UX Research · Indeed
Finding the friction in Indeed’s job search
A usability study of Indeed’s search filters, benchmarked against Glassdoor and LinkedIn with 10 participants.
Role
UX Researcher
Team
Team of 5
Timeline
Sep – Dec 2021
Participants
10 + 2 pilots
Overview
Overview
Our team ran a thorough evaluation of the job search feature on Indeed.com. Through heuristic evaluation and usability testing with 10 participants, we identified pain points and potential improvements to the site’s search filters. Our analysis combined quantitative and qualitative data, so we could give actionable suggestions to improve the platform’s experience.
10
moderated tests
2
pilot tests
3
job sites compared
3
sessions I moderated
My role
As part of a group project for Indeed.com, I worked with four other researchers on the study. I moderated three sessions and took notes for the rest, and we presented our findings to Indeed’s design and research team.
What I worked on
Moderator script
Recruiting participants
Moderated interviews
Note-taking
Affinity mapping
Quantitative analysis
Qualitative analysis
Presenting to Indeed
Process
01
Heuristic evaluation
We evaluated Indeed’s functionality and usability, compared notes, and identified a user flow worth testing further.
02
Usability testing
We tested that flow in moderated Zoom sessions on Indeed and two competitors, Glassdoor and LinkedIn. Participants then completed a SUS questionnaire and answered verbal questions.
03
Data analysis
We analyzed the questionnaire data with statistical methods, and sorted insights from interview notes into an affinity diagram in Figma.
Evaluation
Heuristic evaluation
We looked for potential usability problems on Indeed’s desktop site using Nielsen Norman Group’s 10 usability heuristics.
Example from the heuristic evaluation
Rule 5 · Error prevention
Severity 2
Leaving the search bar blank returns a list of random jobs. Suggestion: prompt users to set filters if they’re unsure what job they want.
Rule 10 · Help and documentation
Severity 0
The tip shown when location is left blank is helpful.
Areas of improvement
01
Consistency and standards: buttons and phrasing styles are inconsistent across the site, which can cause confusion.
02
Visibility of system status: some elements are awkwardly laid out, leading to a suboptimal use of whitespace.
03
Flexibility and efficiency of use: the application format is restricted in many steps, and customizing job-search conditions and notifications is limited.
Recommendations
01
Let users filter: offer job filters (pay, location and so on) on the landing page, and let them choose which job types they get email alerts for.
02
Style elements consistently: standardize buttons and suggestions across the site to minimize confusion and errors.
Testing
Testing plan
01
Moderator script
1
Overview
2
Rules of the road
3
Tech setup
4
Persona reading
5
Instructions
6
Task 1
7
Task 2
8
Task 3
9
Post comparison
02
Recruiting participants
We recruited eligible participants through a screener survey and grouped them into students and non-students.
P1
P2
P3
P4
P5
P6
P7
P8
P9
P10
4 non-students
6 students
4 women, 6 men · ages 21–31
03
Testing sessions
Each session ran about 45 minutes. Participants did the same three tasks on Indeed, Glassdoor and LinkedIn. We wanted to see whether they’d intuitively use each site’s filters to narrow their results.
~45 min per session
3 tasks × 3 sites
Moderated over Zoom
Affinity mapping
Categorization
To make sense of the qualitative data, we built an affinity map that sorted feedback by job search site.
Indeed-specific feedback
Indeed feedback split into “to be improved (filters)”, “to be improved (others)” and “doing well”, showing frustration and confusion with Indeed’s filters.
Commonality in comparison
Participants shared a lot of common feedback when comparing the three sites, and the affinity map reflected it.
Findings
System Usability Scale
Average SUS score per site, with each site’s grade and 95% confidence interval.
Indeed
75.5
Grade C
CI 63.8 – 87.2
Glassdoor
67.5
Grade D
CI 53.4 – 81.6
79.75
Grade C
CI 67.7 – 91.8
LinkedIn scored highest, Glassdoor lowest
Across all 10 participants, LinkedIn had the best average SUS score and Glassdoor the worst.
