# You Can't Cite a River

By Wes Pomeroy. Published July 8, 2026. https://wespomeroy.com/writing/ai-and-source-attribution/

A while back I wrote [an article about whether you can trust an AI](https://wespomeroy.com/writing/can-you-trust-ai/). To test it, I used a checklist: seven things to look for in something you trust. That checklist wasn't mine. It came from Brené Brown. She wrote about trust and named seven parts of it, like keeping your word and owning your mistakes. I borrowed the whole list to judge how much you can trust an AI. I said it was hers, because that was easy and fair.

The ethics there were straightforward. Use someone else's framework to make your point, and you credit the person who built it. Simple. But I've pulled other kinds of knowledge into the files my AI assistant works from, and those aren't anywhere near that clean.

Some of it I never gave credit for, and I still can't decide if I should have. Ideas I can't separate from my own anymore, where what I read and what I lived have blurred together. A company's hidden behavior I figured out and built on, where there's no one to thank. None of it felt like stealing. None of it felt as clean as naming Brené Brown, either.

But something has changed. You can now take in someone else's work - an article, a talk, a piece of code, a whole career of posts - and keep all of it: saved, sorted, and ready the moment you need it. That's a new kind of power, and it drags a pile of old questions into the light. Who deserves credit? What do you owe the people you learned from? Does "I learned it and used it" still mean what it always did? I've been trying to work these out.

## The Act Is Older Than the Tool

Think about a lecture. Someone stands up and shares what they've learned. If you have a good memory, or you take good notes and read them again, it stays with you. You use it in your own work for years. You don't stop to thank the speaker every time it helps. Most of the time you don't even notice you're using it. It's just part of how you think now.

Nobody calls that stealing. The lecture was meant to be used.

An AI does the same thing. It reads what people post in public, and later that reading shows up in whatever it helps you build. The tool is new: a computer instead of a brain, a saved file instead of a notebook. But the act is old. People have done it for a thousand years.

What used to vary was how much you kept. The people who got the most out of a talk were the few with a sharp memory or good notes. Everyone else forgot most of it fast. AI gives everyone that sharp memory.

## What You Agree To When You Post

When you post something in public, you're saying it's okay to learn from it. A note you keep in a drawer is private. A post you share with the world is an invitation.

So an AI reading your public words doesn't break any deal. What changes is how much of your knowledge gets used, and how fast.

People watch what happens to what they share. Say I give away good advice for free. Then I watch it make other people rich while I get nothing. I might start charging for it. Not out of spite. It's just that I made something valuable, and everyone is getting paid for it but me. Sometimes the people using it are better at running with it than I am.

Someone takes a class, or reads deeply on a subject, and over time it just becomes part of what they know - blended in with everything else, hard to trace back to any one teacher or book. What's different now is the gap between what you gave and what others get from it. It can grow fast. Fast enough to sting.

When that happens, knowledge starts to feel like a raw material. Something you can gather, store, copy, and sell. And once it feels like a product, people start putting a price on it.

## A Case That Isn't Clean

Start with one that's mine, from the other day.

I read an article by a strategy director named [Ho-yen Leung](https://www.linkedin.com/in/hoyenleung/), called "[The Slop Economy](https://www.linkedin.com/pulse/slop-economy-using-godlike-technology-scale-mediocrity-rjyfc/)." His point: everything is turning bland and safe, and it isn't really AI's fault - it's years of companies playing it safe. His fix: hand AI the repetitive work so people have room to be bold again. The "don't be boring" part, I already knew. But he put it in a way I couldn't shake.

Normally, that's where it ends. I read something good, feel the spark, and keep a sliver of it - if I'm lucky. My memory is terrible. My brain is like a bag of cats in a noisy, happy bar. Most of what I read is gone by morning.

Not this time. I had my AI read the article, boil it down to a few short rules, and drop them into a file I keep called Campaign Evaluation. That file is full of ideas. Some I came up with. Some I first ran into in other people's work. But I chose every one, put it in my own words, and decided how it gets used - and where an idea started doesn't make it any less mine now. By now they're all stirred together, and I don't see the seams anymore.

