2024.07-2024.08
Last Summer, I joined Tencent's AI Creative Design team as a product design intern. I designed the AI writing feature from 0-1, from researching the problem to creating the design for implementation. The new AI writing feature allows users to have more stable control over the generated content, improve output quality, and become one of their go-to tools for frequent use.
Key metrics post-launch week
Feature Daily Active User
16,000+
Penetration Rate Among MAU
42%
User Satisfaction Rate
+37%
1-week Article Generated
120,000+
Background & Goal
Tencent YuanBao
Tencent Yuanbao is an AI conversational app, launched on both web and mobile in June 2024. At that time, our team set an ambitious growth goal: reaching 1,000,000 DAU by the end of 2024. With this target in mind, we needed to identify key breakthrough opportunities to drive user engagement and accelerate growth.
Opportunity
AI Writing had the potential to become a new growth driver
According to the Research Team’s report, writing-related tasks were a primary use case for YuanBao, but the experience was unoptimized. Users relied on a one-question-one-answer model, which felt limited and inefficient. We saw AI Writing as a potential growth driver to boost retention and long-term engagement.
Insights from the research report
Identifying Core Pain Points in AI Writing
Our Research Team conducted focus groups and surveys to understand how users employing AI in their writing process. After analyzing the collected data, we identified two pain points that stood out. These insights became the foundation for our first phase of the AI writing project, ensuring that our design decisions directly addressed the most pressing user challenges.
Getting a Satisfactory Result Takes Too Long
Users often go through multiple rounds of Q&A, prompt adjustments, and regenerations before achieving a usable result. This trial-and-error process is slow and frustrating.
Result Editing is Frustrating
Even after receiving a relatively good result, users always needed modifications, but the current editing process was inefficient and time-consuming.
These insights became the foundation for our first phase of the AI writing project, ensuring that our design decisions directly addressed the most pressing user challenges.
HMW help users quickly get results that align with their expectations?
Our Solution
Prompt Library - Enhancing First-Try Output Quality
Pre-set prompts help users better express their needs and improve output accuracy. Given that our users come from diverse industries with distinct writing tasks, we introduced a Writing Prompt Library for high-frequency scenarios, reducing user effort while enhancing content quality and precision.
Layout: Stack vs Inline vs Block
With my PM and team, I sketched out multiple layout options for the Prompt Library. Together, we discussed the pros and cons of each direction and quickly decided on the "Block" layout as the best approach.
❌ Visual takes too much spaces, no space for extra info in a card.
❌ Scrolling is not intuitive for a vertical list - have to click the "flip page" button.
✅ There is more room for info, but we don't actually need to show the entire prompt in a card.
❌ It only displayed three prompts at a time, not good for quick browsing.
✅ There is sufficient space for prompt title & a short description.
✅ It can display 12 prompts in one page.
Outline-First Mode - Improving Coherence & User Control
Our model has a degree of forgetfulness, which becomes especially problematic when users write logic-driven and structured articles. Users often struggle to get the expected results due to AI losing track of previous context.
To address this, we introduced Outline-First Mode—when enabled, the AI first generates an editable outline before producing the full article. This gives users more control over structure and flow, ensuring a more coherent final output.
Instead of getting a fully generated text they may need to heavily edit, users can first review and adjust the structure, ensuring alignment with their intent.
HMW make it effortless for users to refine AI-generated content?
Editor
Although the chat interface is easy to use and works well for many tasks, it’s limited when people want to work on writing task that require editing and revisions. We introduce the editor mode, allow users
(This case study is still in progress, feel free to contact me for more information.)