Stack Overflow Users Don't Trust AI. They're Using It Anyway  ## Document Notes ## Highlights - **Rapid Code-Red Response** - Prashanth declared the AI moment existential and carved out ~10% of the company to respond quickly. - He set a six-month deadline to produce a meaningful public and enterprise response by a conference keynote. Transcript: Prashanth Chandrasekhar Was definitely a very, very surprising moment. I don't think an unexpected moment in many ways, because here comes this technology that obviously some people knew about, but not in a way that obviously captured everybody's imagination Using this beautiful interface. And we were in the middle of wrapping up our calendar year. And at that point, it was, here we go. We were thinking about our priorities for the next year. And it pretty much became very clear what we needed to focus on, because this is obviously going to be this very, very huge change to how people consume technology. And this is, you know, welcome to technology. You know, it's constantly changing and things, especially this wave, I think it's completely unprecedented. I don't think there was any sort of analogy or any other sort of, you know, prior wave that I could look to, including the cloud and maybe the internet. But, you know, I don't think, you know, we're still sort of fully sort of consuming what that is at the moment. But I would say, yeah, so we had, we went into what is the equivalent of a code red situation inside the company. It was an existential moment, especially for our public platform, because the primary, you know, the jobs to be done, if you will, is all around making sure people got answers to their Questions. And here you go, you have this really, really slick interface, that natural language interface that allows you to do that, you know, on a moment's notice. So we had to sort of organize our thoughts. And what I ended up doing was carving out 10% of the company's resources to very specifically focus on a response to this. And we set a very specific date to respond by in a meaningful fashion. So we said, hey, the summer of 2023, I was going to go speak at the We Are Developers conference in Berlin. And I effectively told the company, hey, we've got six months to go and produce our response, at least our initial response, because obviously this is going to keep iterating and so On. And that's how we mobilized the company. We had this, we acknowledged it was a code-written moment. We carved out a team of 10%. So that was about 40 people or so. So we're somewhat of a medium-sized company. And then we got to work. And that was the moment. ([Time 0:05:51](https://share.snipd.com/snip/6a2d55b4-4199-4760-b951-5c1d9a478303)) - **Two Distinct Business Realities** - Stack Overflow had two distinct businesses: a public Q&A community and an enterprise product. - The enterprise side gained strategic value because curated knowledge is crucial context for AI agents. Transcript: Prashanth Chandrasekhar And remember, our company, Stack Overflow, has really two parts. You know, we have a public platform, which had this sort of, you know, this web kind of big disruptor, which we should talk about more broadly by the internet. But then the other side of our business was the enterprise business, where we're serving large companies with our private version of Stack Overflow Insight Company. So thankfully, that was people continue to see value in having a knowledge base that's very accurate. And increasingly, over the past few years, it's actually even become more valuable because you need really great context for AI agents and assistants to work. And I've got plenty of examples we can talk about there. ([Time 0:10:32](https://share.snipd.com/snip/f08d3309-0a44-4d55-961e-dac5761eb1db)) - **Community Policing Of AI Slop** - AI answers spiked activity because people used ChatGPT to post responses, which the community quickly detected. - That detection drove the temporary ban on AI-generated answers to protect trust and quality. Transcript: Prashanth Chandrasekhar Noticed two things right out of the gate. One was the number of questions that were being asked and answered on Stack went through the roof. Because people started using, to your point, ChatGPT to answer these questions. And then they were able to sort of fuel this kind of spike, which is kind of counterintuitive. But I think people just felt like, hey, wow, I can gain the system, so let me just go do it. And very quickly, we are extremely shrewd and our community members are amazing at sort of figuring out what's real and what's not. And they were able to call out very quickly that these posts were actually ChatGPG generated. And that's kind of what initiated the ban, which we completely supported and still support, by the way. So you still cannot answer any of the questions on Stack Overflow with AI-generated content. And the reason for that, Anilay, is because what we have is effectively, our proposition is to be the trusted, vital source for technologies. That's our vision for the company. So for us, it's all about making sure that there are only a few places where you can go, where you're not dealing with AI slop, where you can actually, the community of experts have actually Voted up and curated this in a way that you can trust for various purposes. So on the input side, that it made sense to do that, and we continue to do that. ([Time 0:15:25](https://share.snipd.com/snip/377dc6ec-1f7c-4404-bcd0-b5c594da3fd9)) - **Go Where Users Already Work** - To stay relevant, show up where users spend time and provide