In this episode, Kamyl Bazbaz (VP of Comms) talks with Tim Raybould (AI Lead) about the design of DuckDuckGo’s AI-assisted Search Assist — why it defaults to concise answers, and how the new “More” button lets users dive deeper when they choose.

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.

Kamyl BazbazHi, welcome to Duck Tales, where we pull back the curtain on DuckDuckGo to share stories, tech, and people building privacy tools for everyone. Today, my guest is Tim Rae Bold, who leads AI development at DuckDuckGo. Welcome, Tim.

Tim RaybouldThank you, Camille... Baz... Baz?

Kamyl BazbazThat's right. Pretty good. We're going to talk about a new feature that we just shipped on the old SERP at DuckDuckGo, which is the More button on assist. Can you briefly explain what the More button does and why we decided to add it to the AI assist answers?

Tim RaybouldYep, sure. Well, search assist, first of all, is the AI-generated answer that goes on the top of results for about a fifth of queries. That's whenever we think that the query is asking for information that can be answered concisely. And then the main thing we pride ourselves on with assist is that it is just the answer to your question and nothing more. It's unique in that way, I think, across the AI answers industry, in that it really tries to get the job done in a very short amount of words. But users for some queries were asking it to go deeper on topics, and that's the answer to the more button question. So it's a pretty simple feature. Right below the concise answer is a button that says more, and you click it, and then it will go and expand its answer into more of a... maybe like a traditional answer that you'd see in an AI chatbot that has tables and headings and it lays things out.

Kamyl BazbazVery cool. And so what problem do you feel like users were experiencing that led to this feature?

Tim RaybouldYeah, it was really just a request to go deeper on some percentage of topics. The short concise answers are just what the doctor ordered for many, many of them. But the click rate on the more button has been really good so far. It's about 10% of people are clicking that button. Around that portion of queries, the topic they asked about is they just want a little more depth with the answer.

Kamyl BazbazWhat was the biggest technical challenge you had in implementing this expanded answer?

Tim RaybouldMm-hmm. Well, I don't know how far back you want to pull the curtain. Your intro said you were pulling the—OK, well, I mean, the SERP is written in, largely written in Perl. Perl has a—it is challenging to stream the response back in Perl. So when you use an AI tool like DuckAI, our chatbot tool, the response will stream back, which means that every single set of words, it doesn't wait until the end of its response to give it to you. It gives you the response as it's going. And it's a much faster experience for users because they can start to read it as the AI is generating word after word after word. That's called streaming. In SearchAssist, we don't need to stream that concise response. We wait until it's done. We run some safety checks on it even, and then we present the answer on the page. But with the more button, it would just be too long to wait to take that same approach. So we needed to figure out streaming. And in some cases, that's not a hard technical problem at all. In our particular environment and setup, that happened to be the hardest problem. Second hardest problem is you need to come up with a prompt that explains to the AI what type of response we're looking for. Our initial prompt, as we've been saying, is very focused on being concise. But this one, we needed to figure out the right words to tell the AI to say what type of answer we want out of an expanded answer and how we wanted to lay it out and talk to the user.

Kamyl BazbazGotcha. And so how are you—I mean, it sounds like in the development of this, you were trying to balance the speed and simplicity people expect from DuckDuckGo with getting a more comprehensive answer to their question. Does that sound right?

Tim RaybouldYeah, it does sound right. Yeah. And I mean, it seems simple in retrospect that a simple button to click more was the answer, but we, as we were building search assist, you know, we knew we wanted to give more in some cases to some answers. And it was, you know, sort of a lightbulb moment to just say, well, the first step we can take is just to ship the user a button and the user can decide when they want more. So that took the burden off of us to figure out which query exactly deserves to have more of an answer. And then from a speed standpoint, that sort of solves that problem too. Because once they're engaged and clicking more, they've presumably read the concise answer, and the new answer is going to stream back, so they're going to get to read something almost right away. The speed piece there, we have a little bit more leeway to take a few seconds to deliver the answer, whereas the concise answer is coming along with a SERP and people want a fast SERP. So, can't delay that.

Kamyl BazbazWhat's been the feedback so far and then sort of what kind of data are we seeing?

Tim RaybouldIt was the number one feature request before we built it. And it has largely satisfied that request—that people asking for it has died down, as you would expect. Our overall—the main metric we use to track user satisfaction for this feature is the ratio of people clicking the thumbs up over the thumbs down. That's at the bottom of each answer. It says, was this helpful with these two options? So after we shipped the more button, the overall thumbs up to thumbs down ratio increased noticeably. People seemed to like it.

Kamyl BazbazAnd if you're thinking about the differences between DuckDuckGo, what you would see on Google, and then if you are using ChatGPT or like a Perplexity, now that DuckDuckGo has more and is easily connecting to DuckDuckAI, how would you place our offerings in that context?

Tim RaybouldYeah, well, when it comes specifically to these AI responses, I still think the main differentiator is that we play the 80-20 rule really well with the length of the answer at first. In most cases, giving the user the direct answer in a way that they don't have to spend a lot of time parsing or reading is the preferred experience. And we've heard users say that—specifically say that—your answers are so much shorter. I didn't even know I wanted that, but that makes a lot more sense in most queries. Letting the user expand into this more button is the first step into some queries do deserve a richer, deeper answer. It deserves more work from the AI to be able to do for the user. So this more button is the first step along that path. But I think there's other tools that started there and they're coming back to what people want in a search engine—the core search engine experience. They're having to add on to the core AI experience. And we just started the other way around. We started with a great search engine that already gave a bunch of instant answers that were not AI-generated—stocks and sports and weather and things like that. This, you know, our search assist tool fits right into the wider search experience for people.

Kamyl BazbazHow does this fit into, you know, the DuckDuckGo, you know, looking to the future here? How do you see this evolving? What do you think is next for these kinds of—for the search assist answer on DuckDuckGo?

Tim RaybouldWell, I think both of our AI tools will evolve. And Duck AI will get more and more ability to do things that search engines do. And a search engine will get more and more ability to lay out a search engine results page, a SERP, with the benefit of generated content from AI. The difference right now is primarily the types of queries that people are bringing to these two tools. Whether or not people's behavior merges into one tool is anybody's guess. It's a very interesting question. Right now on the search engine, most of the queries are four words or fewer. You're not typing big long prompts or asking long complicated questions. So the search assist lives on the SERP and is built for those types of queries more so. Whereas in Duck.ai we see—we don't see actually, because we don't save any of the prompts. And that can be annoying from a building a product standpoint. We know from other data sets and other tools and benchmarks that when people go to AI tools, they don't just type in four words. They're being a little bit more conversational in the way that they prompt the model. So that deserves a different—yeah, those queries are—the user is asking something different. Obviously the answer is going to be differently natured.

One more thing on the future. So right now the search assist concise answer—it goes to the model one time. One shot is the terminology in AI land. And we give it context from the web and it has all