But P1 and P3 rated Glassdoor highest
Even though Glassdoor averaged lowest, two participants gave it their top score. Both turned out to be non-students.
SUS by group
Non-students
Students
Indeed
68.13
80.42
Glassdoor
74.38
62.92
73.13
84.17
The two groups showed opposite trends
Split by group, Glassdoor scored highest with non-students but lowest with students.
Non-students weigh pay and benefits more
Full-time job seekers paid more attention to benefits and salaries, which Glassdoor handles very well.
Where each site lands on the SUS scale
Timing analysis
Expected task time limit: 7 minutes. We excluded outliers outside the 95% confidence interval to get a more precise average.
Indeed
4 min 34 s
Average task time (274.5 s)
Geometric mean
4 min 20 s
Glassdoor
4 min 19 s
Average task time (259.4 s)
Geometric mean
4 min 6 s
Fastest
4 min 5 s
Average task time (245.0 s)
Geometric mean
3 min 47 s
With only 10 participants the sample is small, so we also used the geometric mean as a middle value to better estimate the trend.
Order may be a main effect
Sites were always tested in the order Indeed → Glassdoor → LinkedIn, so participants got more familiar with the task as they went. P3, P6 and P10’s task times dropped as they progressed.
Why does Glassdoor’s task time vary so much?
1. How well does the participant set the filter?
2. Did they get lost on the “company page”?
Fastest single task: 117 s
Slowest single task: 387 s
Success rate
Indeed
96%
Most successful
Glassdoor
86%
84%
Indeed and Glassdoor list more jobs than LinkedIn, so participants couldn’t always find an ideal job on LinkedIn. Indeed’s results were more precise than Glassdoor’s, so participants found the jobs they expected more easily.
2
tasks failed due to time-out
4
participants fully succeeded in all tasks
7
tasks were partially successful
Two main reasons:
1. Not satisfied with the job they finalized
2. Some differences from the persona
Preferences
Which site participants preferred overall, and why.
70%
20%
10%
LinkedIn 70%
Indeed 20%
Glassdoor 10%
Most preferred
Most common
Least redundant information
Easiest, clearest and most intuitive
Social networking and a personal resume profile
Least preferred
Glassdoor
Lacks description
Unintuitive navigation
Unclear UI and information
Confusing jump from results page to company page
The full scorecard
Metric
Indeed
Glassdoor
Success rate
96%
86%
84%
Ease rate
70%
20%
70%
Average ease score (1 easy – 5 hard)
2.1 / 5
2.8 / 5
2.0 / 5
SUS score
75.5 (C)
67.5 (D)
79.75 (C)
Site preference
20%
10%
70%
Best result for that metric
Insights
Insights from testing
Positive feedback
01
Clean, minimalist UI that avoids clutter and information overload.
02
Conspicuous, intuitive filters make searching easy.
03
Skill-set filters are a unique feature that sets the site apart.
04
Results are accessible without signing up or logging in, preventing frustration and abandonment.
Negative feedback
01
Search results don’t accurately match the selected filters, so irrelevant postings show up.
02
Job descriptions could be better organized; a lot of information is packed into a small card.
03
Filters could be more accurate, complete and user-friendly.
Suggestions
01
Allow multiple selections in filters, such as several companies or locations.
02
Make sure results match the selected filters so irrelevant postings don’t appear.
03
Offer a sliding range for preferred salary instead of fixed options.
Reflection
Takeaways
What I’d do better
01
Randomize the order participants test the sites, to avoid order bias toward sites based on position.
02
Test with more participants for a larger sample and more statistically significant insights.
03
Simplify the persona information shown during testing, so participants don’t feel obligated to use all of it.
What I learned
01
My first full UX research project made me familiar with the process and methods like benchmarking, heuristic evaluation and moderating tests.
02
Technical checks are often overlooked but are a crucial part of usability testing.
03
Make sure participants’ equipment works and is compatible with what they’ll be testing.
04
Tech checks also let participants warm up and let researchers build rapport, making the session less intimidating.