Inside DuckDuckGo

In this episode, Greg (Product) and Karl (Design Engineering) discuss how we build, design and decide when to show Instant Answers. Plus, how we’re balancing instant answers with optional AI generated answers.
Disclaimers: (1) The audio, video (above), and transcript (below) are unedited and may contain minor inaccuracies or transcription errors. (2) This website is operated by Substack. This is their privacy policy.
Greg: Hello and welcome to DuckTales where we go behind the scenes at DuckDuckGo and discuss the stories, technology and people that help build privacy tools for everyone. In each episode you hear about our vision, product updates, engineering, our approach to AI from employees at DuckDuckGo. I’m Greg, I work on the search engine here at DuckDuckGo and with me today is Karl. Karl, would you like to introduce yourself?
Karl: Yeah, sure. Yeah, so I’m Karl. I’m a design engineer here at DuckDuckGo and I work across the search engine on instant answers and our Search Assist AI module.
Greg: Awesome. And that’s actually what we want to talk about today. DuckDuckGo Instant Answers, how we show instant answers for different search queries. Obviously, in a search engine, we show links to websites. But there’s a lot of other content that we show on the page as well, which I think broadly we sort of refer to as instant answers. You know things like if you search for the weather, you know, we might show a weather module at the top of the page. I’ll just share a couple examples of these. You know, so you search for weather and we try to show, you know, relevant information about the weather at the top of the page in addition to then all of the, you know, the links to different websites.
Karl: Nice.
Greg: We have things like unit conversions. If you search for a temperature and you want to convert it from Celsius to Fahrenheit, different calculations, all of these things are kind of within the umbrella of instant answers. We have had instant answers for a long time. I think really early in the history of the company and the product, recognized that when you’re pulling information from a lot of different sources, you have the ability to present that information in different ways to help the user find what they’re looking for, whether that is looking for a website that they want to navigate to, looking for an answer to a question, or looking for something a little bit more complicated. So yeah, we have lots of different instant answers, and thought we’d talk a little bit more about that today. So I guess as a starting point, I’m curious to talk about, and you mentioned being a design engineer, what are some of the challenges involved in designing and building instant answers?
Karl: So I think one of the things that I’ve always found working on these is it’s very easy to just basically take everything we get. So we will work with like a provider. Let’s say if we’re doing where to watch something, let me share my screen actually as well and we can have a look at this. We are working with a provider. So in this case, we’re working with JustWatch and using their API data. When you come to design something like this, it’s very easy to look at that and say like, okay, cool, let’s go to JustWatch’s website and have a look at what’s available. We can see their website, how they present the information they have and they give us. And you could kind of jump to say like, well, we’ll just take what they’ve done. They’ve got that data, we’re using the same data, let’s just copy the way that they’ve presented that information. That would be like the easy solution. But then you’ve got all of these sort of legacy things to deal with, like we talked about a bit earlier, we’ve had instant answers for a long time. We’ve got ways that we present our information on the search engine and we need these modules, whether they’re pre-existing or they’re new, like we need them to all feel like they belong together in some way. So it needs to sit in the search engine nicely, you know, and feel kind of like it belongs. So we can’t just do like a one-to-one copy, which, you know, as designers anyway, we don’t want to just copy what someone else has done. But when the data is so vast and so complex, it can be kind of hard if you’re not familiar with the domain so much as an individual to just figure out like, what do I even care about here? So like, here’s an example with a movie, we were trying to figure out where you can watch this, but like, where are you in the world? You know, because that’s like the first thing that’s probably one of the biggest challenges is we can’t show you the same thing if you’re in the US or the UK or Australia. You’re going to see something different.
Greg: I think you touched on a lot of important pieces to this that make this challenging. One is just understanding the user’s context. Obviously, we have limited information about our users based on our privacy policy and our privacy standards. We can approximate things like location accurately enough to be able to do things like the weather one that I showed, or this where to watch example. But that does limit us in what we are trying to do where we’re not trying to overly personalize the things that we’re showing down to a very specific user profile. And so, we have to kind of generalize and try to figure out just from the search query and that limited bit of information, what is it that we’re going to show to try to answer the user’s query. The other thing you touched on that I think is important to highlight is just the fact that this search query here is very specific and I think has a pretty clear intent where to watch a specific movie. Sometimes the intent is a little bit harder to unpack. And so you might have more different things that you could show. If you just searched for the movie title, for example, maybe someone’s trying to figure out where to watch it. Maybe they’re interested in the release date or different news about it. If it’s an upcoming movie, maybe they want to know what the cast is like. So trying to figure that out and then present the information in a way that gives users the ability to see what they’re looking for and also then refine their search down to the more specific thing. I think that adds some challenge as well.
Karl: No, I agree. I think probably one of the biggest things that I found difficult around this is exactly that is the like figuring out the query, figuring out the query space, and then defining the data set, the golden data set, as we call it, that helps us to understand when do we trigger? How do we decide what we trigger? And entertainment is especially challenging because we have three different modules that are part of the class of entertainment. We have the titles module, which is the overarching piece that tells you, you know, this movie is from this date. This is a brief synopsis and you can get to the two other modules. So this where to watch module. And then also you can get to the cast module from that parent module. But like you say, you have to trigger these differently and figure it out. And so we can do these really clear ones where someone puts the keyword in that we’re looking for, but sometimes we’ve got to try and use all these other signals from what we get from the organics or like other pieces of information that we’re retrieving to say like, actually we think this is best and you will know better than I, but like it’s incredibly hard and sometimes, you know, it doesn’t matter how hard you try, the intent might feel really obvious to the person typing it, but as far as we can derive from that information, we can’t always figure out the right thing to show you. And so we do our best, but you know, there’ll be misses sometimes.
Greg: Yeah, I want to ask you about something that’s on your screen right now because you can see, you searched the query Project Hail Mary cast and we’ve got the cast module at the top left there. We also have that Search Assist box to the right side that attempts to kind of answer the question. I guess, we’ve talked a little bit on previous episodes about Search Assist and what it’s doing. You know, how would you describe how we think about showing those two things next to each other, the different kind of roles they play, and maybe how that’s evolving over time?
Karl: I think, yeah, that’s been a big challenge for us because AI summarization is a really useful tool for some people. It’s not for everybody. And obviously, we’re super careful about giving people optionality and letting you say, I don’t want to see this. But for people who want it, people who are interested in seeing that, we don’t want to just always put that front and center as the key answer to every query. Because in a case like this, for example, Search Assist is giving you the same information, but it’s giving you it as text. And maybe we’re giving you a little bit extra too. But if I’m specifically trying to ask about the cast, there is a high chance it’s because I’m thinking of an individual who I’ve seen in the film, or I’ve heard is gonna be in the film, or what have you. And so I’m trying to see their face to be like, yes, okay, so it is Ryan Gosling who’s in Project Hail Mary. Okay, cool. And so in this case, like that is probably what we feel the most critical piece of information. So we want to give you that module upfront. But at the same time, if you’re someone who’s interested in getting a kind of AI summary as part of your query, we don’t want to then say, just, we won’t show you that tool then. And so we have this like convenient little right rail slot where we can say, actually, we’ll just move this over here. So you get kind of the what we think is the best of both worlds where we can deprioritize Search Assist in this case where the intent is really clear. And then in cases where we’re not so sure, we will do our best to try and figure out like maybe Search Assist is the best answer of this or maybe it’s a different module.
Greg: Yeah, I’ll also note, we have different sort of settings that you can use for Search Assist everywhere from, I never want to see Search Assist at all up through, show it to me as often as you can. You obviously have it set to show often here. I do too. But yeah, I mean, for this kind of query, certainly the use case you spoke to of like, I’m sitting with my family watching a movie, go, wait, who’s that? This is the type of thing where, okay, you can get that information really quickly. But I guess to your point, we recognize with a search engine, you’re walking a line between giving just the immediate information that a user is asking for and then giving them the ability to dive deeper, which a lot of people are using a search engine for. It’s sort of a, people come and type all manner of topics and things into that search box. And we kind of have to be ready for any of it and different user expectations. I would say we do, I’m curious to the extent to which you agree with this, like we do generally try to keep things relatively uncluttered, like we’re not trying to sort of throw everything at the page all at once. You know, I think from my perspective here, this is sort of one of the more kind of rich search results pages you would see on DuckDuckGo. You’ve got the news module down below. You’ve got some kind of imagery within the organic links. And that’s about it. You know, I guess from a design standpoint, how do you kind of see the role of just like the information layout and kind of how much we’re showing on the page at a time.
