AI integration in LinkedIn’s core search experience

Context

The product team identified a gap with members engaging with search queries serendipitously. Understanding this gap, we hypothesized that integrating generative AI into LinkedIn's core search experience could allow members to interact with serendipitous searching. Reimagining a new search experience, the product team took a bold approach by creating a novel concept experience that leveraged generative AI into LinkedIn's core search experience to explore with participants and see if this would be a viable solution for LinkedIn members. 

Research goals

• Evaluate the discoverability with the new placement of the search navigation experience at the bottom of the mobile screen. 
• Evaluate members’ attitudes and perceptions by relocating the placement of the search navigation experience to the bottom of the mobile screen. 
• Surface attitudinal research insights by combining generative AI and search navigation into one seamless experience. 

Research questions

• How do users react to a unified input box that combines both generative AI and Search? What are their concerns and expectations?
• Do members know where to go for Search?
• Do members understand the generative AI sparkle icon is connected to the input box?
• What are members' perceptions of interacting with Coach to perform search queries?
• What are members' perceptions of removing the search box from the mobile experience?
• What are the most critical concerns regarding privacy and data security that members might have, and how can we address these concerns?

Stakeholders

• VP of Consumer
• Director & Principal PMs
• Manager & Principal Designers
• Directors of UXR
• Staff UXR for generative AI

Timeline

• 2 months total.
• 1.5 months to workshop with stakeholders and solidify concept design for testing.
• 1 week to design research study
• 1 week to recruit and perform concept test. 
• 3 days to analyze and synthesize data for final deliverable.

Recruit audience

Considering this concept was a bold and new search experience, we selected a general audience familiar with LinkedIn's search experience. Having feedback from this type of audience would be valuable since this concept could impact a core experience on LinkedIn. More importantly, the team wanted to collect insights in a fast and iterative manner with the concept still in its early phase. 

Methodology & Research Approach

Research method

After workshopping with key stakeholders, an unmoderated concept test using the UserZoom platform was recommended. This method was selected for many reasons: 

• UserZoom provided recruiting from a general audience pool.
• Performing an unmoderated concept test would allow the team to receive research insights quickly. 
• Leveraging UserZoom would allow the product team to duplicate the research design and test iterative designs in an agile manner.  

Research approach

We conducted 30, 20-minute unmoderated concept test sessions by testing four different concept tasks using Figma with follow-up sections of open-ended, participant-driven conversational prompting/experimentation (to answer foundational questions):

• Scenario 1: Searching for a member's profile with only one search query using conversational command prompts.
• Scenario 2: Searching for a member's profile by regenerating and refining their search using conversational command prompts.
• Scenario 3: Searching for Company/Organization and exploring additional information using conversational command prompts.
• Scenario 4: Searching for a specific topic and exploring additional information using conversational command prompts.

Sample size

Considering the limited bandwidth that participants exhibit during unmoderated research sessions, we chose a total sample size of n=30 to accomplish the needs of the research.

• We sampled N=30 general audience members who were from the US. Additional participant criteria: Familiar with the LinkedIn app and search experience, a mix of demographics (age and gender).
• We exposed n=15 participants to Scenarios 1, 2, and 3 and exposed n=15 participants to Scenarios 1, 2, and 4. Additionally, we exposed participants 

Impact

The insights from this study gave the product team confidence to move forward with exploring the concept further to create it into an MVP with LinkedIn's core search experience.