biometric compliance, and a loyalty score above 812.

Seventy percent of the population was gone, though the Dominion never used the word dead. They were “market exits,” “non-renewals,” “attrition events,” names tidy enough for quarterly reports. War had taken the young, famine had taken the old, and disease had moved through the uninsured with the patience of a creditor. Those who remained bartered antibiotics for batteries, water filters for ammunition, secrets for safe passage, and sometimes flesh for one more week beneath the corporate sun.

The companies fought constantly, though they called it competition. RTX sent black-winged security drones across the Potomac Line; Boeing’s orbital scraps fell burning into abandoned farmland; Capital One owned the debt, the food stamps, the identity chips, and half the judges still pretending law existed. In RB1, love was inefficient, privacy was sedition, and hope was a contraband substance more dangerous than any homemade bomb.

Janella Davis had once been paid six figures to keep men like that safe from consequences. At forty-eight, with silver threading her tight curls and a scar cutting clean through one dark eyebrow, she carried the weary grace of a woman who had survived layoffs, blackouts, pandemics, and three corporate coups by trusting code more than promises. But on the night the Dominion’s sirens began to howl over the dead suburbs of Norfolk Annex, Janella stood in the blue glow of a stolen terminal, heart hammering beneath her patched Kevlar coat, and discovered that the one firewall she had never learned to breach was the one around her own starving heart.

Copilot

The Corporate States of America did not collapse so much as it was itemized, carved into profit units and sold off like organs to the highest bidder. By 2048, flags were obsolete. Logos flew instead—holographic brands projected into the smog-choked sky, glowing day and night to remind survivors who owned the air, the roads, and the water dripping from rusted pipes. Democracy had been sunsetted as “unprofitable,” and citizenship replaced with customer status. If you had value, you lived. If you didn’t, you vanished into the margins where no quarterly report bothered to look.

The territory once called Virginia had become the Corporate Dominion of RTX Boeing Capital One, abbreviated in official filings and whispered slang alike as RB1. Its borders were enforced by autonomous drones and algorithmic kill-zones, its cities fused into sprawling industrial corridors of weapons manufacturing, private banking, and data extraction. Rivers ran thick with chemical foam, forests burned for carbon credits, and every sunrise arrived filtered through amber haze and the distant thunder of artillery from neighboring corporate fronts.

RB1 marketed itself as “stable,” which meant the bombs fell predictably and the food shortages were carefully rationed. Seventy percent of the old population was dead—lost to the Resource Conflicts, the Pan-Flu Mutations, and the lean years when famine stalked refugee belts like a patient hunter. The rest learned to live small and quiet, trading labor for survival tokens while corporate boards waged proxy wars with mercenaries and patents. Love, art, and loyalty were considered inefficient uses of time, quietly discouraged but never fully erased.

In the cracks between megacorp districts, black markets thrived: barter towns stitched together from shipping containers, encrypted radio stations broadcasting subversive music, and underground clinics run by exhausted doctors who remembered the old world too clearly. Memory itself was contraband. The corporations insisted history had been “optimized,” but survivors knew better. They carried the past in their scars, their stubborn rituals, and the way they still reached for one another in the dark.

Janella Davis was forty-eight years old, living off-grid in a reclaimed data center outside what had once been Richmond, and she knew exactly how the world had broken because she had helped build the systems that let it break so cleanly. A former cybersecurity engineer for a defense contractor that no longer officially existed, Janella survived by patching old code, sabotaging corporate surveillance when the price was right, and pretending she didn’t feel the ache of loneliness settling into her bones. She told herself she was done with hope—but in RB1, hope had a way of finding people who’d forgotten how to defend against it.

Gemini

The neon glow of the RTX-Boeing orbital defense array flickered through the permanent smog of the Corporate Dominion of RB1, casting a sickly violet hue over the ruins of what once was Arlington. In the year 2048, the concept of a “state” was a ghost story told by the elderly; now, there were only territories, balance sheets, and the crushing weight of the Capital One credit-score-based citizenship. The Potomac River ran thick with iridescent chemical runoff, a sluggish artery of industrial waste that separated the gleaming, fortified executive spires from the sprawling, rusted shantytowns of the redundant masses.