Karl: Yeah, that’s a good question. I’m a huge clean freak when it comes to design as well. It’s kind of hard trying to walk the line because when we talk to people who use the search engine, we’ll try to understand, how do you feel about the page they’re presenting? Especially if we try and compare to anyone else, what do you think about it? And we always get kind of mixed results, I guess, because some people love a nice clean layout. Some people want to see loads of stuff because they’re just so used to being bombarded with richness of like images here, there and everywhere. And one of the biggest challenges is that we obviously want to meet people where they are. We want to help give them what they’re looking for, but we don’t want to go to the kind of degree that we just say like, we’ll absolve all responsibility and just give you everything that’s available. Like here’s every image and every kind of video and stuff’s auto playing and like just taking up all of your space. Like for me, like you said, this current SERP that we’re looking at is this layout of the page. Like this is very rich for us. And it’s like at that kind of point of balance where I think we are okay with it. And we’re like, that’s good. But you know, if we imagined that Search Assist also had an image at this point as well and other things in there, it starts to make it really difficult for somebody to figure out like, where does my eye need to go, what do I need to look at in this particular point? So we’re trying to kind of keep it really clean and say, you know, cast is this most important piece right now. That’s you’ve got big images, big module, and then we kind of de-prioritize as you go further down. But it is probably one of the hardest challenges and it’s why whenever we’re developing a module like this, it’s not the kind of thing that we can develop in isolation. You know, you can’t sit in a Figma file just with this design of a flat version of a cast module and be like, you know, that looks great. I love it. Because the minute you put it in the SERP, you realize, you know, okay, I’ve got to deal with all of these other visual things and maybe we need to turn some stuff off or change something. And that’s I think, you know, being a design engineer here, it made it possible when we developing these entertainment modules for me to do that. So I was given the kind of the rope to get into the code base and start to put these things in a SERP and play with them. And I guess you just, you get a different feel for what works and what doesn’t work when you actually see it in real place in the code base.
Greg: Yeah, definitely. I’ll also add, similar to the Search Assist setting, we have a setting that lets you basically turn the instant answers off if that’s what you want to, recognizing that some people like more and some people want less. Although I think in a lot of cases, we see users keeping them on and finding a lot of value in them. I’m curious, as we maybe get towards wrapping up our conversation, putting your user hat on for a minute. What are the instant answers that you use the most?
Karl: Well, I will say that I use my entertainment ones all the time, of course, but that’s also just as a kind of, I guess, a curious person who works on them. I have to keep an eye on them. Search Assist is obviously one of my, again, most frequently used, but I think conversions is a surprising one that I use a lot, and I think that’s especially the case because being from the UK but working with a lot of my American colleagues, when people are just chatting about Fahrenheit, I have no idea. So I have to end up getting in there and converting that. But also great for time zones and things like that when you’re trying to figure out when you can do a meeting, you know, like this meeting, for example, I can jump into that. And then weather, again, like one of the simple things that just, it saves a ton of time when you’re already in a search engine doing other things, just pulling up the weather. IA is a super handy one. What about you? I feel like you’re gonna be a sports fan.
Greg: Yeah, well, we haven’t talked much about sports, but I do use the sports modules a lot. The sports ones are interesting because, you know, with something like conversions, you know, that’s pretty consistent as far as the queries we see day in, day out, people like looking for conversions. Sports is going to depend on, you know, the events that are happening, right? Our American football module is probably not triggering very much right now. But when the season starts in the fall, it’ll start showing again. And certainly, the day of the Super Bowl, it’ll be shown quite a lot. I’m a big Philadelphia sports fan. I have some queries bookmarked that I just look at pretty frequently. I find, personally, the instant answers for sports are actually kind of my favorite way to get some of this information. You know, if I’m looking for like the NHL standings and, you know, or kind of, you know, when do the Flyers play their next game or when did they play their last game? Like this search query is really quick for me to do. It loads quickly. I kind of know what all the information is here and where it’s going to be. So I find this pretty handy. And we’ve actually just this week, I think we’ve shipped some new sports. We’ve shipped Formula One, we’ve shipped NASCAR, and we’re planning to add a bunch more sports as well for people who want to get their sports information from DuckDuckGo. So yeah, I guess anything else we missed that we want to talk about before we wrap up?
Karl: I think we’ve covered pretty much everything, yeah, I can think of.
Greg: Yeah, I’m sure we could do future episodes deep dive into more. We didn’t talk about local at all, which is its own answer that has a lot of complexity to it. But yeah, I guess we can wrap up there. Thanks, Karl. This was fun.
Karl: Yeah, awesome, thanks.
Greg: All right, bye.


This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit insideduckduckgo.substack.com

In this episode, Gabriel (Founder) and Julia (People Operations) discuss hack days (our version of a hackathon), how we encourage participation, and some of the product changes it’s led to.
Disclaimers: (1) The audio, video (above), and transcript (below) are unedited and may contain minor inaccuracies or transcription errors. (2) This website is operated by Substack. This is their privacy policy.
Gabriel: Hello everybody. Welcome to Duck Tales again. I’m Gabriel, the founder of DuckDuckGo. And with me today is Julia. Julia, do you want to introduce yourself?
Julia: Yeah, I’m Julia. I am part of the PeopleOps team. Been at DuckDuckGo for three years and a half. So yeah, almost four. Very excited to be here.
Gabriel: That’s cool. Yeah. It feels like a long time. I’m really bad at time. I know I’ve been working with you for a long time. You’re wearing a DuckDuckGo sweater from our... one more swag like that one. Yeah. Nice. Okay. So today we’re here to talk about hack days, which is something, it’s not exactly PeopleOps, but it’s something that you also just work on here. You’ve been responsible for it for a while. So yeah, tell us, tell us what hack days are.
Julia: Yes.
Gabriel: I know they’ve preceded you and I can talk about that too, but talk about your journey with hack days.
Julia: Yeah. So it’s, I love actually owning hack days because it’s not as... it’s related to culture, but not so specific to HR and PeopleOps. But hack days actually, it’s also known or mostly known as hackathons. A lot of companies in the tech industry do them. It’s a combination of the word hack, meaning creative exploratory programming, and marathon, which is something that you do fairly quick and in a short amount of time. So it’s kind of about working very intensively for a short amount of time and see what you can accomplish. It became popular in the late nineties, beginning of two thousands. And we just happened to call it a little differently. So we call it hack days. At DuckDuckGo, the way that we do that is about three or four weeks throughout the year. We do our hack days. It’s from Wednesday to Friday. So from Wednesday to Friday, we allow folks to just put all their regular work duties on the side, on pause, and using good judgment, of course, and work on anything they want that relates to DuckDuckGo. So full creativity, autonomy, and collaborating with other folks and in other domains. So that’s what it is.
Gabriel: Yeah, so a couple things with it. One is we, you do your largest part of this, so tell me what you think about this, but we’ve really tried to encourage people to do it too. So you don’t have to do it. I mean, that’s one thing, you can continue working your normal working day, but we’ve really tried to encourage as many people as we can to participate and also as we’ve grown to collaborate with others. With the idea being here that when you step away for those kind of three days, you know, without the constraints of like regular project scoping and oversight, and you just kind of left to like build from scratch, you can come up with and try new ideas. And especially if you collaborate and cross-pollinate, and as a result, I mean, we’ll talk about this, but a lot of good things have come about it. So we’ve been really trying to encourage people, but I think some people are kind of reluctant to do that. Worried they’re going to fall behind on their work or whatever. So yeah, so maybe you can speak to how we try to encourage people.