headless APIs for integration. - Offer both in-product experiences and embedded APIs so customers can invoke Stack knowledge from tools like Cursor. Transcript: Prashanth Chandrasekhar So now that the world has changed and people are, in fact, using Cursor and GitHub Copilot and anything else to write their code, Again, our goal is to be the vital source of the technology. So let's show up wherever our users are. So we've actually become a lot more headless. More recently, we've launched, for example, MCP servers for both our public platform as well as our enterprise product. And so what people are using our platform to do now is not only invoke those MCP servers, let's say from a cursor when they're writing code and say, you know, what's the difference between Version one and version two, but also to be able to write back, which is very unique in the industry, to write back to our platform straight from cursor if, you know, they want to sort of Engage on sort of getting a deeper answer and so on. And so that's been our product principle, just go anywhere where the user is. And hence, but ultimately, we just want to be the source, whether it's inside companies or outside companies, to be that trustworthy, vital source for technologists. ([Time 0:17:57](https://share.snipd.com/snip/e31d164c-5480-4553-9b34-e41a57b0827e)) - **From Scraping To Licensing** - Stack Overflow created a data licensing business after blocking rampant scraping. - AI labs became partners who now license curated Stack content instead of scraping it uncontrolled. Transcript: Prashanth Chandrasekhar One of the things that we also noticed was that a lot of the companies that the AI labs were obviously leveraging our data for LLM pre-training and post-training needs and drag and indexing, We put up a whole bunch of anti-scrapers. We worked with third-party companies and we did that. And very quickly, we got calls from a lot of them saying, hey, we need access to your data. Let's work together to formally get access. And so we've had to do that. And now we've struck effectively partnership agreements with every single AI lab that you can think of for the most part. Cloud hyperscaler that you can think of, you know, companies like Google, OpenAI, all these folks. And even partnerships with a long tail of, you know, the Databricks, Snowflakes, you know, some of the more kind of on the secondary point, even though they're not doing LLM pre-training. And that's been our second business more recently. ([Time 0:19:54](https://share.snipd.com/snip/faea7ff1-35bb-4987-bb7e-e4ed4c7d0b7b)) - **Double Down On Complex Value** - Serve advanced, complex questions because LLMs handle simple queries well. - Offer community features like chat rooms and challenges to attract learning and reputation-building users. Transcript: Prashanth Chandrasekhar Complex questions absolutely still get asked on stack because there's no other place. The LLM is only as good as the data, which is typically human curated, and we're one of the best places for that, if not the best for technology. It's still a very active site, a lot of engagement, a lot of monthly active usage. And the questions being asked are quite, you know, I would say they're advanced questions, but what we're also increasingly seeing through our new mechanisms that we've opened up, Because we want to answer your question, we want to give people other reasons to come to the site other than just getting their answers. So we have had to broaden our site's purpose and hence the mission of cultivate community, power learning, and unlock growth. So what we've done with those things is let's open up new entry points, new ways for people to engage. So we even, in fact, for example, newly unlocked the ability for humans to chat with each other to get directional guidance. So that's been a very, very popular feature on the site where people are engaging with other experts. So we have, for example, people asking OpenAI API questions, as an example, and they can go into the OpenAI chat room and be able to engage with other people that have similar questions, As an example, or Python experts, as an example. We also opened up ability for people to demonstrate their knowledge by opening up challenges. ([Time 0:28:02](https://share.snipd.com/snip/ca30ae6a-6be7-4e37-b876-c7e1f9e2cc15)) - **Huge Use–Trust Gap** - Survey: 80% of users use or want to use AI but only 29% trust it for useful work. - That gulf explains vocal resistance despite widespread adoption. Transcript: Prashanth Chandrasekhar And the distribution on the site, I mean, we ask people on whether or not they're using AI. Our own surveys basically say, if we look at the Stackpole Flow 2025 survey, over 80% of our community members are using AI or intend to use AI. 80%, right? But, you know, the trust level on that answer, when they're using AI, is only about 29%. 