PrivacyAkinator: Articulating Key Privacy Design Decisions by Answering LLM-Generated Multiple-choice Questions

Paper | CHI 2026

What is PrivacyAkinator?

PrivacyAkinator is an interactive tool that helps developers identify and articulate privacy design decisions by answering dynamically generated multiple-choice questions. Instead of requiring developers to fill out complex privacy assessment forms, PrivacyAkinator walks them through one decision at a time — surfacing choices they might otherwise overlook.

Key Idea

A single feature can involve many hidden privacy decisions. For example, Zoom's attention tracking feature involved 13+ key decisions (opt-in vs. opt-out, individual vs. aggregate scores, retention period, etc.).

PrivacyAkinator maps out this space using a verb-based privacy representation with three layers:

Layer	Description	Examples
Data flow	How data moves through the system	Collect, Process, Store, Share
Stakeholder interactions	How users engage with the data flow	Consent, Control, Notice, Audit, Access, Request, Influence
Design properties	Specific choices on each node	Data type, retention period, collection frequency

The system generates two types of questions:

• Exploratory — discover new nodes (e.g., "Should users be notified before tracking?")
• Exploitative — fill in properties on existing nodes (e.g., "How long should data be retained?")

The taxonomy was grounded in 10K privacy news articles to capture real-world privacy decisions.

Setup

Prerequisites

• Node.js 18+
• Anthropic API key

Install & Run

npm install

Create .env.local:

VITE_ANTHROPIC_KEY=your-key-here

npm run dev

Citation

@inproceedings{10.1145/3772318.3790408,
  author = {Li, Qiyu and Wong, Yuen Sum and Wong, Yuen Kei and Yu, Longxuan and Jin, Haojian},
  title = {PrivacyAkinator: Articulating Key Privacy Design Decisions by Answering LLM-Generated Multiple-choice Questions},
  year = {2026},
  publisher = {Association for Computing Machinery},
  url = {https://doi.org/10.1145/3772318.3790408},
  doi = {10.1145/3772318.3790408},
  booktitle = {Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems},
  series = {CHI '26}
}