Julia: Yeah, it’s an interesting part of... So our culture has a lot of autonomy and one of our values is to build trust. So we treat people with trust, we trust folks when they are hired, we trust our hiring process. So we know they’re going to use good judgment, they’re going to put their work on the side if they can, they’re going to keep doing the things that they have to do and balance things out. But it is interesting because as folks come from other organizations, they do have... like some folks, some people have a fear of like, no, I’m going to leave my work behind. I don’t know how this works. Like, is it going to be okay if I actually put it on hold? Thankfully, we’ve been doing it for a while and a lot of people participate. So I feel like after you’ve gone through the first one and you see how many people participate, you’re like, this is actually a big thing here. And it’s cool to do it. So we try to motivate in a way that we build our culture, just like letting people know, like we trust you. We trust you and you can make that judgment call. We’re all grownups here. We all know it’s a business. We all know what we have to do. So it’s, I think that’s the way that we motivate. We also try to ask people to plan ahead. So we remind folks when we post the whole calendar of the hack days in the beginning of the year. So folks already know when that’s going to happen and they can kind of make sure that things that are more pressing in their regular work is taken care of in advance. And we announce a couple of weeks before it happens, so folks also have that just as a reminder. So those are some of the things that we do to motivate them. And they can also, if they’re not able to do the hack days that week, they can actually make it up to hack days in another week. So if that week actually just becomes something like, you know what, now I have to actually focus on my work, I won’t be able to pause things, they can do that in another week that seems more suiting for them.
Gabriel: Yeah, I was going to mention that if you didn’t, because it’s a subtle thing that we eventually did. I think another subtle one, but it’s just, it’s not even something we’ve done, but to your point, since so many people participate, is a lot of the leadership also participates. And I think, yeah, and I participate sometimes. And so when people see that, I think it may give more permission for people to do so, you know? Yeah, exactly.
Julia: You participate! Yeah, that’s an example thing.
Gabriel: And then on the other side of it, I know we’re gonna wrap this up, but maybe not as a good time, is like, you know, I agree, we have a good cadence now of like encouragement and lead up and stuff, but then the end of it is also kind of fun, as like kind of the end of the hack days. So what happens at the end of hack days?
Julia: Well, so we do hack days for several reasons, but one of the reasons is engagement. And one of the really cool things about hack days is you have from Wednesday to Friday, for those who don’t know, every Friday at DuckDuckGo, we have an all-company meeting. So we all get together. We talk about company updates. It’s kind of like what a town hall, a traditional town hall call would look like, but with our cool twist. So for hack days weeks, we actually try to, we cancel all the other meetings and that’s the only one that stays. And during that meeting, we have a showcase specifically for hack days. So that week is just about the showcase. We allow folks to talk about what they worked on for two minutes. We actually have to time it, which is kind of, it’s one of the things that...
Gabriel: We have a lot of people now. Yeah. Yeah.
Julia: I do not like. I’ve been really hard not to interrupt people and to just let them do it, but we are growing as an organization, so it becomes quite challenging as we grow the amount of hack days demos that we have. But during that call, you just get to listen to everything that folks created in three days. And it’s really fascinating because you get to see, well, for me, for example, I’m not in the tech team, so it’s really interesting to see how tech folks are thinking about the product, what are they doing to make things better? And from any other domain, you just get to see how people work and how fast they can accomplish things and how, you get to know more of the products because obviously they’re demoing things. So it’s kind of a win-win even if you don’t participate because you just get to learn a bunch and you just leave that call so motivated and so inspired because you get to see we are really a big bunch of like really smart people who are overachievers. So it’s really crazy what we can achieve like in three days. And it’s just like beautiful to see and you just leave that call feeling proud and feeling inspired and then you get your own ideas from it and you get to reach out to the folks who worked on things that you found interesting and just like talk to them about your feedback on that. So that is really special and I think that is the most special and impactful part of hack days, aside from, of course, the fact that it ends up impacting our products because several things are shipped and also internal processes are improved throughout those projects. So, yeah, it’s really exciting and really beautiful to see.
Gabriel: Yeah, agreed. I mean, let’s talk about that last part for a bit. So you can work on anything. We don’t even say you need to work on something that needs to ship. You know, like people can work on kind of pie in the sky ideas or little improvements they want to make. But some people do work on either internal improvements for like work process or development process or actual changes in the product that do end up shipping or spark ideas that end up shipping. And we do encourage that. I mean, so we had one, I’m just thinking of the Duck Tales episodes we did. We did one episode with, I think with Rachel, with the AI, no AI image filter that came out of hack days. Are there any others that jump out to you of like projects we did the last year or so that we ended up shipping?
Julia: Oh yes. I love that one. It’s so interesting that you say that because that is one of my favorites, basically, as someone who creates some imaging myself, as someone as an artist myself, I actually really appreciate that functionality of being able to filter out AI images and just be able to see what was actually created by humans. So I love that one. There are many, many things that I think were impactful. In the internal processes piece, it’s hard to say it because I would have to explain the actual process.
Gabriel: There’s been a lot of good internal improvements though. Yeah, like people make people’s lives easier. Yeah.
Julia: Yes, for sure. And there are small things like, for example, the automation that we did for when you’re out of the office, you actually get to update all your tools at once instead of having to go one by one. That’s really impactful and saves a bunch of time. We also had one for that you can just read the title of an Asana task in Mattermost, which saves a bunch of time so you don’t have to click to see what it is. So these small tweaks that are really important. From the product perspective, I think there was one about hiding distracting items from any website that I think is really, is very connected to what we do in our ethos of like user-first approach and giving people optionality. Also like vertical tabs. I feel like it’s one that folks ask for a lot. So that’s cool.
Gabriel: Yeah, the last hack days we had tons of vertical tab projects. We haven’t shipped those yet, but I think they will be coming and people will be very excited about that when we do.
Julia: Yes, we’re excited about those. I feel like it’s a big one and people love it. And I think, did the Easter eggs come from hack days?
Gabriel: Oh yeah, yes I did. Yeah, yeah, that’s true. I forgot about that. That was the project that I worked on.
Julia: That’s a favorite. Everybody loves it and gives a bunch of ideas. I think that’s really fun and very delightful. So yeah, some of it, but you probably know way more because you’ve been doing this for...
Gabriel: No, those are just a good example. So yeah, I mean, the point I just had was like, yeah, I think without really even trying, people really do improve things and we should kind of ship them and they make them into the product. I guess we could end with something about hack days are part of what we’ve been calling special days, or I don’t know if you’ve changed the name of that, but I think that’s what we still call it. And kind of like, it started with just hack days a long time ago before you joined, but then we had tried some other things too. Like we basically found that like, you know, about a quarter, four times a year is the right cadence for hack days, but there’s still some needs for other days to kind of take some time apart from regular work. And at some point in the past we had called them low-hanging fruit days or quick wins days. And it was kind of like these things that kind of fall through the cracks that maybe take just a few days or even you can knock out a bunch in a few days, but that we never end up doing because they’re small little tasks and they just don’t get prioritized. So that’s evolved, it’s evolved since you’ve been here too, but now maybe you could talk about what we do in the other months just briefly. Like I think we’re just calling this quick win days now, right?
Julia: Yeah, so for the sake of everybody who is listening, at DuckDuckGo we have something that we call special days. So special days is the umbrella terminology for hack days and quick wins. I feel a little bit bad for quick wins because hack days has way more hype to it. It’s a known terminology in other places.
Gabriel: Yeah. Just like this episode, we’re tacking it on to the end, but I think it’s still important just to mention because it’s an interesting concept that we’ve come up with.
