Welcome to another episode of the Mixed Signals podcast from us here at Semafor Media, where we are talking to all of the most interesting and important people shaping our new media age. I'm Max Tani, I'm the media editor here at Semafor, and with me as always is our editor-in-chief Ben Smith. Ben, when was the last time you were on Reddit?
Probably this morning. Yeah, this definitely this morning. I'm traveling and I wanted to check out a restaurant.
Yeah, that's actually my experience with Reddit. I feel like I solicit it for information when I don't know a better place to go, don't know maybe a local media organization to rely on, and also when I'm having issues with my car. Nobody ever taught me how to use a car and it's really good at explaining basic information. But the reason I wanted to ask is because our guest this week is Reddit's CEO Steve Huffman. Steve's a really interesting guy and someone Ben you've known for quite some time.
Yeah, I first met Steve a few years ago when he came back to Reddit. He co-founded Reddit 21 years ago and then returned as the company moved through a funny process. It was acquired by Condé Nast, let go wild for a bit, taken public, and faced a takeover by its own community. Steve was brought in to bring the hammer down and right the ship, which he did pretty effectively, and now he's running a company that's right in the center of the AI revolution.
Right. One thing that has changed the company most and shaped how people think about it is that Reddit forms the backbone for LLMs trained on it. Essentially it's now regurgitating and spitting out a lot of information people get when querying ChatGPT, right?
Yeah, I think the language of LLMs is largely the language of Reddit. I guess we're all Redditors now.
We want to ask about how Steve feels about LLMs training on information shared by Redditors. We want to ask about Reddit's transition from a sometimes dark and not safe for workplace platform to one that's incredibly ad friendly. Its business has grown tremendously as it built up its advertising business. And we also want to ask about how he thinks about people trusting random Reddit users more than members of the media like you and I, Ben, which is a little depressing.
Yeah, the way I think about it: Reddit used to be the worst place on the internet. Now it's kind of the best place on the internet. That's partly Reddit getting better and partly everything else getting worse.
Well, we have to take a short break, but we'll be right back with Steve Huffman after this.
Well, thank you so much for joining us, Steve.
Yeah, great to see you. My pleasure.
And I think today, June 22nd, you have spent exactly half your life at Reddit. Is that right?
I think that's right. We started Reddit in 2005 and launched on June 22nd, 2005. I was 21, and today is Reddit's 21st birthday and I'm 42. So it's been half my life on this thing.
You have this kind of thousand yard stare as you say that.
I've tried to reflect on this so many times. Reddit has been bigger than I ever imagined it would be since August 2005. Ever since that moment, I feel like I'm just on the ride. Everything's been bonus.
That's a very positive and open-hearted way to see what I've always thought of as one of the quite difficult CEO jobs in the world. I was Googling your Reddit handle, spez, this morning and the first thing that pops up from a year ago is a user asking, 'What's the thing about you / spez? I see so many people insulting that user and I don't know the reason.' What is the reason?
We've all forgotten why we hate me. There are reasons. You'd have to ask the users, but certainly when we made the API pricing change three years ago, that was upsetting. And I inadvertently stirred the pot more than necessary. So people don't like spez for a while. But if we can go deep for a sec, you're touching on a transformative moment in my life. When I came back to Reddit in 2015, I'd been gone for about five years. I was so stressed and just wanted to be happy. I thought the opposite of stress was happiness. Then I read a famous book, Man's Search for Meaning by Victor Frankl. The idea is you need to find purpose in your life. I realized my job at Reddit is stressful because it's interesting. That's a real privilege and honor to have such an interesting job even if it's really stressful. So I became grateful for the stress and then I became much happier. It's a tough job, and over the years I've done better and worse at it, but it's a gift and I'm thankful to have it.
Before we got here, I was walking down the Croisette watching your TED Talk where you said Reddit was like a city, more of a community than other platforms which are more like stages. I'm curious how you think about people bullying you on Reddit, spez, or making fun of you, within the frame of trying to create a healthy community that Reddit has become known for. A corner of the internet that's less toxic than my corners of choice, X or Twitter.
