Case Study: 

Shopping on Search

Moderated usability testing + semi-structured interview (n = 7) 

What is Voice Access?

Voice Access, also called Voice Control on iOS, is a voice command interaction model that gives users hands-free control of their device. It is often used by those who have upper-body motor impairments.

Common Voice Access interactions include:

The Ask

Motivations for the study

Google's 2025 company-wide accessibility OKR required top product flows to be validated against accessibility standards, and Shopping on Google Search had not yet been assessed for upper-motor impaired users.

For users

Upper-motor impaired users (~10% of the population, spanning arthritis to spinal cord injuries) should be able to shop via Search independently and efficiently.

For the business

~95M consumers shop via Google search. Search earnings ($198B annually) make up 60% of Alphabet's revenue, with a potential $19B at risk if Shopping on Search is inaccessible to upper-motor users.

The underlying assumption

The team assumed flows were broadly usable, no upper-motor complaints had surfaced and other accessibility issues had already been addressed via the 2024 blind / low vision study. In reality, teams rarely hear about these barriers directly (designers are unfamiliar with them, cannot test AT themselves without expert help, and AT like VC relies on specific support systems that are complicated to prioritize).

Research question(s)

• Can upper-motor impaired users complete top Shopping on Search tasks independently and efficiently?

• What accessibility issues exist across the top 10 shopping user journeys?

• Which fixes should be prioritized to reach OKR compliance within the quarter?

Problem Solving Under Pressure

↪ Screening gaps and a tight OKR clock forced fast pivots that doubled our learnings

My Role

• Sole researcher: Full ownership of study design, recruitment criteria, moderation, analysis, synthesis, and presentation

• Support: 2 recruiting / operations assistants (sourcing, pre-screening, scheduling)

• Primary Stakeholders: Senior project manager (PM), product UX researcher

• Final reporting was a prioritized recommendation share out to a mixed audience of designers, engineers, and PMs

The Challenge(s)

• Voice Access participant screening: Participants conflate virtual assistants (Siri, Alexa) with Voice Access, and my screener questions were not specific enough, leaving fewer Voice Access users than initially planned. I coordinated with recruiting on signals to watch for during screening calls, then updated and disseminated the screening question bank.

• Tight OKR timeline: Actionable insights were due within 3 weeks of interviews, so I ran thematic affinity mapping against quantitative scores to move straight from analysis to prioritized fixes.

• Unplannable participant: A late interview confirmation meant I could not cancel a participant who did not regularly use Voice Access. I worked with product partners to quickly build questions on AI integrations with Shopping on Search, uncovering new insights and reducing future external research need.

Method Spotlight 

↪ Moderated usability testing

Goal of this research

Validate that Shopping on Search meets Google's accessibility standards for upper-motor users and provide prioritized recommendations to achieve compliance.

Usability Testing (n = 7)

• 7 upper-motor impaired users (2 Voice Access, 1 stylus, 4 no assistive technology)

• 10 tasks, based on common user journeys, measuring success rate (3pt scale - pass / partial pass / fail) and perceived ease of use (5 pt unipolar scales, target ≥ 3.5)

• Task scored based on number of major and minor risks, with quantitative scores cross-checked against qualitative theme frequency

Reasoning behind the method

• Pros: Live moderation captures how assistive technology actually behaves against real components, surfaces workarounds users would not self report, and lets me probe failures in the moment

• Cons: Recruiting assistive technology users for live remote testing is difficult (even joining a Google Meet is a barrier for some participants), which limits coverage of assistive technology types like keyboard only and switch access

Takeaway

• 6 of 10 top Shopping user journeys are Unhealthy

• Voice Access users completed only 3 of 10 tasks without difficulty and took twice as long

Voice Access number labels persist behind dropdowns

Blocked 2 top tasks, Voice Access users could not select dropdown options because base page labels stayed active behind them

↪ Recommendation

Disable number labels for base page under dropdowns

Voice Access number labels missing from search results

Blocked 2 top tasks, search results (the core of the experience) could not be selected via Voice Access

↪ Recommendation

Check all search results can be chosen via Voice Access number labels

“Brand Information” module tabs reset the entire page

Affected 2 top tasks for all users, every tab switch refreshed the full page and returned users to the top, forcing excessive re-scrolling that hits upper-motor users hardest

↪ Recommendation

Have the module itself refresh, or keep the user at the same point in the page

Impact & Decisions Made

Takeaway

My research turned an unscoped compliance goal into 3 shippable fixes the team scoped directly into their bug hotfix cycle.

What was assumed

User journeys were broadly usable for upper-motor users, no complaints had surfaced and the 2024 blind / low vision work had addressed known accessibility issues.

What we found

Voice Access users were blocked by labeling and refresh issues on 6 of 10 top user journeys, issue themes extend beyond tested assistive technology (keyboard only, memory difficulties), and several top issues only required minor component tweaks.

What was decided

The team scoped the 3 priority recommendations into their bug hotfixes, restoring healthy scores on 5 of 10 tasks toward OKR compliance. My findings also initiated a design systems update for dropdown and number label consistency across Search.

Downstream wins

• Improved accessibility on top user journeys projected a reduction in purchase drop off among studied users by 50%

• Identified a significant module bug affecting all users, not just assistive technology users

• AI integration insights from the pivoted session reduced future external research need