Julia: For sure, for sure. So quick win days basically are things that you can do quickly. So it’s the same concept in a sense that it’s from Wednesday to Friday and you get to like pause certain work using good judgment and work on quick wins and things that you can accomplish really quickly or within that timeframe. I also do want to mention though that some things initiate as a hack days or a quick win. And then we realize the potential of that idea. And even if the person cannot end or finalize the whole thing in three days, we might just keep going because after the showcase, we realized the impact of it. So we have a very structured way that we work at DuckDuckGo. So having that space for creativity is really important. So going back to quick wins, those days are pretty much about getting low-hanging fruit taken care of. And a lot of people, a lot of our functional teams actually plan around it a lot. So what they do is that when they are working on specific projects and they see quick wins within those projects, they kind of just separate that for when we have the quick wins. And then they get a bunch of stuff done. So that’s really exciting because it becomes very productive. And it becomes a conversation within the teams. Like what are the small improvements that we can do in specific things or what are the little steps that we can go beyond in certain projects? So it actually becomes really impactful to have those days. So right now, as we evolved hack days and special days in general, we have tried to combine them. We have tried them separately. We have tried having themes connected to them. So at some point, the themes were connected to our objectives, which are our project roadmaps that are contributing to our top priorities and organization. So we have tried several things. Right now, what we do is that we have them separately, and it’s every two months. So basically, you have three hack days, three hack weeks a year and three quick win weeks a year. We try to skip weeks that are super busy like when we have our performance review process or things like that, people are involved in giving feedback. But yeah, that’s basically what those are. Also have recently implemented the... the difference with quick wins is that we don’t have the showcase, but we do tell people like, you worked on something really cool, do a demo. There’s always room for demos in our company meetings. So that’s really fun. And we are trying to increase that. And we also try to encourage people to work on not necessarily create something or work on something, but also learn something. So if there is a different domain that you want to learn about, or if there is... right now, we are highly encouraging people to test AI capabilities and what it can do, how far can it go? What are things that we can accomplish with it? What are the different tools that are out there that we can use? So quick wins also serve that purpose a lot. Like you don’t have to create something, just like learn, explore, do things that are going to help you do better work. So yeah, for hack days, we also implemented Hack Days Awards, which is one of the reasons why there’s a bit more of a hype with hack days than quick wins. So at this point we have an award, we have four awards.
Gabriel: You...
Julia: They are connected to what we’re trying to do with our product. So delight, dependability, and discovery. And then we have a fourth one for improving internal processes. And you, Gabriel, select the winners, which is really exciting for folks to just have that. And yeah, we give them a prize. We have swag that’s with special logos for our hack days. So that’s really exciting. We love swag at DuckDuckGo. So that is something that motivates people.
Gabriel: Heh heh heh.
Julia: Yeah, so I hope that answered your question about quick wins.
Gabriel: Okay, last thing, last thing. Yeah, it totally did. And part of the reason I asked was, I like about quick wins days is too, what you said is it’s not just engineering. Like it’s very clear that anyone has quick tasks across all the functional teams. Because with hack days, and this is maybe a problem with the name, like I think people assume it’s just engineering, but we have really tried to encourage non-engineering. And to your last point there, I’ll make this point that if you can... you can add anything for us to close out if you want. You know, we have been encouraging people to learn new things or use AI and I feel like AI has now really opened up in the last few days a clear path for even more non-engineers to get involved. Like, John, who’s been on this before, did his automations for... anyone can now like make automations or even build a small website and just try different things.
Julia: Yes.
Gabriel: And my hope is that we get even more kind of non-engineering, product design folks doing stuff.
Julia: Yeah, it’s so interesting because usually when we’re trying to automate something, we ask IT, like we, as in PeopleOps and folks who are not tech, that are in the non-tech teams, we usually need support to automate anything. And now we’re like able to automate things ourselves. And it’s really exciting. So we’re kind of like, are we, are we coding now? Yeah, I like it. Coding.
Gabriel: Yeah, yeah, you are.
Julia: And I mean, of course it’s all very basic stuff that we’re doing still as people who are not in tech, but it’s really exciting to see what you can accomplish with AI and what you can accomplish with just having that autonomy and freedom to just go and do it and do it yourself. And it’s all about curiosity, which is one of the things that I really love about hack days and the freedom that you get. It’s, you really, you can work on anything and you can explore anything and you can go in any team and be like, hey, I want to learn something about finance or the legal department or whatever. And just like actually be able to explore and get out. Even the folks who are in tech get out of the tech zone and kind of explore other things they can be impactful with. And so yeah, I love hack days. I think it’s super exciting. Like we, at some point we had less frequency and people had a very emotional response to it. So it’s clear to me that people really love it and that it has a huge impact. We have an engagement survey question about hack days and it’s usually pretty positive. So yeah, I’m very proud of the program. We do have some challenges like the growth and how we’re gonna manage the amount of demos and all of that, but I’m super happy to own it. And this was really exciting. Thank you. Thank you so much for having me here. It’s fun to talk about.
Gabriel: Yes, thank you for coming. I also love hack days and thank you everyone for listening. See you next time.
Julia: Thank you. Bye.


This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit insideduckduckgo.substack.com

In this episode, Peter (Product), Stephen (Design) and Balint (Windows) discuss updates to our browsers, from the most popular custom themes, to why over 40% of our iOS users have made DuckDuckGo their system-wide password manager.
Disclaimers: (1) The audio, video (above), and transcript (below) are unedited and may contain minor inaccuracies or transcription errors. (2) This website is operated by Substack. This is their privacy policy.
Peter: Hi, and welcome to DuckTales, where we go behind the scenes at DuckDuckGo and discuss the stories, the technology, and the people that help build privacy tools for everyone. In each episode, you’ll hear from DuckDuckGo employees about our vision, product updates, engineering, and more recently, approaches to AI as well. My name is Peter. I am on the product team at DuckDuckGo, often working on our browsers, which we’ll talk about today. And joined here today, I have my colleague Stephen.
Stephen: Steven, work on the product design team.
Peter: And Balint.
Balint: Hey, I’m on the Windows Developer Team.
Peter: Awesome, and today we’re going to talk a little bit about our browsers, as I mentioned. So many people don’t know we have browsers. They’ve come to know us through DuckDuckGo Private Search, which of course you can use in any browser. But we’ve for many years offered browsers on iOS and Android devices. And more recently, in the last few years, we’ve expanded that to Windows and Mac desktop computers as well. Our browsers are amongst the most popular in the market. For example, in the United States on iPhones, we are the most popular browser after Safari, which comes built in, and the Chrome browser, which of course many people have come to know over the years. So our browsers are used by lots of different people. When we started building our browsers and as we expanded our browsers to desktop computers, we started to prioritize the features and functionality in our browsers based on sort of two things. One, having privacy at their core. Of course, that’s our brand promise, and we want to deliver on that. But also, we wanted to make sure our browsers were easy to use replacements for the privacy invasive browsers that people have been using for many years. And so we prioritize a lot of features based on what would make it easier to use. We base it on people’s feedback, which we’ll talk about a little bit more. And often internally, we talk about focusing the features we build on the three Ds. We call that dependability, discoverability, and delight. The features should just work. It should be easy to find. And they should be delightful if we’re doing them right. So today, we’re going to talk about a couple different browser capabilities that fall into a few of these buckets that we thought would be interesting to share a little bit about. First, we’ll talk about custom theming. Stephen, do you want to tell us a little bit about that?
Stephen: Yeah, custom theming to me is pretty simple. It’s where you can go into your browser and just choose which color you want to use. It can be that green or blue or purple. I think we have a range of colors to pick from.
Peter: And why did we build custom theming?
Stephen: Originally, we didn’t plan to. We were going to try to keep it simple and just have a light and a dark mode. But we got into it we discovered that people really like to customize their UI. It’s just like a fun way to make browsing more fun and personalized. And they really like that. And it also can help you pick your browser out from other browsers. So it’s really also a functional improvement.
Peter: Do you want to give us a little bit of a demo on how that works?
Stephen: So we built two ways to get into this. If you’re on the new tab screen, you can open the customized sidebar and we have a theme picker up here with a few colors. You can pick gray or blue or green, purple. Lots of people love purple. And you can also change the theme from light to dark or just to use whatever the system decides. So we support all of these. And if you want, you can also change the new tab background screen to be separate from the rest. And we’ve spent quite a bit of time making sure that this works on all surfaces. So you can see it in the settings and the bookmarks in the history and elsewhere. Also, we have a neat little feature here where you can change the app icon and the dock to match the theme. So that’s a little more continuity and a little more fun added to the feature.
Peter: Stephen, were there any challenges in building this as we started to introduce it?