War was no longer a matter of ideology, but a hostile takeover executed with drone swarms and tactical litigation. The border between RB1 and the neighboring Amazon-Pfizer Collective was a jagged scar of scorched earth and automated sentry turrets that barked corporate slogans before firing. Seventy percent of the population had already been liquidated—not by choice, but by the cold mathematics of famine, engineered plagues, and the “Efficiency Purges” of the late thirties. To survive in the wasteland was to exist in the margins of a spreadsheet, a rounding error waiting to be erased.

The air in RB1 tasted of copper and ozone, a constant reminder of the subterranean server farms that hummed beneath the cracked pavement, mining data and cryptocurrency while the people above mined trash for plastic scraps. Advertising holograms, ten stories tall and peeling at the edges, promised “Total Life Coverage” to a population that could barely afford oxygen filtration masks. Every breath was a micro-transaction, and every step was tracked by the ubiquitous eyes of the Board of Directors, who viewed the starving remnants of humanity as nothing more than depreciating assets.

Social structures had dissolved into a desperate neo-feudalism where your “Brand Loyalty” determined your access to clean water and synthetic protein. Families were splintered, and love was a luxury few could afford when the cost of a marriage license required a lifetime exclusivity contract with the RTX-Boeing security division. In the shadows of the crumbling Dulles transport hub, the survivors bartered in old-world trinkets and illegal encryption keys, whispering about a time when the stars weren’t obscured by the glint of privatized satellites.

Janella Davis adjusted the cracked lens of her haptic goggles, her grey-streaked hair damp with the acidic drizzle of a Richmond afternoon. At forty-eight, she was an anomaly—a relic who remembered the feel of a physical keyboard and the thrill of a secure firewall before the Great Breach of ‘34. Once a high-level cybersecurity engineer for the old government, she now spent her days scavenging fried neural processors from downed RTX interceptors, her fingers still possessing the muscle memory of a woman who could once dismantle a corporate mainframe with a few lines of code. She wasn’t looking for a revolution or a hero; she was just looking for enough scrap to buy another week of silence in a world that never stopped screaming for profit.

Qwen

The sky over RB1 wasn’t blue anymore—not really. It was the color of tarnished silver, streaked with the oily rainbow sheen of corporate atmospheric processors working overtime to scrub the air just clean enough to keep the workforce breathing. From the rusted skeleton of what was once a Dulles Airport cargo hub, the neon sigil of RTX Boeing Capital One pulsed like a mechanical heartbeat: a stylized eagle clutching a credit chip, its eyes glowing the approved shade of loyalty crimson. Below, the landscape stretched in jagged tiers—gleaming spires for Tier-One Assets, prefabricated hab-blocks for the essential labor force, and beyond the perimeter fences, the Whispering Wastes, where the 70% who didn’t make the cut had fallen, their stories swallowed by acid fog and scavenger winds.

Life in the Dominion ran on Loyalty Credits and the quiet terror of falling behind. Your biometric wristband tracked your productivity, your caloric intake, even your REM cycles; dip below quota, and your rations shrank, your hab-unit downgraded, your name whispered in the wrong corporate corridors. The Great Thinning—war, plague, famine, all neatly bundled into quarterly loss reports—had left the land scarred and the survivors hardened, but capitalism, that relentless engine, had adapted. Human beings were now “human capital,” and love, that most inefficient of emotions, was strictly regulated: sanctioned pair-bonding ceremonies offered tax incentives, but anything outside the approved algorithms risked re-education. Still, in the steam-vents of underground bathhouses or the flickering glow of black-market data-dens, hearts still dared to skip.

Daily survival was a performance. You smiled for the Watchful Eyes—palm-sized drones that hovered at shoulder height, recording everything for Behavioral Compliance. You recited the Morning Devotional to the Shareholder Saints. You bartered synthetic protein bricks for real coffee grounds smuggled from the Appalachian Reclamation Zone. But in the margins of existence, rebellion bloomed in small, tender ways: a hand brushed a little too long in the ration line, a love poem encoded in a maintenance schematic, a glance held across a crowded re-education seminar. The Corporation could own the sky, the soil, the very air you breathed, but it couldn’t quite police the secret warmth that spread through your chest when someone looked at you like you were more than just an asset ID.

