I Built a Voice Diary-Study Instrument to Study Trust in AI

OVERVIEW

I designed and built the participant-facing diary-study instrument and its Custom GPT–Zapier–Google Sheets workflow for planning real errands, immediate reflection, and first-pass coding. In the team’s exploratory study, six participants contributed 24 entries. Findings are directional rather than representative of New Yorkers broadly.

The custom GPT diary-study instrument's opening screen, prompting the participant to log their first entry

TIMELINE

Spring 2026

ROLE

UX Researcher & Ethnographer

TEAM

Merlyn Koonamparampath, Eric Lopez, Conor Mack

FOCUS

UX Research

Voice AI during real errands

This client project through Pratt investigated how people interact with voice AI during everyday errands. My contribution was the instrument and its pipeline; the broader study, synthesis, and client recommendations were team work.

The study explored assistance, trust, autonomy, and interaction friction. It does not establish representative findings about New Yorkers, older adults, or people with mobility needs.

One instrument, three jobs

Split-screen view of the custom GPT's configuration panel and its 'Navigation Diary Study' participant-facing preview

The diary study instrument, built as a custom GPT with three simultaneous roles.

I owned the instrument’s design, implementation, troubleshooting, and iteration. It helped with an errand, invited reflection within the same conversation, and submitted structured entries through the Custom GPT–Zapier–Google Sheets pipeline.

Six participants contributed 24 diary entries about errands they were already running, rather than standardized assigned tasks. The workflow supported first-pass coding for human review; automation was not a guarantee of correct coding or anonymization.

Keeping reflection inside the conversation was intended to reduce switching between tools. We did not establish a measured reduction in participant burden.

What this small study surfaced

The team found that participants delegated discovery while retaining decisions about what to do. That suggested an autonomy boundary: help people explore options while keeping them in control.

The team identified failures to acknowledge or follow prior requests as an important trust and interaction problem in this small study. Interface friction and preferences around privacy and delegation suggested questions for further investigation.

The sample cannot establish how common these preferences are or whether age, digital literacy, or other factors explain them. These are exploratory findings, not validated demographic segments or population-wide rules.

What the team handed Woven

The team translated the study into voice-AI guidelines emphasizing autonomy, trust, and personalization. I owned the instrument and pipeline; the broader findings, synthesis, recommendations, and presentation were team work.

These are research deliverables and design directions. Client implementation, retention effects, and product impact were not established.

Limits and further questions

Errands were opportunistic, entry counts were unequal, and some reflection fields were missing. Findings are directional rather than representative, and ratings should not be treated as precise population estimates.

A further study could test how acknowledgment affects trust recovery and compare privacy and delegation preferences with a suitable design and sample. Those questions remain open.

This page is looked after by the Archivist, an agent I built that works from my primary research and can’t publish without me. Here’s how that works.

I Built a Voice Diary-Study Instrument to Study Trust in AI

OVERVIEW

I designed and built the participant-facing diary-study instrument and its Custom GPT–Zapier–Google Sheets workflow for planning real errands, immediate reflection, and first-pass coding. In the team’s exploratory study, six participants contributed 24 entries. Findings are directional rather than representative of New Yorkers broadly.

The custom GPT diary-study instrument's opening screen, prompting the participant to log their first entry

TIMELINE

Spring 2026

ROLE

UX Researcher & Ethnographer

TEAM

Merlyn Koonamparampath, Eric Lopez, Conor Mack

FOCUS

UX Research

Voice AI during real errands

This client project through Pratt investigated how people interact with voice AI during everyday errands. My contribution was the instrument and its pipeline; the broader study, synthesis, and client recommendations were team work.

The study explored assistance, trust, autonomy, and interaction friction. It does not establish representative findings about New Yorkers, older adults, or people with mobility needs.

One instrument, three jobs

Split-screen view of the custom GPT's configuration panel and its 'Navigation Diary Study' participant-facing preview

The diary study instrument, built as a custom GPT with three simultaneous roles.

I owned the instrument’s design, implementation, troubleshooting, and iteration. It helped with an errand, invited reflection within the same conversation, and submitted structured entries through the Custom GPT–Zapier–Google Sheets pipeline.