Stephen: Yeah, so like I said, we didn’t really have themes on the roadmap when we started. So we kind of had to go back and rethink the way we were styling things and build a whole theme system so that we could apply a bunch of different themes and make sure that it was, we could have more themes in the future. So now it’s all centralized and easy to work with. So we could probably add more or maybe even enable customization at some point, user customization.
Peter: And would you say, what’s the response been from our users so far?
Stephen: It’s been pretty positive so far. We got a lot of feedback saying, thank you for adding this. And it’s been pretty positive and monitoring people using it. It’s going up. So that’s good.
Peter: And you’re saying the certain colors were the most popular, I think you mentioned earlier.
Stephen: Yeah, so I think the most popular color is, let’s see, it’s sleet blue, I think. And then the second most popular color was violet. It’s kind of neck and neck there. Personally, desert is my favorite, but it was not the most popular.
Peter: Got it. Cool. Where are we going next with it?
Stephen: We’re looking into adding a dark mode for web pages that don’t have it, which is going to be pretty interesting. And like I said, we may, depending on if we get more feedback on how many themes we have, how many colors we ship with by default or expand that out to letting people choose their own colors.
Peter: Yeah, the dark mode for sites to use dark mode, that’s starting to roll out. We have that on mobile devices, on Android starting to roll out on iOS. And it’s something, as you said, we’re looking at on our desktop browser. So I think our users can look forward to that in the near future, which is awesome. Great. Let’s talk about password management next. Maybe Balint start with what is a password manager? Maybe people don’t realize that in their browsers they have typically password managers. Maybe just grab that at a high level first.
Balint: So the password manager is the thing that comes up when you go to a login screen and the browser offers you to fill your username and password automatically. And it’s a critical piece of feature because one of the most dangerous attack vectors used to be credential stuffing where an attacker got hold of your passwords from one site. And if you have been reusing the same password on other sites, they could just brute force. They could just try the same password with your username on other sites and usually get a hit. And the password manager is a perfect defense against this sort of attack because it takes the burden of having to memorize passwords off your back.
Peter: That makes sense. Yeah, I guess, you know, over the last 10 years, a lot of you hear a lot about third party password managers like 1Password or Bitwarden or LastPass and more people are using those. But I think it’s probably safe to say that most people use their browser as their password manager. Maybe they don’t even recognize that it is a password manager, but it is, as you’re describing overall. Why did we choose to sort of build it into the browser and start to build out a lot of those capabilities? Is it because that is the easiest way for people to protect themselves, as you’re describing?
Balint: I think it matches perfectly well with the 3Ds that we have. So it’s a core part of dependability. You want your online experience to just work, not have to memorize passwords and not have to think about where those passwords are stored. Of course, if you explicitly want to, you can use a password manager, but it is very nice to have a secure default built right into the browser. And we take care of the whole lifecycle. So when you register to a new site, our browsers will suggest secure passwords, which are randomly generated and long enough, take care of storing them, and if you turn on sync, can get your passwords synced across your different devices so you don’t have to re-enter them on every single device you have.
Peter: And we’ve had the core capabilities in our browser product for some time. What’s been the user response and maybe describe where we’re going with it next.
Balint: So we see on our anonymized dashboards that uptake for these password managers is very high and we keep adding new features. For example, recently over the past months, we have shipped a feature for Android and iOS devices where you can use DuckDuckGo’s password manager as the main system password manager and you can fill passwords in other applications from the DuckDuckGo database. And uptake for that feature has been phenomenal. We have almost half. So a bit more than 40% of our iOS users enabling us as their primary system-wide password manager. We have more cool features in the works. My personal favorite is something that is called TOTP. It is that six digit number you see on certain websites, which is used as a second factor authentication. This is something that many password managers already offer, but browsers don’t really tend to. We would be one of the first on the market to ship this. And it would present a second security layer that is much harder to phish, so much harder to steal than a conventional password.
Peter: And do you think we’ve hit challenges in building this along the way? And what are some of those challenges?
Balint: Certainly. So password management is all about security and figuring out a way that lets us store the information on the devices securely in a way that we can access them, but an attacker cannot access them. And we have been very proactive in using operating system level security features. For example, on Windows, we are storing them in a way that is secured by Windows Hello. So that is Windows’s own built-in encryption mechanism. So breaking our password manager is similar in difficulty to cracking Windows’s own encryption, which makes it a pretty hard target.
Peter: That makes sense. So when a user goes to access a stored password, that’s why the browser will cause Windows to trigger the Windows Hello authentication overall. Makes sense. And then you talked a lot about storing all these passwords. And, you know, as I mentioned earlier, a lot of people don’t know that we even offer browsers to start with. And so if they come to us and install our browser for the first time after many years of using another browser, presumably they’ll have a lot of passwords in that other browser that they need to get into our browser somehow. Maybe we can talk about that and how we’ve been iterating on the import capability, so a little bit.
Balint: Exactly. So we offer an import capability which you can do as part of onboarding. So when you first come to our browser, one of the onboarding steps is the offer to import your stuff, but you can also do it later on from main, from more tools and import on Windows. Similar logic is available on our other platforms as well. And one of the recent additions to this has been the so-called multi-import where you can very simply select not just one target or one source, but several of them, we can import directly from the most popular browsers on the market. And it’s been trimmed as well. So if everything goes well, it is just one or two clicks and you are done. We support importing passwords and also bookmarks. And this also sees pretty good uptake. So once we have updated the visuals for this multi-import, we have been seeing 13% more people choosing to import and over 70% more stuff getting imported, so 70% more passwords and bookmarks. And that is pretty good news because it reduces the friction. So when you come to DuckDuckGo, you don’t have to restart your digital life from scratch. Instead, you can continue from where you left off, except in a more private manner, of course.
Peter: That’s awesome. Yeah, my experience is, you know, over the years, people tend to often use more than one desktop browser, sometimes one for work, sometimes one for personal or for separation of different context. So the fact that we make it easy in one step to import from multiple at a time seems to be really beneficial and delightful to users overall. That’s great.
Balint: Yeah, and we are also innovating on our mobile platforms as well. So for example, we have been first to market with a feature that lets you import from mobile browsers directly. So not via a desktop browser that gets synced to your mobile version as well, but directly on the mobile browser itself. And we are planning to take care of the newly released iOS API, which will further streamline imports on the iOS platform.
Peter: That’s great. Yeah, we have that both on iOS where you can import from Safari and then we also have directly imported from the Google password manager on Android, which is great for people. Awesome. Well, thank you, Balint. Thank you, Stephen. I think as a summary, what I’ll say is if you are using our browsers or if you’re not and you choose to start to use them, we’d love to hear your feedback. We do very much listen to user feedback. There’s ways right in the product to send us feedback. You can fill out the form. We absolutely look at those, read those. If you go on our subreddit and discuss our browsers and capabilities there. We absolutely look at those threads and we take all of that as input to help make prioritization decisions about the features we’re building. So thank you very much and see you next time on DuckTales.
Balint: See you.


This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit insideduckduckgo.substack.com

In this episode, Gabriel (Founder) and Zac (SVP, Insights) discuss AI adoption, common user concerns, and why we’re building AI that users can control and customize.
Disclaimers: (1) The audio, video (above), and transcript (below) are unedited and may contain minor inaccuracies or transcription errors. (2) This website is operated by Substack. This is their privacy policy.
Gabriel: Hello, welcome back to Duck Tales. I’m Gabriel, founder of DuckDuckGo. I have with me today Zac Pappis.
Zac: Yep, I’m Zac. I’ve been here for about 14 years now, so I think we’ve had the pleasure of working together for a very long time. And we’re on our insights team, so generally doing a lot of market research or user research and generally trying to give us better insight into what our customers and future customers want.