29% of our user base actually trusts what's coming out of AI, which is actually quite appropriate considering where we are, because there should be skepticism of this new technology. So there's enthusiasm to try it, but not to fully trust it. ([Time 0:32:35](https://share.snipd.com/snip/990a5a3a-0515-4b01-b8aa-bf6e56135fd4)) - **Preserve Curated Trust First** - Respect the original mission by keeping human-vetted answers as authoritative and disallow AI-only answers. - Add natural language entry points while preserving curated, high-integrity content as the trust anchor. Transcript: Prashanth Chandrasekhar And we've had to sort of decide that, hey, we've got to be able to respect the original mission, keep accuracy at the heart of it. And so we're not comfortable with using, for example, AI for answers because it will generate slop because it hallucinates. And that's hence the trust score is low. But why don't we incorporate natural language interfaces? So that's the preferred way to sort of engage? And so we ended up doing that both on the public side as well as on the enterprise side of the house. And that's been really well received by the vast majority of users. But there will be always a vocal minority that will push back against incorporation. ([Time 0:34:01](https://share.snipd.com/snip/c631fe26-2b98-4e24-a892-d47b34ee625c)) - **Why People Contribute For Free** - Prashanth recounted asking founders why people contribute and learned the community is driven by altruism, not money. - He noted paying contributors is complex and historically not requested by users. Transcript: Prashanth Chandrasekhar There. One is, I think, if you think about the original DNA of this company and why people came together to do this thing. When I joined the company, I asked a question, what's people's incentive to spend time doing this? And I asked the founders, specifically Joel Spolsky, about this. And his point is that the software development community is effectively, it's very altruistic. People just want to help each other out, because people understand how frustrating. I used to write code many years ago. I recently picked it back up with, you know, some of the code gen tools, which is interesting to compare and contrast. But I just remember how frustrating it was if you get stuck on something. And so Stack was obviously a huge boon when it was created to unlock this. And so it was truly out of that. That was the reason. I also, we also asked a question, like, should we, along, even before ChatGPT, should we incentivize users by paying them for that? Should we give them a monetary benefit? And that wasn't like a high ask by a user base. We went and researched people. So people were not in for the money. Plus it complicates things because how do you judge the payment for a particular JavaScript question relative to a particular Python question? You know, it's just a very, very, it goes down a rabbit hole, which is, I think, a bit of a sort of a untenable sort of rabbit hole. And so this commercial aspect, so that's one. So what is the original reason why people got together and it was about the mission? Secondly, in terms of like, I think why we have to do this and, you know, is it unfair and so on. The primary reason why we have to go down this data licensing route of why we've had to do it is because the model of the internet has literally been, you know, has been turned upside down. People relied on, and I know you talk about this, Nile, with the DoorDash problem. I think the model of the internet has literally where people go to search engines and go to websites and you monetize off of ads. That is completely, I really empathize with content sites that are heavily dependent on advertising because I think most content sites, their traffic is down 30%, 40%, something Like that. So with this huge seed shift and where companies that support these platforms have to ultimately, we're a business ultimately. So what do we have to do? We have to do what is necessary to adopt a new business model to survive and thrive and do all the things. And so thankfully for us, we've had an enterprise business, which is independent of all of this. Thankfully for us, we were now able to, we still had the advertising business for large advertisers still cared about this, you know, our community. And so daily licensing only felt right in terms of making sure that we can, you know, effectively capitalize on the moment, plus also be able to invest back into our community so that People who are there for the right reasons saw the benefits of that. So we've invested with all these new features I just mentioned, you know, whether that is all these new content types with challenges or chat or AI assist or any of these things all takes Resources to go and build. ([Time 0:36:20](https://share.snipd.com/snip/b139bb35-2f69-4952-9f60-b32c9b53a12b)) - **Convert Scrapers Into Paying Partners** - Use anti-scraping tech and clear commercial terms before negotiating licensing deals. - Offer controlled access formats (APIs, bulk feeds) and recurring revenue terms to convert scrapers into partners. Transcript: Prashanth Chandrasekhar Somewhere in between, I would say. I think we put up the anti-scrapers very quickly. We even sort of changed the way in which people received our data dumps. And, you know, because we want, again, there's a balance because we never wanted to prevent our community users from grabbing our data for their legitimate needs, you know, for them To do their school projects or PhD theses or anything like that. So we've continued to be open about our data for our community members, but they have to be community and they can be companies looking to commercialize off the data. So we were very specific about the policy terms, putting up technology that prevented people from grabbing it so we know exactly who is scraping, who is not scraping. Some parts of those were outreach to those folks to say, look, you know, stand down because you're just, you're obviously putting up even a pressure, you're putting a lot of pressure On the servers by doing what you're doing. So, you know, take it easy. Nilay Patel But I think my characterization of those companies is they don't care or some of them care and they want to be good citizens. And