That's a big question. We could spend all day on this. What you're getting at is a conflict in values. We started Reddit as a place where users decide what they talk about. One founding principle was that consenting adults should be able to talk about whatever they want. We came from a place of humility, not getting in the way. Historically Reddit empowered users. But sometimes they do things that aren't nice, creating bad communities. That creates a conflict between free speech and our intuition that something feels wrong. But the point of Reddit is that we don't have preferences. For the first decade, we didn't have a safety team or policies because it's antithetical to Reddit. But we saw bad behaviors emerge. We debated this conflict for a long time. Now we start with speech and openness but we're building a place people want to be. It's like the difference between laws and decorum. One of the hardest moderation challenges is how to moderate spez. People get really mad at me. On any other user we'd call it harassment, but if we moderate that, it'll just spiral. I've learned some decisions are just hard. There's a line we can't always perfectly articulate, but we do our best.
I have one question. Do you moderate yourself looser than you would somebody else?
Oh for sure. I don't moderate myself.
Or no, I mean you, Reddit.
Yeah, yeah. I don't do it. I don't want my hands on this.
We think about it. A very common diagram on Reddit is 'fuckspez'. That is generally allowed. If we ban that, what's left? An empty website. But there's a line beyond that where we consider the tone we want. What behaviors are acceptable? I ask the team to take a light touch, but we still want a sense of community. It's not easy.
Yeah. There's so much anger at tech CEOs these days. Maybe it's reassuring that the CEO of Reddit has the single worst experience on Reddit.
Oh, I don't have the worst experience on Reddit. That would be the spammers. I love Reddit. Even when people are angry, they might be emotional but they're generally not irrational. There's something there. They say 'fuckspez'. Are they still mad about the API thing? Maybe. Did they have a legit reason? Yes. There's information there.
Yeah, we can't hide from it. I tell the team whether it's users or a bad piece in the press, find the truth in what they're saying. There's something in there. I don't care if you don't like the message or the messenger. There might be something for us to respond to. Deal with that.
To shift gears, you guys are one of the central players in the AI story. I'm curious how much of the content from Gemini, Claude, or ChatGPT is Reddit?
A lot. We don't know for sure and they don't tell us, but it's a lot. One data point: OpenAI's last public research paper, I think GPT-2 or 3, said Reddit was about a third of their training set. Things have evolved, but it's significant. People on Reddit talk about everything in a very natural way, like how people speak, versus books or Wikipedia. The coverage is the whole human experience. People go on Reddit for their hobbies, interests, passions, current events, whatever they're going through. So to the extent people want to know about the world, there's a lot of Reddit.
How do you think about your relationship with frontier labs? You're in business with OpenAI and Google, you're suing Anthropic. How much of Reddit's long-term business is as a word dealer to frontier labs?
Our core business is a platform for community and conversation, and our business model is ads. On the side, we have formal relationships with Google and OpenAI. There are growing opportunities in using Reddit as an intelligence platform. But community and conversation is Reddit. The side extensions are bonuses. We're learning how to do that best while considering user privacy. The fact is you can understand a lot of the world through Reddit. Even before AI, Reddit was open because we believed information wants to be free. But we've been forced to evolve on a new, less free internet. So where does a free platform like Reddit fit? That's the path we've been navigating.
Is there any logic to making an exclusive deal with just one company given Reddit's unique value in AI training?
There is logic, but I'd rather have relationships with everybody. It's better for our business, more resilient, and better for the ecosystem. In this space, we don't want one AI winner or one search winner. That's bad for the internet. In AI particularly, the technology is so powerful and new that we need more players, not less.
Did Reddit intentionally become an AI feeder or was that an accident?
Neither intentional nor an accident. It was not intentional on our part, and I don't think it was an accident for companies that used our data without permission. There isn't an LLM on Earth that didn't do that.
Do you think people are happy that Reddit is feeding information back to them that fills gaps news media weren't providing?
I think Reddit being in the models makes them better. I'm proud Reddit has been part of humanity's advancement. These companies went from public research, which we support—we still give data to universities. We haven't been precious historically, but we've been forced to become more protective because we're being taken advantage of. Now, commercial use requires commercial terms. If you're doing something for free for the good of humanity, we're open to supporting it. But there's an industry around laundering data. There was a different world where we'd just give it away, but we no longer live in that world. People are taking it, enriching themselves, and lying about it. So we need to protect our data, which has resulted in formal agreements, lawsuits, and locking down access in ways we wouldn't have considered before.
