Research Repository

Prompt Engineering | UX Research

Background

As the primary researcher for my team in my current role, I have conducted over 100 hours of moderated interviews and run dozens of studies. The results of these studies have been recorded in a number of different formats, ranging from formal slide decks to Mural and Lucid boards. Given the amount of research I've done, particularly in the area of aftermarket products, I wanted to find a way to organize all of my findings in one place. Using Claude, I was able to code a searchable research repository that I could share out with other members of my team.

Initial Prompts

I started by gathering all of my various studies - some in slide decks, others screenshotted from collaboration boards on Mural, Lucid, or Figma. I then uploaded these to Claude, asking it to create an expandable card for each one. I refined my initial prompt several times, clarifying exactly what I wanted, before moving forward with the results it gave me.

Each card contained a title, a date, contributor names, the goal of the study, a summary of the results, and more. Some cards were more accurate than others - in many cases, Claude emphasized findings that were less pertinent and overlooked important nuances. I also encountered a few hallucinations, including falsely attributed user quotes. Because I was involved in all of the studies, and there were a limited number of them, I was able to proofread each one for accuracy. However, it did raise concerns for me about creating a repository like this on a larger scale.

Personas & Themes

I also created two additional tabs within the repository. One contained Personas, originally created in a Mural board but reformatted by Claude into expandable cards. The other contained overarching themes found throughout multiple studies. Identifying these common themes was one of the most valuable things I was able to do using Claude’s capabilities, as Claude was able to compare a dozen studies at a time in a way that a human wouldn’t be able to. I asked Claude to cite the studies that it pulled each theme from, and its reasoning behind doing so. The themes it surfaced were consistent with themes I had noticed over time - none of them were groundbreaking discoveries, but the ability to see which different studies these themes appeared in was helpful.

Expanding the Repository

After proofreading my initial batch of studies, I created a prompt for adding studies on an individual basis. Some of these studies were missing data, and I found that while Claude still insisted on adding incorrect placeholder data, it did learn to ask me if I would like to edit the placeholder data rather than trying to pass it off as accurate and wait for me to call it out as a hallucination.

As I added more studies, I decided to add a keyword search, so that users wouldn’t need to scroll through the entire repository to find what they were looking for. I had Claude add a basic search bar, as well as specific keyword filters. When the user searches or filters by a keyword, that word is highlighted within the study card.

Accessibility

When I had finished building the repository, I wrote a prompt asking Claude to run an accessibility audit using current WCAG guidelines. It presented a list of fifteen recommendations, ranked by priority, that would improve the accessibility of the repository. Many had to do with ARIA attributes, but it identified a few contrast and labeling issues as well. I asked Claude to implement these changes, resulting in a more accessible, usable experience.