Gabriel: Cool, and yeah, happy to have you. And we’ve been friends for 15 years. But we are going to talk about AI today. You’ve done a lot of research in AI over the last year. And we obviously recently put out this kind of Yes AI, No AI public poll. And we’ve talked about AI a few times on the show here, just in terms of product, but also just more generally. So our approach to AI has been a bit different than the other tech companies. Obviously we’re asking people if they want AI, no one seems to ever be doing that. But more generally our approach is to make AI features that are private, of course, because we’re a privacy company, useful, which I think lots of customers have different opinions on what AI is useful and what is not useful across all sorts of products. And most importantly, I think to this discussion, optional. So we’re making all of our AI features optional. You can turn them off or tune them, actually. We’re going to get into that too. So that’s been our approach, private, useful, optional. And we thought we would talk about, with you today, who’s been actually doing a lot of research with consumers and looking at other research, kind of how those three things kind of thread through the research. So if we were gonna, you know, obviously that’s a big topic, maybe we can start at the highest level and work our way down. Like, yeah, highest level landscape of AI right now, kind of like what are you seeing in all of the research trends that you’re looking at?
Zac: Yeah, I mean, generally, I think what we feel just as being consumers and people downstream of a lot of technology is that it’s been pushed on us as well as the general market, despite not really asking if we wanted it. And the market moves so fast that I think we feel the consumer has been left behind and no one really ever got an informed choice or consent into how prevalent AI has become in every product. And we think that’s a mistake, right? So for users and the companies that are building it, that choice as you laid it out is to make these things really truly useful and optional and of course, private. But that’s not what we’ve seen. And I don’t think that’s how most people have felt in terms of the ways that the companies have rolled these products into, or I should say rolled them out to their broader consumer base. Lots of tech companies have put AI overviews into products or turned AI on by default without really asking for consumer consent. It really reminds me a lot of what had happened through the 2010s and the cookie era or social logins or a lot of cases where technology just sort of appeared for people without a lot of demand for it or really a lot of cost to use it or a lack of concern into how exactly their data was going to be used or how would it impact the usefulness of the product they were using.
Gabriel: Yeah, that makes sense. I mean, giving them the best intentions, like, even though we kind of just screwed the approach, I think somebody saw a lot of people’s assumptions seem to be it was super useful and everyone’s going to love this. And I think to some degree it’s turned out not as useful for lots of people in different scenarios. And then some people just don’t want it shoved down their throat. You know, they would like the choice. They might use it anyway. So with that in mind, like, I know you actually — so like that’s kind of the general consumer sentiment, but like in terms of like actual adoption, like data points, where are you, what are you seeing right now?
Zac: Yeah, so we see that this is everywhere. We also see that people are using it a lot. Some studies that we have looked at and run ourselves, as well as those that we’ve evaluated from external companies and polling services, or things like Pew had run a study in June of last year showing that there is a lot of variability when it comes to who is using AI and who has a strong preference for it. A good example might be that 29% of parents use AI daily, but 15% of non-parents use it. And that’s almost twice the rate from parents and non-parents that are using it. That’s just the type of disparity, I think, that we see across different consumer types. So older or younger, more educated, less educated, there is a real preference and usage gap between these groups. And that doesn’t mean that this product is going to be right or the specific implementation is going to be right for everybody. And so that level of specificity or attenuation for exactly how the product appears to people is really important to us. So despite the fact that you have lots of parents, 30% of parents using AI daily, I think even more, 50% of Americans are actually concerned, more concerned than excited about AI. And that’s gone up over the last few years. So you have these two things happening. You have a lot more exposure and usage of AI and more familiarity with it. But as people are becoming more familiar with it, their concern is going up. And that to us is evident that someone’s really living with a choice that they didn’t really get to make.
Gabriel: Yeah. Well, I think one thing is usually said in there and it goes to some other data points I’ve seen very recently, correct me if I’m wrong, but there was another Pew that came out really recently on teenagers and it was about, I think about half the teenagers were kind of regular daily users. And then there was another one, just about ChatGPT usage and it was like maybe 40% of desktop people were weekly users. What I see on that is I do see that’s obviously very significant adoption in three or four years. However, that’s still like majority of people close are not daily users. You know, so like the one headline I take from what you said and what I’ve seen is that yes, there is a lot of adoption, but this idea that everyone’s using it all the time is that narrative just seems not true. And it relates to your concern points. Like the people who are using it are concerned. People who aren’t using it are also very concerned. And so there’s just like generally a building concern. There are obviously lots of different issues people have with AI. So like, how do you like think about that concern? If you kind of try to piece it apart, like, where are people expressing, how are people expressing that concern, I guess, and how is it related to their...
Zac: Yeah. I think it’s something that we know pretty well because we’ve seen a similar type of concern with tech overreach before. Just some stats that we have on hand to share because I did look into this a little bit before today. There was a study by Resilience [CHECK] in December, so pretty recent, December 2025, showing 54% of those polled, US adults that they surveyed, have avoided AI-powered features. And in our own polling of US adults, we’ve seen something closer to like 13% of people actively disabling AI search or browser features in their browser to protect themselves. And when we look more deeply at why people are doing that, taking, you know, kind of these extra steps either to tune or to completely avoid the way that AI has been integrated into their products, we see a couple of, I should say, familiar concerns. There is a concern that companies are rolling these new technologies into their products so quickly that they come with new types of privacy trade-offs or data security concerns. And of course, we’ve seen that in the past with Cambridge Analytica and other cases where a lot of data collection can just increase the surface area, the risk surface area for having that data. So when asked why people were avoiding AI or were turning these features off, I think 51% had said they reduced data sharing because of AI, meaning a behavior that they’re taking to proactively not share as much as a result of AI being in the product. And when asked what they wanted, the majority of answers from those folks were opt-out rights, data traceability, and disclosure. None of those things are no AI. They’re consistent more with a theme that would be control, right? Not no AI. They just want to know when it’s being used, how it’s being used, and to have some input and flexibility into where it’s applied to the product.
Gabriel: Yeah, I mean, those seem like totally legitimate concerns to us. I think more broadly, like what I’m hearing is, I mean, that’s, privacy is one of the main concerns people have. That’s obviously why we’re building private AI and giving people that control. The second is that people do want options to your point. It’s not just yes or no. It’s, like, yes, some people, but it’s a smaller percentage like you pointed out, totally want to get rid of all AI because maybe they have more objections for various reasons. But it seems like the majority of people actually just want it to be useful and private. So what’s useful to somebody may not be useful to another person. And so if there are 10 AI features, maybe they want to engage with six out of 10. Maybe they want to turn the dials on them a little bit different. And so it’s like this broader customization of AI thing to make it useful that we’re trying to do with our search features that I don’t think other companies have approached in the same way. They’re just kind of like all on all the time, you know?
Zac: Yeah, exactly. Funny enough, there was an Ernst & Young study pretty recently, it might have been January or February of this year. It was a poll of 500 or so US business leaders, so people in an executive position or director position in some kind of a large company. I think these were all companies of over 40,000 employees or something pretty large. And from that survey, 78% of the company leaders polled said that their adoption is outpacing their ability to do good risk management. And 45% of the same people polled said they had a confirmed or suspected data leak via these unauthorized AI tools. So they’re really prioritizing speed over the exhaustive vetting that they would need to do to either ensure that they’re actually producing something that’s safe, both from a privacy and security standpoint, and also useful, that they’re getting it into a product really without understanding user needs or how the product is being adopted by those who are really core to their business.
Gabriel: Yeah, that’s actually a really good point. I mean, it’s like we expect these numbers that we’re citing to change over time, right? And you’re already seeing that like concern could go down if you address people’s needs for transparency and control and also concern could go down as people discover, you know, actually useful features and how they’re using it. I think part of the issue is there hasn’t been that control and transparency. And then part of the issue is I think to that last point is things are intentionally moving too fast for people and change, you know, it takes time to understand these technologies, to get good risk management, to like get good processes in place that, you know, don’t exploit your data and have other security and privacy risks. And so, like, I imagine that longer term, some of this will settle down, but it feels, I guess, probably not the level, it just — if I were to summarize some of this, it just feels like it’s moving too fast for people a bit and that everything needs to slow down a little bit.