Well, we have a lot more to get to with Steve, but we have to take a short break, so we'll be right back after this.
In this week's branded segment from Think with Google, I talked with Google's VP of marketing, Josh Spanier, about the massive increase in search queries. So, Josh, in Google's last earnings call, Sundar Pichai said search queries are at an all-time high. What's behind that?
Search has always been synonymous with 10 blue links. Today, Google search is unbelievably more competent, capable, and intelligent than ever. When consumers use advanced features—planning a trip, booking a meal, investigating quantum tunneling with AI mode—they're searching more than ever. They're transitioning from queries to quests. A query becomes a whole series of deeper explorations using Google's technology. We're way beyond 10 blue links.
What should marketers be doing about this?
These quests result in more insights, signals, and ways to find customers as they think about your product and shop. AI Max for search is a year old, one of the most successful products in Google's history. Marketers who turn it on see a 7% increase in conversions—a remarkable figure for a 20-year-old product. Google's AI investment is helping marketers find customers for less money and drive more conversions. Everyone's winning.
Where can people quest for more?
Head on over to thinkwithgoogle.com and search for AI Max for search or AI mode or anything AI. There's so much great content there.
Shifting gears, we're here in Cannes Lions, the biggest week in advertising. Reddit's ad business has grown significantly. You talked about community safety. When you started thinking differently about safety, was that the moment the ad business changed? What's behind the growth in advertising revenue?
Having a better posture on safety was a driving force. Ten years ago, the first and last question from customers was brand safety. Advertisers are sensitive to where their ads appear. We spent years evolving policies, building the safety team, and enforcing it. Now it rarely comes up; it's a home field advantage because our moderation has come so far. Also driving growth is Reddit's growth. The platform is much more diverse. It used to be all 20-something dudes, 10 million clones of Steve. Now it's balanced male and female, with a wider age range, including 70 and 80-year-olds. The topic surface area is much wider. Every brand's customers are on Reddit, probably talking about your brand or space. So we've become relevant to most brands.
For our listeners, can you explain why the user base changed so much? What happened?
I think it's simple but hard to do. Reddit is for everyone. Communities are a natural byproduct of people. If you don't get in the way and multiple people are in the same space, they will...
Community. That's all Reddit is. It's just a natural reflection of what people do. And they will talk about things that are interesting to them. They will look for a laugh. They will give and receive advice. It's just what human beings do. And so the design of our platform is to just let people be people. And as a result Reddit just kind of grows and expands. So to your specific question, why was Reddit demographically narrow in the past?
Demographically, Steve, in the past? Yes. And why is it broadened now in the present?
Well, we were doing things that limited. For example, our UI. I have this theory: our UI was so bad in the past it appealed to a demographic that has a high pay tolerance for bad UIs. What is that demographic? Gamers. Video gamers have terrible UIs. So people who play games are comfortable with Reddit. Okay. We have all dudes. Every time we made Reddit easier to use and easier on the eyes, we grew a little bit. Every time we made Reddit a little safer — by safer I mean the odds of you opening Reddit and seeing something that makes you think 'Ooh, this isn't for me' — every time we got better at that, we grew a little bit. There's a cycle where as we grow a little bit, we get new users, those users create communities for themselves, within those communities there's content that attracts new users. Every time somebody asks what's your strategy for growth, I say 'make Reddit better.' And that's literally it: we'll make it faster, easier to use, safer, easier to find communities you like. Every time we do that we just grow. And at times where we've plateaued it's because our product has hit a quality plateau.
A few months ago we were talking with Adam Mosseri, CEO of Instagram, and we were asking him about what he's doing to cut down on the AI slop on the platform. He had a really interesting answer: he didn't really care whether the slop was AI slop, he cared more about slop on the platform. I'm curious what steps you're taking to limit bots and AI creating content on Reddit, so it isn't swamped by bots creating content that then gets regurgitated through Google or LLMs.
