Why does suggesting questions and showing citations matter more than a feedback widget in a chatbot?
A thumbs up or down button mostly helps you, the team running the chatbot. Suggested questions and citations help the person using it, right when they need help. In our testing of Ask LukeW, those two patterns were what decided whether people got value at all.
Chat interfaces have a blank slate problem. You open one up and wonder: what can I do here? Suggested questions answer that. In our January 2025 usability studies, they were a key part of helping people get the sense of what this is for and also get to a good result. They also made them feel confident that they understood the whole concept. And people actually use them: a huge percentage of all interactions kicked off from one of these suggestions.
Placement mattered a lot. Our first versions hid suggestions behind a light bulb icon and then a "what else can I ask" button, but in both cases, the burden is on people to take this action. We ended up putting them directly in line as almost part of the answer, so going from an answer to a next step felt like one continuous flow.
Citations handle the other big question: can I trust this? Even just seeing where information comes from gives people a real trust boost. In testing, those little numbered footnotes did two jobs. They built trust, and they explained to people how it worked: these answers are assembled from real content, and here's where each piece came from.
So where does a rating widget fit? Back in 2023, a conference attendee asked me about letting people rate answers and then educating them to ask better questions. Suggested questions already do that second part. They show people what a good question looks like before they ever get a bad answer.
Feedback controls are fine to have. Just build the things that get people to value (suggestions) and help them trust it (citations) first, and treat ratings as a nice extra.