Zac: Yeah, I mean, it’s just kind of spitballing here, but it seems almost like a double-edged sword. You both have people who are in charge of making these types of product decisions, rushing them out the door without fully understanding them. And then as consumers who are also, you know, privy to, or I guess experiencing the downstream effect of other product changes, we feel it. So you’re getting this double whammy of having to participate if you’re somebody that works in any industry right now, which probably a lot of them are impacted by AI. It’s likely that your organization is dealing with a lot of these same challenges, not really having the right oversight or internal expertise to understand the risks. A lot of pressure from the market and competitors moving so quickly and you feel like you’re going to get left behind. And it’s understandable why some people might feel like the industry or the changes are moving faster than they can really make sense of them. And I think that’s what we feel both as people somewhat responsible for creating consumer technology, but also as consumers ourselves. We see them in the products that we use every day from Apple and Microsoft, et cetera.
Gabriel: Yeah, well, let’s take a few minutes to just talk about our Yes AI, No AI campaign we did. So we put out this essentially public poll. It really wasn’t to our users, and I’ll get to that in a second, about asking, are you Yes AI or No AI? And we understand, we just talked about how it’s all nuanced and it’s about control. So we understand it’s a bit of a binary kind of thing that we’re asking people to choose, but it was kind of like, your finger in the air, just kind of say what side you’re on. Now part of that was because we think that the people who are concerned inside just haven’t had a lot of voice. They haven’t really been listened to. So this was a bit of an attempt to allow and show that and have an opportunity for a tech company to kind of listen and see what’s out there. Obviously the poll ended up very skewed. Interesting though, is like overall numbers were like 85-15, something like that. But if you look when you polled our actual users on like a platform, it was more like 50-50. And what happened was the whole poll went viral in the No AI community. And kind of my theory is, you know, not a lot of people are speaking to this community. We did. And so it went viral there when there’s everyone speaking to the Yes AI community. They don’t really have a reason to vote Yes AI, you know, but people really do want to express their No AI vote. So I thought it was interesting. We didn’t really know what to expect. That was the hypothesis. And that’s really what happened. You know, it seemed like, I guess, I guess my read on it and tell me what you think is like, I think some people are out there and maybe listening to this and being like, I still don’t believe there’s no AI sentiment here. And it’s like, I think we’re here to say, yes, there is, you know, like we have so many data points here that show there’s a large percent of people who are concerned with different AI things.
Zac: Yeah, exactly. The campaign was awesome. And I know it’s not something that we typically do. So it’s great to see just a response from it. But it was really meant, I think, to point out, correct me if I’m wrong, the gap that we didn’t, that no one really gave consumers a choice for. So this gap between yes and no, where most people aren’t in this absolutist camp. Even if I think you have some anti-AI sentiments, it’s more than likely, I guess, just given a kind of a bell curve that for most people, they fall somewhere in the middle. It can be useful for some things, you know, in certain conditions or if, you know, it’s kind of an opt-in or something that really is explicit for the user. But in other cases, maybe not. And I think the experience that this campaign had kind of really drawn on was what we talked about earlier, just a flood of AI without really a lot of ramp up and consumers not really getting a choice to speak out about it. And that certainly, as you pointed out, the people who were really pro AI were kind of getting the life that they wanted to live in and the world was really bending in their direction. But the No AI crowd probably didn’t see or kind of people on the other parts of that bell curve didn’t really see anything coming for them. They felt probably like it was and it’s still seemingly like an AI-powered world that we’re heading into. But without understanding what that is, it’s certainly scary. And certainly with the other concerns that we’ve seen in the data that we’ve just shared here today, there are privacy concerns. There are systemic concerns and how that’s going to impact the rest of the products that they use. So if it’s something that’s getting built into Amazon, how does it impact my Echo device? How does this data migrate from one process or one product to another? And all of that, I think, just comes in tenfold with AI because it is such a sensitive topic for people. And the type of content that you engage with in AI is uniquely different from, say, something that you would type into your browser. It’s a lot more personal, it’s a lot deeper, and certainly from the history that has grown from that, it can be too personal for people and really jeopardize a lot of the concerns, bring a lot of concerns forward that they had in the past with cookies and just general corporate tracking. So one thing that I think ours does really well that we’ve seen a lot of positive response for is the fact that you can use it with no accounts. You can just go to duck.ai and start using it. No accounts, it’s not training on your data, you can stop using it when you want, you can turn it off if you’d like, and all of that optionality and that level of control is just not something I think we have seen in any other product. So it would be interesting to see.
Gabriel: Yeah, agreed. And people have asked specifically on the back of that campaign, kind of what are we doing? And it’s important to say, well, first of all, all that optionality was built in already before we did that. So you could have turned Duck AI off completely from search. You can turn Search Assist off within our search results on Duck AI itself. You can choose what model provider you want. Like you don’t have an account, like you said. But I’d say additionally, we also created this domain, noai.duckduckgo.com, as well as yesai.duckduckgo.com. And those have now built in like all AI on and all AI off. If you really want to be on each end of the extreme and you don’t want to tune it, we took the time to kind of like make those two bespoke experiences for people on the end. And we have seen a decent uptake on the No AI side. Just thinking about, you know, closing out a little bit, like, is there anything on the product side or going forward, you know, in addition to all that you want to talk about from like a research perspective or anything.
Zac: Yeah, I could talk for hours. I don’t know if the podcast can sustain that but yeah, it’s just something that you just said was interesting because I did run into my neighbor yesterday who is kind of aware that I work at DuckDuckGo and was asking me about the AI, you know, just in general, not ours, just in general, what’s going on with AI and they were really delighted to hear about noai.duckduckgo.com and I directed them to our search overviews and let them know how you can tune kind of the frequency that those appear in search. And she was like really taken aback by that. I think that’s the kind of experience that we’re trying to manufacture more of is that like the need clearly matches the product that we’re providing. And that connection happens almost instantly where people recognize that that’s what they’ve been looking for. So for us, I think that’s a lot of trust and control and really turning that into what you would call like a delightful UX to getting back to, I think, the podcast as well. So not really AI that confuses you, AI that should be there when it’s truly helpful and how and where that gets embedded into the product is a lot of what our research is focused on going forward. So you’re going to be seeing a lot of how we integrate and make AI easy to get to, easy to get out of, and easy to switch from when you’re navigating between, say, traditional search, browsing, AI. If your phone is in your pocket, your laptop is in your bag, and all of those contexts where a technology can be helpful, but you may not know exactly which one can be. So we want it to be present when it is helpful and kind of hidden or tucked away when you would like to invoke it, but otherwise out of the way.
Gabriel: Cool, that’s a good place to end. Well, thank you, Zac, for coming on.
Zac: Thanks for having me. This was great. Thank you. Bye.
Gabriel: Cool, thanks everybody for listening. See you next time. Bye.


This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit insideduckduckgo.substack.com

In this episode, Gabriel (Founder) and Caine (CTO, first employee) discuss the history of our search engine, why now is the right time to build a full web search index, and how our scale makes us uniquely positioned to ship, learn and iterate quickly.
Disclaimers: (1) The audio, video (above), and transcript (below) are unedited and may contain minor inaccuracies or transcription errors. (2) This website is operated by Substack. This is their privacy policy.
Gabriel: Hello, welcome back to Duck Tales. I haven’t been here in a while. And I am Gabriel Weinberg, the founder of DuckDuckGo. And I have with me someone who I don’t think has been on Duck Tales at all yet, but you should know, Caine Tighe, who I know very well, who’s the first employee of DuckDuckGo and now our CTO. Caine.
Caine: Hi Gabe.
Gabriel: We’ve been working together for a very long time. And we’re here today to talk about something we’ve both been working on. Caine more than me, but I’m working on it some, which is our web search index. So as some background, first some background. DuckDuckGo started as a search engine, as many people know, and it was actually started by me. I was by myself for a few years. And the first thing I did was start crawling the web and building a web index.
Caine: Yeah, for sure.
Gabriel: But you know, I soon realized that that is very expensive, especially as one person. And there were other places to get a web index at the time. And what was more interesting was maybe adding value on top of the web index. So building other indexes, this was a time, this is the mid 2000s, you know, there weren’t, there obviously wasn’t AI, but there wasn’t even really many instant answers on search engines.
