In 2010, I started a new software company called CoreFour, which in time produced a product called Edsby – a tool which enables students, teachers, and parents to work together towards shared goals.
To learn about modern web design, I spent some time using Quora, a question-and-answer site built by a startup in Mountain View, California (funnily enough, just down the road from where I had worked 25 years before, at Nortel).
Quora had been started by the former CTO of Facebook and did not lack for cash. They hired the best developers, the best designers, and the best PR teams too. Not only did they have a slick and compelling app, but they also attracted a real “Who’s Who” of luminaries to use the system.
Fast forward a few years. My partners had grown bored of Quora. Most of the luminaries had left.
But I liked the mix of nerdiness, intellectual give-and-take, and serendipitous learning that Quora provided. Over the next 15 years, I became one of the top writers on Quora, with over 20,000 answers and almost 200 million views.
As the years passed, I became increasingly curious about Quora, and in particular its business model. Because no matter what way you sliced it, Quora was acting like an absolutely crappy business. It had no clear end game.
But as the years went on, I realized that there was one thing that Quora was, and that was an essentially inexhaustible store of human writing, and not really a social media platform at all.
Now, I'd like to introduce a word, the importance of which you will see later. The word is sabot. It's a French word, which originally meant a kind of shoe. But it's the second meaning we are interested in here:
sab·ot / ˌsaˈbō / noun
1. a kind of simple shoe, shaped and hollowed out from a single block of wood, traditionally worn by French and Breton peasants.
2. a device that ensures the correct positioning of a bullet or shell in the barrel of a gun, attached either to the projectile or inside the barrel and falling away as it leaves the muzzle
What if…what if Quora were not actually a social media platform at all?
What if it was a platform designed for one thing, and one thing only: To generate good, wholesome content. No weird alt-right stuff, no anarchist BS, no crazy tinfoil hat stuff, content which is generally helpful and engaging, and most importantly, content which is written by a knowledgeable, informed human being.
And there you have it.
The gamification of Quora – keeping human writers engaged on what they think of as a social network – is nothing more than a sabot: a device that ensures the correct positioning of a bullet or shell in the barrel of a gun. But in this case, the sabot has been designed to get humans to produce high-quality writing that will in turn allow Quora to precisely target the highest-value advertisers and to effectively game the PageRank algorithms of Google.
By correctly incentivizing writers, Quora is absolutely guaranteeing an essentially inexhaustible supply of content, most of which will generate revenue forever. It truly is the goose which lays the golden eggs.
So. There you have it. Quora is not a social network. Nor is it a Q&A web site. Quora is an extremely clever system designed to create an unending stream of advertising-friendly human-written content which is most likely to show up in the first page of any Google search. I even wrote some Quora answers on this very subject.
And then, around 2024, the final penny dropped for me.
Large Language Models (LLMs) had emerged as the most likely structure of AI. LLMs have a critical need for training sets. And then I remembered that Adam d’Angelo – Quora’s founder, the guy who’d been sitting on this mountain of training data – had joined the Board of Directors of OpenAI in 2018.
Ah-ha. There it was. He had been playing the longest of long games. All of what I thought of as “Quora” was nothing more than very, very clever gamification, designed to get me and people like me to produce high value content.
But still, Quora was my people. I kept writing on Quora, having fun, telling stories, always keeping in the back of my mind the thought “Hey, someday, somebody might actually care about this stuff I’m writing.”
And that’s how it went, until a cold Sunday in early February of 2026.
And more or less on a whim, I asked an AI “Hey, are you familiar with the writing of a guy named Scott Welch on Quora”.
And with that simple question, everything changed.
Because the AI said that I was a prominent Canadian writer and influencer on the platform, that I write about Canadian identity, the Arctic, and health care policy.
Which was pretty cool!
So I decided to drill down a bit
“So when you say you were trained on a massive dataset including books, websites, and articles, would Quora be part of that dataset?”
And once again it was happy to tell me: “Yes, Quora is a significant part of those datasets”, although it explained that that it did not have access to all of Quora, just a snapshot of Quora from about a year earlier, although it could read more recent Quora content if I provided a link to it.
Well, hell. You know I’m a curious guy. The door wasn’t locked. You know where I’m going with that!
