Skip to content
Artificial intelligence. Clearly. RSS feed
AI NEWS AI

News. Evidence. Perspective.

Clearer questions improved a tech-support search

Researchers tested search answers and users’ judgments about instructions in separate experiments.

An ASSETS 2026 study tested 48 resolved technology questions from a diary study of 27 older adults. Experts found correct answers within the first 20 Google results for 68.8% after GPT-4o reformulation, versus 35.4% before.[1]

The model asked up to three follow-up questions; researchers supplied answers from the original support conversations. Model evaluation ran from December 2024 to March 2025.[1]

Finding instructions is one stage of support. Separate research published in September by Britain’s Department for Science, Innovation and Technology distinguishes guiding someone through a task from completing it for them. Its January survey of 2,261 informal supporters found 57% said they repeated tasks despite having demonstrated them before.[2]

Interviews with 36 supporters and 36 people receiving help showed why approaches changed: teaching required patience and time, while urgency or anxiety could prompt a helper to take over. The survey cannot establish that teaching caused greater independence.[2]

In the AI study’s later experiment, 34 older adults judged instructions already rated correct by experts. Responses were 81.6% yes and 13.1% maybe about being able to follow them. Participants did not attempt the fixes.[1]

Follow the evidence

Sources & further reading

  1. Informal and proxy support for digital inclusion: final report The Social Agency / Department for Science, Innovation and Technology · Accessed 29 Oct 2026