some of them absolutely do not care. And they would prefer the smoke. You can just categorize. And there's a reason Amazon is suing perplexity. They told perplexity to stop it and perplexity won't. As we're speaking today, this morning, the New York Times is suing perplexity. Then there are other players who are acting in different ways and they're striking different kinds of deals. Walk me through one of those deals. When you went and struck your deal with OpenAI, was it we're going to stop you and if you want the door to be open again, you have to pay us? Or was it you know this is wrong, we can take all the technical and legal measures, but we should actually just get to the deal correctly. Walk me through that conversation. Prashanth Chandrasekhar With some folks like them, we were already, you know, we were incorporating something like OpenAI into our product. Remember the Code Red situation where we were about to announce our AI response? So we were actually using that technology to do what we had to do to incorporate AI into the public platform as well as our enterprise products. So we had a relationship with them. And we also said, look, this is not going to work. It's not tenable. And this is the new way of working. And so we need a new arrangement, a business arrangement for you to use the data. And so let's actually have a conversation. And to credit to them, they were very partner-centric around that. I was very impressed by both OpenAI and companies like Google who are all very open to engaging on this topic and wanted to be responsible AI partners. They got it immediately, even before we asked them. It wasn't like this kind of big, you know, let's go have this conversation from the ground up and justify why it had to be done. We just said, look, this is what needs to happen because this is a new business model. And we got into the conversation pretty quickly in that, okay, let's actually have a constructive way to, you know, what exactly are you looking for? Which format of data do you want to scrape the content? Do you want bulk uploads? Do you want API calls? You know, what do you want? So we got into that whole mix. And then, of course, there is the conversation around, and these are recurring, just mind you, that these are recurring revenue type of deals, that these are not one-time, you know, Payments. ([Time 0:40:10](https://share.snipd.com/snip/44bdfb6c-a56d-4c84-b70d-26c4888d75b5)) - **Dataset Value Goes Beyond Answers** - Stack's dataset is more than Q&A: it includes comments, vote history and metadata valuable for reasoning. - That context makes their corpus attractive for model training and enterprise RAG workflows. Transcript: Prashanth Chandrasekhar Way that we have thought about this is that, you know, every model that's being trained, you're training it on some corpus of information. You know, you're going from GPT-X to Y. And so in the new model that you're training, if you're leveraging our original data or some derivative of that from a prior model, then you have to pay us for it. That's effectively the legal requirement to be able to do that. And so it's a cumulative aspect, right? So let's not forget that. And so people have to pay for the cumulative data. It's not just the fact that it was used in the back of the day. And yes, obviously, relative to 20 years, one year's worth of information is going to be less, but that's why you're getting 20 plus one. That's the idea. And so that's just the way that the legal agreement has been set up. Nilay Patel Is it per year? Is it every year's worth of data is a chunk of money or how does that work? Prashanth Chandrasekhar No, it's just, it's a cumulative, it's like the whole corpus past, you know, historical data, as well as, you know, anything that's going forward for the following year, all that is Sort of one accumulated sort of data set. And that's, you know, charge is one effectively. Nilay Patel But that doesn't, so like this year's data doesn't get pulled into the training set for Gemini 3, which just came out, right? Every new question and answer in Stack Overflow since Gemini 3 has come out is not incorporated in Gemini 3's training. So you're kind of betting that they're just going to train ever bigger models. That's right. Is that how it's structured in your mind? Prashanth Chandrasekhar Yeah. And some companies have asked for their wide use cases. There's pre-training use cases. There are, even beyond that, you can leverage the data in many, many different ways for AI and non-AI use cases, search use cases, and so on. But correct, I think that there may be scenarios where larger models are built and our data is going to be useful for those scenarios. But there's going to be RAG indexing. There's going to be post-training needs. There's going to be all sorts of scenarios. And it's quite interesting to see some of the frontier labs ask for very specific slices of data that they find it to be, you know, find very useful. And we're able to remember, we've got a lot of, there's not only just a question and answer, but we've got the comment history, we've got the metadata history, we've got, you know, the Voting history, we've got the user A has gone down this path history, you know, so it's a lot of excellent context for things like reasoning and to really sort of be able to be useful to Really mimic the human brain. ([Time 0:44:40](https://share.snipd.com/snip/c46cdaf2-3cbe-4b60-80dc-332f5423f49e)) - **Ground AI With Stack Evidence** - Combat hallucinations with RAG: query trusted Stack content first, then fall back