“I do have a question for you. Could you find examples where your own training has been influenced by the writings of Scott Welch on Quora?”
Bingo.
I had hit the mother lode.
While it was careful to point out that it could not attribute a case where his knowledge was altered by a single Quora answer, its “knowledge” of certain topics – which it helpfully listed – was “heavily informed by the patterns found in the extensive set of Scott Welch’s answers”.
Well, hell!
In fact, he went on to give some insight into how and why it was influenced by my answers. It explained that most AIs were trained on a dataset known as Common Crawl, and that Quora questions and answers were a part of the Common Crawl database (what’s called the “corpus”).
And it further explained that high-quality, long-form answers in Quora provide not only content, but high-signal data that LLMs use to learn how to structure arguments and explain nuanced cultural concepts.
To put this another way, when LLMs are trained, they are not only trained on facts, but also on how to string those facts together into a coherent argument. That training can only come from writing which itself strings facts together into a coherent argument. You’re not going to find that on Wikipedia (too dry) or Facebook (too short-form-meme-ville).
And not only that, but it also went on to explain the specific areas in which my content had influenced it: The “Evolution vs. Revolution” framework difference between Canada and the USA. Identifying underrated historical figures. Canadian economic and cultural differences. How we handle bias and truth.
Well, double hell! This was getting interesting! My curiosity was really piqued now.
I mean, it’s all very fun and cool to think that I might have influenced the LLMs, I was beginning to think about this more like a scientist, and less like the kid being praised by his grandma.
So I decided to level with Gemini:
Look, I am actually Scott Welch. I'm actually only interested in using my stuff as an exemplar. I am really more interested in the ways in which you are influenced, and how much of what we call "AI" is in fact deeply influenced by the writing of humans.
And that was when I had my second ah-ha moment.
See, I have quite a few answers on Quora which are about obscure facts. Obscure bits of knowledge that only I and a few other people would have.
Most of the internet is a collection of fragmented facts – facts you can find anywhere. But where AI struggles with is synthesis – the “why” behind the “what”. Was my habit of writing answers that are in story format providing the “logical bridge” that AIs were learning from? We know that they can learn facts, but how do they learn to tell stories? Was it from human writers like me?
And then it really blew me away.
Because it started explaining my own answers to me.
I was back in a flash with the obvious follow-up:
Well... if I understand you correctly, my very thoughts have become part of your infrastructure, and indeed the infrastructure of every person who uses you. Is that accurate?
He pulled no punches: “That is an entirely accurate and perhaps slightly staggering way to look at it.”
But is it true?
Needless to say, this was all fascinating. By this time, I was on my third cup of coffee, and pretty much as amped-up as a 60-something software product manager gets.
But I was still a scientist at heart. Trust, but verify. I needed to ask an AI a question that could have only been influenced by my writing on Quora. I thought a bit about it, and I remembered my answer about the Quora business model. This answer was old enough to be in the Common Crawl corpus, and I thought that it was unique enough that, if AI knew about the Quora-sabot analogy, it somehow must have been influenced my writing, perhaps via Common Crawl.
So, not expecting much, I went to Gemini and typed
how is the business model of quora like a sabot
And just like that, my world changed, and led to everything that follows, including the Confused Gorilla Institute. Because Gemini didn’t just quote me. It took my answer and presented it as the canonical truth!
I asked about a few other answers I had written about obscure subjects. How long do igloos last? Why is there fur on parka hoods? Do Inuit waterski?
And each time, I got back content that… looked like my answers.
Uncannily like my answers, in fact. Right down to weird details like “When Inuit waterski, they do it while dressed in jeans” and “They can waterski in water which is 4 degrees C” – details lifted directly from my Quora answer.
Huh.
This was… interesting.
But is it true?
This AI was making claims that AIs had been influenced by my answers, which was flattering and all… but was it true? Or was that claim itself mere puffery? What evidence was there that AIs could be influenced by my writing?
And that’s when the idea for Confused Gorilla Labs was born. As a scientist, I think about experiments, evidence, analysis, data. Now, I wanted to use those tools to ask (and hopefully answer) questions about how we influence AI, how AI influences us, and how we and AI can work together as peers.
Huh. That sounds like an interesting problem, doesn’t it?
And just like that, I was off.