Caine: I mean, that’s what we were working on together at the very, very beginning. Like we were working on, you know, you had the knowledge graph. It wasn’t called a knowledge graph at a time, but you were doing all the structured content from Wikipedia and otherwise. We worked on some other smaller indices. So yeah. And then actually fun fact, in hiring our backend project is still based on some of the original spam and content farm crawling, like one of the projects is based on some of the spam and content farm crawlers that you originally wrote. So that lives on 15 years.
Gabriel: Yeah. So we were doing lots of indexing and lots of crawling. Yeah, exactly. Just not, you know, we started, but then we stopped doing a full web index, but just as examples, right? We started like the code that you were talking about indexing Wikipedia, which became our knowledge graph, you know, which is, powers a lot of answers, which also we used when we started working on AI answers. We’ve been doing local indexing for, you know, over a decade, local businesses and things like that. You know, then all sorts of kind of niche indexes that involve some crawling like lyrics and things like that. So indexing technology is not new to us, despite what some people say about it. Sometimes we do lots of search indexing, but we hadn’t been doing a full web index until relatively recently, last few years-ish. But now we are. And so the question is, the questions and why you’re here, and we’ll talk about it for a few minutes, is kind of why, what’s going on, how, all the main questions, which we’re obviously not all gonna answer today, but we can start with, I think, kind of the why, but why are we well positioned to work on this? So to speak, and you’re kind of at the center of it, so I think you’re a good person to ask.
Caine: Yeah. I mean, I think, the why now is a mixture of like our needs. Like we want to support our own AI use cases. That’s we have two primary agent, agent driven products out. Search Assist, which is on the SERP, search engine results page, duckduckgo.com. And then we have Duck AI, which is our chatbot. And both of those products are hungry for this kind of data. So it, yeah, it just makes sense for that.
Gabriel: Yeah, in particular, right. You could maybe talk percentages, but like there’s some percentage of search results now, what is it like 25%, I think that have Search Assist answers. And then, you know, the percentage better made for Duck AI, but some significant percent call the web, 15% maybe.
Caine: Yeah, I always do. I do, I have my numbers based on, absolutes make more sense. Yeah, yeah, just, yeah.
Gabriel: You know what? Bad question. Ignore numbers. Doesn’t matter. Good percentage of queries and Duck AI prompts require web search. And so we need a web index for it, essentially, right?
Caine: Yeah. I mean, I think on the chatbot side, it’s really good, like to ground. If you’re deciding whether or not to ground and you’re on the line, you should probably use RAG, retrieval augmented grounding, and go out to a third party data set. For us, raising the standard of trust online. We want to do that because the more that you ground, it’s known empirically, the answers are better. So we err on the side of grounding where I think maybe not everybody does. So it’s really good that we need to build our own index in order to be able to accommodate that. So that’s kind of some of the why now. And again, it’s on Search Assist and it’s on Duck AI. One of my favorite parts about this whole thing is like, we’re very used to working for customers, like our end user. For the search index, duckduckgo.com itself is the customer. A very nuanced, unique thing for us to be able to serve ourselves, which creates this really tight feedback loop internally. So it’s been cool to like use our own and we are live for, you know, some amount of the traffic today. That’s just growing day over day for these use cases.
Gabriel: That’s a good point because like, I think in terms of like our position, well, positioned to do it is, you know, being live, you know, maybe we talk about that a little bit, but like that creates a feedback loop that we have that a lot of people don’t have because we have, you know, many, many millions of people using our search engine and now Duck AI. And so we’re getting constant feedback about the relevancy of the search results that we’re serving, not to mention the fact that we have almost 20 years of evaluating relevancy ourself on our own search engine.
Caine: Exactly. Yeah, exactly. So like, humans are unsurprisingly and appropriately more, more critical of results than agents are. So it kind of creates this higher fidelity feedback loop because, you know, through our, through anonymization and whatever else, like we can privately understand what is most relevant on the internet for customers and users. And that really helps us to, positions us to be pretty competitive in the space quickly. So like, I think that’s kind of interesting and it’s exciting and like the true DuckDuckGo way as you and I know well, like we like to ship stuff. And so it’s been really cool to, yeah, like it’s just been really cool to be using it already and in production, our own index. And it creates that flywheel and we could, you know, use buzzwords like reinforcement learning and this, that, and the other thing. But at the end of the day, it’s just really the relationship of consuming your own internal API product. That’s the flywheel and allows you to like establish relative priority really quickly and be like, I ran this experiment. Like we really think this query set is going to be well suited to our own index. And then we tried and we’re like, we’re not working that well on that. Let’s move to this other one. And then it just changes the game for how quickly you can iterate, which has been really exciting for, and I know the team’s really excited about it too, because engineers like to ship things. So that’s been cool.
Gabriel: Perfect, I think that’s a good intro. But let’s do, to your point about buzzwords, let’s do a few more buzzwords in terms of like, just give us the broad tech flow, like, you know, without getting too deep into anything, but just to give people a sense of kind of how it works and then maybe.
Caine: Yeah, so kind of the way that I think about this is like a little pipe or a train or whatever. You have your frontier that kind of is the web that you’re looking to crawl, like, because you have to pick what your frontier is. Then you crawl that. Crawling, all of these components are extremely complicated by themselves. To crawl, it means like, you need to crawl politely. Some sites want you to crawl, some sites don’t want you to crawl. And so to be a good trustworthy netizen, you have to respect those things. And that’s an important part of crawling. It’s also important to have the bandwidth and the throughput to crawl at the scale that you need to crawl. And so fortunately for us, we’ve had a lot of experience with that, so we have that. The rendering side, you have to, when you fetch content, you have to render, including JavaScript and everything else like that. The only way to get the content is to literally run the whole webpage. Otherwise, like you get no content. So that’s quite an expensive process. So we kind of do a naive approach and then a more complicated rendered approach. Then you have content extraction, which is like the next step, where you think about your title, your description, your headings, metadata, main body stuff, where you extract the content, what the page actually means. And then we’re very fortunate in today’s day and age to have semantics. So semantic search is a big part of the pipeline. And what that means is what people are calling embeddings. And you calculate embeddings on extracted content. And then we use a database, which I quite like, called Vespa. And it’s all ingested into Vespa. In my opinion, kind of your indexing, your ingestion, your features, how you calculate those things, and how they describe, that’s a big description of your product. Because it limits what you can do in the ranking phases, which we could get into. I don’t know if you want to get into the ranking funnel as well or just the data pipeline.
Gabriel: I think that’s a pretty good overview. I mean, I think, we’ll do more of these in the future, but like, if people have questions, they can write me or write us, but we need to get an email going where people, we have the proper feedback for this podcast.
Caine: A mini series as it were. Yeah, because I’d like to bring in, you know, I can give you the overview, but there’s a lot of experts on the team that, yeah.
Gabriel: Yeah, so I think what we plan in general to come back and we can talk deeper about specific things in the future, but I think that was a great overview. I think we hit it. Anything that we missed that you want to cover or we’re good?
Caine: No, I mean, I think the space is quite interesting and it’s been really fun. Like when we started out, you know, I started out with you 15 years ago and like, we were doing something different back then. The concept of a customer being an AI agent was not really a thing. And I think AI agents as search customers is quite interesting because like, they generate the queries. AI agents are generating the queries. They’re there. The queries are formulated quite well. They don’t feature misspellings. AI agents can consume large amounts of information quickly. Like lots and lots of information. They can speak in embeddings. So I just think it’s a really exciting time. And I think we made the right call to kind of forgo it until more recently. Because now we’re quite well equipped to support ourselves. And we have the products out that really express the needs and the requirements of an API like this. So I know I’m very excited for that, and the team’s very excited for that. So yeah, we should do more of these and get deeper into what a polite crawler really means, as opposed to just glossing over it, because it’s important.
Gabriel: Cool, well, thank you, Caine, for joining, of course. We’ll do more again later. If anyone’s got questions, email them in to us. Otherwise, see you till next time. Bye, everybody.
Caine: Later. Thanks.


This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit insideduckduckgo.substack.com