Readers/Information Retrieval/Related Projects
Q&A
[edit]One very early exploration we are looking into is Q&A. The idea is that some Wikipedia articles might feature frequently asked questions coupled with answers surfaced from the article text.
We shared some of our early user testing results and design concepts in our Research Report: Improving How Readers Find Information on Wikipedia.
We also shared these ideas with the English Wikipedia Discord in January 2026, and found through that discussion that LLM-generated questions with human-written answers were confusing and unlikely to be accepted by the community.
As a result, the team is currently exploring other ways of generating questions. We are now looking into using human-written semantic search queries as a basis for Q&A.
- We already have some natural language queries in the data (4-7% of all searches in English Wikipedia over the years) and are exploring whether we have enough data to try a test in a subset of categories that are unlikely to be controversial. If the current data are sufficient, that would let us use anonymized and aggregated questions to test the idea on a small scale sooner, without needing to wait for semantic search to be fully launched.