That file isn't there to think for me. I do the thinking. It's there to give my assistant good context. So when I hand it a campaign brief or a rough creative and ask for a first-pass read, it checks the work against those rules, flags the risks, and points out the openings I told it to watch for. I could do that myself, if I remembered all of it and stayed sharp. It just does it faster, and catches more. I built the file. My assistant reads from it every time I put it to work.

Now the questions - and they only bite in the future. Picture next month. I feed the AI a live campaign, and that file helps judge it. In that moment:

One: do I owe Ho-yen credit? Two: do I tell anyone a tool was involved? Three: could I even show you which flag came from his article? (That last one has a name: provenance, the trail back to the source.)

Frame it one way, and the answers are easy. A guy read a good article, and it shaped how he sizes up his work. Nobody credits every article that shaped them, or announces they take good notes, and nobody can untangle their own influences anyway. We've never asked that of anyone.

Frame it the other way, and it gets slippery. I ran his work through a machine, turned it into rules, and built them into a tool I lean on for every campaign I touch from here on - already blended past the point where I could pull his part back out. Same act? It doesn't feel the same. And I have a hard time saying exactly why.

One thread here is clear, so start with that.

The rule for copying is old and simple, and AI doesn't change it. If I copy your exact words, I owe you credit. If I take your whole idea and call it mine, I owe you credit. How I did it doesn't matter - a machine could copy it for me, or I could have a great memory and type it out by hand. Same thing. That rule is why the Brené Brown piece was easy: I used her whole framework, so I named her.

But Ho-yen's article isn't like that. I didn't copy his words or borrow one clean idea. I boiled his thinking into rules and stirred them in with the rest. The old rule was built for things you can point at. Ho-yen's rules, once they're stirred in, aren't something I can point at.

## Where the Trail Runs Out

So if I'm this careful, why do I still feel uneasy? People have always folded what they've picked up into their own thinking. What's changed is the strength of it. The old way was a light beer - a bit of knowledge, watered down by a memory like mine. What I can do now is ten shots of moonshine. Same drink on paper. Nothing like the same kick.

Giving credit carefully and being able to give it at all are two different problems. The second one is slipping away.

That Campaign Evaluation file is where it starts. Ho-yen's rules didn't land in an empty drawer. They joined years of other notes - from articles, talks, code, and half-remembered conversations - each one reshaped to fit how I work, each new batch stacked on the last. Do that long enough and you can't trace any of it. The ideas have mixed and split and changed so many times that no single source is left. Nothing was hidden. The sources just aren't separable anymore. It's like drops of water in a river. You can't credit a river.

People have a name for a softer version of this. They call it provenance decay, the trail fading over time. What I'm talking about is stranger than fading. The trail was never one trail to begin with. And AI makes it happen faster and wider than any notebook could, mostly out of sight.

## Knowledge That Belongs to Everyone

Follow that far enough, and knowledge starts to belong to everyone and no one. You can't trace it. You can't own it. Part of me thinks that could be a good thing, even a beautiful one, for a world that's spent centuries fencing off what it knows.

But the road there is bumpy. A lot of our rules, our markets, and people's reasons to share were built on two things: knowledge being scarce, and people getting credit. Take away the trail, and you take away one reason to share at all. Some people will keep giving it away. Others will lock it up and charge, or go quiet.

The good ending and the messy road can both be true. We won't see some of the damage until we're standing in it. Planning ahead softens some of it. The rest lands anyway.

## Where That Leaves Me

I'm in a strange spot. I'm careful about credit. I name my sources, and I keep a folder of half-read studies so I don't misuse them. I give credit even when no one would ever check.

And here I am, describing a future where being that careful won't even be possible. The knowledge won't hold still long enough to trace. That's how the tools will work.

So for now, my answer is small and practical. Keep the trail while you still can. You can't un-learn something, so you may as well know where it came from. Be honest about the parts where the trail runs out, instead of faking a certainty you don't have. And if you make your living giving knowledge away, think hard about what you put in the river and what you keep on the shore.