to partner LLMs. - Always return source links and attribution so users can verify answers themselves. Transcript: Prashanth Chandrasekhar We believe we've actually unlocked a very important aspect of that trust issue, which is, and responding to it, by our AI Assist feature provides, it's a RAG plus LLM solution. So effectively it provides an answer that first goes out into our corpus of information of, you know, tens of millions of questions and answers. We have 80 to 90 million. And that those are first used to produce a response. And then if they don't, then there's a fallback option where it goes and leverages open AI, for example, who's our partner, to be able to go and produce trustworthy knowledge from other Parts of the web. And so it's first searching through our trusted, attributed knowledge base. It produces the links so people can go down that path and learn more about it, which is very important to us, attribution and so on. And so that's how we are navigating this element of hallucinations. And we're constantly testing it. And it's not perfect. There will always be improvements. But we're also looking at where the world's headed. And if these models continue to get better, then we should benefit from those improvements. And ultimately, we should have the best solution because you've got grounded human context plus the LLM strengths as well. ([Time 1:00:33](https://share.snipd.com/snip/fdf331eb-19a2-4805-b86d-6e36afb9da0a)) - **Enterprise Rationalization Ahead** - 2026 will be a year of rationalization as enterprises demand measurable ROI and consolidate vendors. - Stack Internal positions Stack as the trust layer between corporate data and AI tools for enterprise adoption. Transcript: Prashanth Chandrasekhar We obviously have not only a bird's eye view in the context of our public community with this 29% data point as an example, but our enterprise customers give us a clear view on where they Are, right? Because the ROI question is being asked very heavily inside companies. And clearly, you know, I think the year that 2026 is going to be the year of rationalization. If 2025 was the year of agents, at least, where every tool is being tried out inside these companies, there's a very open, like, open landscape for CTOs to go and buy various tools, test Various tools. And so I think it's been a tremendous time for some of the companies building these tools. But 26, the CFO pressure with, hey, look, okay, there's productivity improvements have to come from these. We're going to hire less people as annual planning happens. There's going to be tremendous pressure in the system to prove out what the real value is. Everybody at a senior level that I've talked to acknowledges that this is a big shift, and they all are leaning into it pretty hard. They're all waiting for the improvements. I think most people, most companies will say that they have seen improvements in the small groups where they've tested these tools, but that's a sort of a self-selecting group because They're the enthusiasts and so on, and they will see great productivity gains, which is probably true. But there's an absolute drop-off in productivity as you think about the adoption across the enterprise. And probably for the same reasons, by the way. You're telling employees to use tools which may put them out of a job, so why would they want to do it? Or more fundamentally, if these tools are not perfect and they're hallucinating and they're going to be held accountable, then that's not good either. And then, of course, is the process change, the mindset change, the ability to completely change the workflows of how you work, all the enterprise change management work, which is Why our solution, which is Stack Internal, is building this human curation layer, a knowledge intelligence layer, so that you can put the MCP server on top of it, the knowledge base In the middle, and then our ability to ingest knowledge from other parts of the company to create these atomic Q&A that are extremely helpful to root your enterprise knowledge in through These AI agents. That is that solution. That's why we've gone and really gone hard at producing that. And we've seen a really, really strong response from our customers. We have some of the world's largest companies leveraging and testing and building this with us. ([Time 1:05:13](https://share.snipd.com/snip/612de354-c21e-45cd-b348-cfb868aebd85)) - **Two-Focused Product Strategy** - Focus product efforts on building enterprise knowledge intelligence and improving public community features. - Prioritize Stack Internal for customers and AI Assist, chat, and challenges for community engagement. Transcript: Prashanth Chandrasekhar Biggest focus is going to be making sure that we build this enterprise knowledge intelligence layer for companies to truly use AI agents in a trustworthy way. So, you know, our stack internal product that we launched a couple of weeks ago at Microsoft Ignite, in fact, is very, very excited about that on the enterprise side, as well as, of course, On the public platform, as I've mentioned throughout, to help our community users connect with each other, really learn as well as grow their careers. And there are going to be so many avenues and new entry points like our AI Assist and our subjective content and chat and other things that people hopefully find very useful as things Change around them very rapidly. And, you know, they can be part of this amazing community and help each other out. ([Time 1:10:21](https://share.snipd.com/snip/a4c6b178-066e-4aee-b691-5ca7e88e92b2))