The New Siri’s Real Power Is Access to Personal Context

Summary

Caleb Sponheim from NN/g argues that Apple's new Siri AI (beta, September 2026) derives its advantage not from raw model intelligence, but from its on-device Spotlight index. This index allows for the integration of personal messages, photos, and app content into responses. ChatGPT and the older version of Siri cannot achieve this without the user manually pasting data.

Why it matters

The article shifts the AI assistant race from the question of 'who has the smartest model' to 'who can act on the user's own data with the least friction.' This distinction is critical for every product strategy in the agentic AI space. The text also highlights a concrete user experience failure mode: because privacy-focused apps like Signal are excluded by default, context can be silently lost. This leads to a dangerous ambiguity for users between 'I couldn't find it' and 'it doesn't exist.' For designers and product teams, this serves as a comprehensive checklist: context access, model capability, transparent source attribution, and honest messaging about limitations must all be present together.

Highlights

• Siri AI found a brother's messages about three restaurants and correctly noted that the user had declined those invitations. ChatGPT, however, searched connected work accounts and found no relevant results.
• When the user asked for 'photos from last weekend in New York,' the old Siri returned generic web images and ChatGPT returned news links. Siri AI, on the other hand, retrieved the actual travel photos without the user needing to specify 'my photos.'
• Siri AI missed a message about Saturday plans in Signal because the app is not indexed by default. The response also failed to explain which sources were searched, creating confusion between 'not found' and 'does not exist' scenarios.

Critical take

The evaluation is based on a single user's three tasks. The author explicitly acknowledges that the results cannot be disentangled into contributions from the model versus the retrieval system. Therefore, the thesis that 'context beats intelligence' is more of a user experience framework than a measured finding. Additionally, the article serves as an indirect marketing funnel for NN/g's paid 'Designing AI Experience' course, which somewhat undermines the otherwise useful critical lens on Apple's distribution process.