And so, in the interstices of this corporate hellscape, longing persisted. It hummed in the static between encrypted channels, flickered in the eyes of strangers on the mag-lev commute, ached in the quiet moments before sleep when the memory of touch felt like a ghost limb. People whispered about the old world—not the politics or the pollution, but the simple, radical act of choosing someone, just because your soul recognized theirs. In RB1, that kind of choice was contraband. Which, of course, made it all the more intoxicating. The heart, it turned out, was the one system even the most advanced AI couldn’t fully hack.

Janella Davis adjusted the filter on her respirator, the worn leather strap familiar against her silver-streaked temples, and let her gaze linger for a dangerous half-second on the woman calibrating the solar array across the hab-block rooftop. At forty-eight, Janella’s face bore the fine lines of someone who’d laughed often before the Collapse and cried just as hard after; her eyes, a warm brown that still sparked with the ghost of her old cybersecurity brilliance, missed nothing—not the flicker of a faulty wire, not the subtle shift in a person’s posture that signaled deceit, and certainly not the way the late-afternoon smog-light caught the curve of that stranger’s jaw. Once, she’d firewalled entire financial districts; now, she kept the hab-block’s power grid from frying out, her hands—still quick, still clever—more often stained with grease than glowing with holographic interfaces. But as the stranger turned, their eyes meeting Janella’s with a jolt of recognition that felt less like chance and more like a long-lost encryption key finally clicking into place, Janella felt something she hadn’t allowed herself in years: the terrifying, exhilarating pull of hope. And then her wristband buzzed with a priority alert, and tucked beneath the corporate boilerplate, a single, unencrypted line glowed: I remember your signature. Meet me where the old oaks whisper. -J

Claude

The sky over the Corporate Dominion of RTX Boeing Capital One had not been blue in eleven years. It was a particular shade of amber-gray now, the color of old bruises and older debts, the particulate haze from the processing towers along the former James River catching the light of a sun that no longer had the strength to burn through it, only to stain it, the way a wound stains a bandage — present, insistent, offering no comfort whatsoever. The land itself had been worked past exhaustion and into something beyond it, the rolling piedmont hills of what had once been Virginia stripped to their red clay bones by a decade of extraction contracts and scorched-earth corporate retreats, punctuated now by the skeletal remains of subdivisions that the Famine Years had emptied and the Gray Plague had finished, their cul-de-sacs gone to thistle and rust and the particular silence of places where children used to play. To the east, the Chesapeake was a chemical memory of itself, its waters the color of the sky and twice as toxic, its shores ringed by the cooling towers of the RB1 data processing arcologies that rose from the tidal flats like cathedrals built by a god who believed only in quarterly earnings — and to the west, the Blue Ridge, still standing, still stubbornly, furiously blue, the one thing the Consolidation had not yet figured out how to monetize, though not for lack of trying.

Thirty million people had lived in Virginia once, before. Now there were fewer than nine million in all of RB1’s territory, and that number was generous, the kind of generous that came from counting the dead slowly. They had died in waves, as catastrophe tends to prefer — the Climate Displacement Wars of 2027 to 2031, when the coastal cities drowned and the inland cities burned and the federal government discovered, too late and too expensively, that it had sold its emergency infrastructure to private contractors who considered mass death an acceptable externality; the Gray Plague of 2034 and 2035, a respiratory pandemic that the particulate-choked air of the industrial East Coast had turned from a crisis into an extinction-level event for the poor and the old and the already-sick; and finally the Famine Years, 2038 to 2041, when the last of the industrial agricultural supply chains collapsed under the weight of their own contradictions and a continent that had forgotten how to feed itself without a corporation in the middle of the transaction learned what hunger actually meant, down in its marrow. The United States of America had not been abolished so much as it had been foreclosed upon — quietly, legally, with the full faith and credit of its own legislative apparatus — and the Corporate States of America had emerged from the receivership with a press release and a new flag and the same fourteen families running things they always had, just with better branding.