Six participants contributed 24 diary entries about errands they were already running, rather than standardized assigned tasks. The workflow supported first-pass coding for human review; automation was not a guarantee of correct coding or anonymization.

Keeping reflection inside the conversation was intended to reduce switching between tools. We did not establish a measured reduction in participant burden.

What this small study surfaced

The team found that participants delegated discovery while retaining decisions about what to do. That suggested an autonomy boundary: help people explore options while keeping them in control.

The team identified failures to acknowledge or follow prior requests as an important trust and interaction problem in this small study. Interface friction and preferences around privacy and delegation suggested questions for further investigation.

The sample cannot establish how common these preferences are or whether age, digital literacy, or other factors explain them. These are exploratory findings, not validated demographic segments or population-wide rules.

What the team handed Woven

The team translated the study into voice-AI guidelines emphasizing autonomy, trust, and personalization. I owned the instrument and pipeline; the broader findings, synthesis, recommendations, and presentation were team work.

These are research deliverables and design directions. Client implementation, retention effects, and product impact were not established.

Limits and further questions

Errands were opportunistic, entry counts were unequal, and some reflection fields were missing. Findings are directional rather than representative, and ratings should not be treated as precise population estimates.

A further study could test how acknowledgment affects trust recovery and compare privacy and delegation preferences with a suitable design and sample. Those questions remain open.

This page is looked after by the Archivist, an agent I built that works from my primary research and can’t publish without me. Here’s how that works.

I Built a Voice Diary-Study Instrument to Study Trust in AI

OVERVIEW

I designed and built the participant-facing diary-study instrument and its Custom GPT–Zapier–Google Sheets workflow for planning real errands, immediate reflection, and first-pass coding. In the team’s exploratory study, six participants contributed 24 entries. Findings are directional rather than representative of New Yorkers broadly.

The custom GPT diary-study instrument's opening screen, prompting the participant to log their first entry

TIMELINE

Spring 2026

ROLE

UX Researcher & Ethnographer

TEAM

Merlyn Koonamparampath, Eric Lopez, Conor Mack

FOCUS

UX Research

Voice AI during real errands

This client project through Pratt investigated how people interact with voice AI during everyday errands. My contribution was the instrument and its pipeline; the broader study, synthesis, and client recommendations were team work.

The study explored assistance, trust, autonomy, and interaction friction. It does not establish representative findings about New Yorkers, older adults, or people with mobility needs.

One instrument, three jobs

Split-screen view of the custom GPT's configuration panel and its 'Navigation Diary Study' participant-facing preview

The diary study instrument, built as a custom GPT with three simultaneous roles.

I owned the instrument’s design, implementation, troubleshooting, and iteration. It helped with an errand, invited reflection within the same conversation, and submitted structured entries through the Custom GPT–Zapier–Google Sheets pipeline.

Six participants contributed 24 diary entries about errands they were already running, rather than standardized assigned tasks. The workflow supported first-pass coding for human review; automation was not a guarantee of correct coding or anonymization.

Keeping reflection inside the conversation was intended to reduce switching between tools. We did not establish a measured reduction in participant burden.

What this small study surfaced

The team found that participants delegated discovery while retaining decisions about what to do. That suggested an autonomy boundary: help people explore options while keeping them in control.

The team identified failures to acknowledge or follow prior requests as an important trust and interaction problem in this small study. Interface friction and preferences around privacy and delegation suggested questions for further investigation.

The sample cannot establish how common these preferences are or whether age, digital literacy, or other factors explain them. These are exploratory findings, not validated demographic segments or population-wide rules.

What the team handed Woven

The team translated the study into voice-AI guidelines emphasizing autonomy, trust, and personalization. I owned the instrument and pipeline; the broader findings, synthesis, recommendations, and presentation were team work.

These are research deliverables and design directions. Client implementation, retention effects, and product impact were not established.

Limits and further questions

Errands were opportunistic, entry counts were unequal, and some reflection fields were missing. Findings are directional rather than representative, and ratings should not be treated as precise population estimates.

A further study could test how acknowledgment affects trust recovery and compare privacy and delegation preferences with a suitable design and sample. Those questions remain open.

This page is looked after by the Archivist, an agent I built that works from my primary research and can’t publish without me. Here’s how that works.