Spoken Attribution Update for the Knowledge Space

A few years ago, I was tasked with rethinking how Siri handled spoken attribution for web-based World Knowledge answers.

This work was driven by a new tool which could extract and combine key information from multiple web sources into a single, conversational response. The answers were more natural, but attribution became harder to handle gracefully. World Knowledge responses could pull anywhere between 1 to roughly 3-4 sources. Verbally listing sources at the beginning added friction and made otherwise fluent answers feel not as user-friendly.

Leading with sources no longer made sense when the answer itself was synthesized.

The problem wasn’t whether to attribute. It was how to do it without interrupting the experience or undermining trust.

My Role & Outcome

I designed when and how Siri would give credit for information and decide when to mention it, moving attribution from the beginning of the response to the end while introducing a simple, scalable system based on source count.

The result: Siri went from saying “Here’s what I found from {source} …” to [see below] …