Readers/Information Retrieval

Overview

Hybrid Search Phase 0

Hybrid Search Phase 1

Hybrid Search Phase 2

Related Ideas

Wikimedia Foundation projects Information Retrieval Responsible for improving the search experience on Wikipedia. Group: Product and Technology Start: October 2025 Team members: Adam Baso, Eliza Blackorby, Dmitry Brant, Peter Fischer, Eric Gardner, Martin Gerlach, Tais Lessa, Justin Moore, Astuthi Odit, Mike Raish, Amal Ramadan, Sucheta Salgaonkar, Justin Scherer, Seddon, Hilary Swan, Sherry Yang Management: Jazmin Tanner ← Reading/Web

Improving How Readers Find Answers on Wikipedia

The problem

This page documents how we are exploring making it easier for readers to find the information they’re looking for directly on Wikipedia, including when they type full questions or natural phrases instead of precise keywords.

Recent research shows 4–7% of all search queries on Wikipedia are phrased as natural language questions (e.g. “What”, “Who”, “How”). Question-style searches like "Who invented planes?” or queries such as “Biggest city in South America” are significantly less successful at leading users to the right article compared to simple keyword-based searches. When users move away from keyword-style queries on Wikipedia, the results can feel buried, incomplete, or unexpected.

A reader wanting to know who invented planes and searching in-Wikipedia search; e.g., "Who invented planes".

Additionally, many readers turn to external search engines or AI-powered tools to reach Wikipedia content, or even to get synthesized answers based on Wikipedia rather than starting their search here. An estimated 78% of reading sessions originate from external search engines. with 90% of those coming from Google Search These external tools excel in this area because they analyze the intent of the query rather than relying solely on direct text-matching.

When readers consistently find it easier to access Wikipedia’s content through third-party platforms than on Wikipedia itself, the site risks becoming a background data source rather than a destination for learning. Particularly, only 0.8–2% of pageviews are from internal search. This matters because it contributes to a decline in on-wiki readership and affects how people experience and value Wikipedia. Over time, it also weakens opportunities for readers to understand article context, and explore related knowledge, and encounter, or even join, the volunteer community that builds the encyclopedia.

Improving how readers find information during their first visits helps them experience Wikipedia as a place for discovery rather than just a source reached through others. Doing so supports the Foundation’s broader Readers strategy to increase reader retention: when new or casual readers can easily find what they’re looking for, they are more likely to return, become active readers that explore more deeply and eventually become tomorrow’s contributors.

What We're Exploring Together

Before considering production-level changes, it is important that we understand from the community’s perspective:

• If the problem we seek to address is real
• Potential risks or alternative ideas
• What success looks like

To learn more about community perspectives, we’d especially appreciate feedback on these foundational questions:

• What, if any, examples or patterns have you noticed as it relates to readers (or you personally) struggling to find specific information on Wikipedia? Alternatively, what has worked well?
• Are there existing community led initiatives or tools you would like us to learn from as it relates to making information easier to find?
• If we explore showing short excerpts or sections from articles in search results, what safeguards or criteria would you like us to consider in guiding that approach?
• What would success look like to you?

Risks and Considerations

We are mindful this work raises important questions. Early risks we want to be attentive to include:

• Context loss: Could excerpts misrepresent article intent?
• Attribution and trust: How can we ensure sources are clear and verifiable?
• Retrieval quality and representation: If meaning-based retrieval methods are used to better understand reader queries, how can we ensure they don’t disadvantage underrepresented topics, smaller languages or nuanced content areas?
• Overreach: How do we ensure this complements, not replaces, community-authored knowledge?
• Infrastructure and privacy: Can testing occur within the Wikimedia systems without compromising user data and our values?
• Editor Experience: How do we ensure information retrieval improvements for readers do not negatively impact editors, for example, by changing how they find, link to, or verify content within their workflows?

We know that there are likely other risks we haven’t yet considered or encountered. As we learn more this section will evolve and be updated. We invite you to share with us any risks you think we should be aware of.

Getting Involved and Staying Informed

This work supports the Wikimedia Foundation's Readers annual plan objective:

"Readers from multiple generations engage, and stay engaged, with Wikipedia, leading to measurable increases in retention."

Below are some ways to get involved with this effort:

• Share your perspective on the problem space as a whole and as many of the questions in the What We're Exploring Together section that you take interest in.
• Add this page to your Watchlist to follow updates.

Additionally, if you’d like the team working on this problem space to join a synchronous meeting or space off-wiki, please do not hesitate to let us know on the project talk page.