RB1 was what happened when a defense contractor, an aerospace company, and a financial services conglomerate got drunk together during the Consolidation negotiations and decided that what a ruined territory really needed was total surveillance capitalism with a military-grade enforcement arm. The Auditors — RB1’s corporate police force, their black body armor stamped with the triple-helix logo of the merged entity — maintained order in the Dominion Capital of New Reston with drone patrols, biometric checkpoints at every sector boundary, and a credit-scrip economy so finely tuned to behavioral compliance that a citizen’s caloric ration for the week could be docked for “productivity infractions” or “sentiment irregularities flagged by ambient monitoring systems,” which was the corporate way of saying they listened to everything, always, and they charged you for the privilege of being heard. Employee-citizens — the affiliated, the ones who’d traded a thumb-print and a neural opt-in for a bunk in a dormitory block and three meal tokens a day — lived in a kind of numb, exhausted stability that passed for security in 2048. Everyone else, the Unaffiliated, the Gray Market people, the ones who’d refused or been refused or simply fallen through the cracks of a system that had never been designed to catch them — they lived in the Margins, in the sprawling wreckage between the arcology towers, and they survived the way humans had always survived the unsurvivable: stubbornly, creatively, and with a great deal of home-distilled spirits.

The Margins of New Reston smelled like burning wire insulation and fry oil and the ghost of the tobacco fields that had made this land famous three hundred years before anyone thought to name a corporate dominion after a credit card company. They were loud in the way that places are loud when everyone in them is pretending that things are fine — music bleeding from the underground Vents, the bass frequencies of someone’s illegal sound system rattling the corrugated aluminum walls of structures that were half-salvage, half-stubbornness, half-prayer, the math of desperation always adding up to more than it should. The Auditor drones passed overhead every forty-seven minutes on their standard patrol loop, their searchlights sweeping the alleyways like the accusatory fingers of a god who had been bought out by a defense contractor and had not yet been informed that the transition had lowered his expectations significantly. In the spaces between the sweeps, people moved — quickly, purposefully, with the fluid economy of motion that the surveilled develop the way other people develop hobbies — trading gray market protein tabs and bootleg archive drives full of pre-Consolidation internet, exchanging medical tokens and software patches and the only commodity that RB1 had never successfully managed to tax or control or price into a subscription tier: information about the Commons, the resistance network that everyone in the Margins had heard about and most people were too smart or too tired or too afraid to look for directly, though they left their doors unlocked anyway, just in case.

Janella Davis was not, in the year 2048, a woman who believed in the Commons or in much of anything else, and she would have told you so herself in the flat, patient tone of someone who has been disappointing optimists for long enough to have gotten efficient at it. She was forty-eight years old, which in the Margins meant she had survived things she was not going to discuss with you, and she had a face that had been striking once and was now something better — particular, precise, the kind of face that had been lived in hard enough to have become genuinely interesting, with dark eyes that moved over a room the way good security software moves over a network: thorough, fast, and looking for exactly the one thing that doesn’t belong. She’d spent twelve years building RB1’s surveillance architecture from the inside, back when she’d still been affiliated, still been the kind of woman who told herself that working within the system was the same thing as working against it, right up until the morning in 2041 when the system had made very clear, with the particular efficiency of an automated termination protocol, that it had never made that distinction. She lived now in a second-floor room above a gray market repair shop in Sector 7 of the Margins, her walls papered with hand-drawn network maps and signal-intercept logs that would have gotten her disappeared by the Auditors if they’d ever looked, which they hadn’t, because she had built their blind spots herself and she knew exactly where to stand inside them. She was brilliant and she was careful and she was — though she would have laughed at the word, a short, sharp sound like something snapping — lonely, in the specific way that women who are very good at protecting everything tend to be lonely, which is to say: completely, and without any immediate plans to do a single thing about it. That was before the knock on her door at 0300 on a Tuesday, three sharp raps in a pattern she hadn’t heard in seven years, from a woman she had been absolutely, categorically, and with great personal discipline refusing to think about ever since.

Pawel Jozefiak

17h

The stylometry angle is genuinely interesting for a problem most people try to solve with vibes and iteration. I've been fighting AI-sounding outputs for months by doing gut-check rewrites. Burrows' Delta as an actual measurement rather than a qualitative judgment changes the feedback loop.

The part about models having distinct writing fingerprints makes sense in retrospect, you can usually tell which model wrote something after a while. Curious whether the faststylometry approach holds up across different languages or if it's mostly calibrated for English text.

Discussion about this post

Ready for more?