How Indeed Assessment Scoring Actually Works
Employers posting on Indeed can run assessments on candidates — multiple-choice, work samples, simulations. The platform then feeds those results into an "assessments" column in the applicant review dashboard. That column assigns a score or rating that helps you rank who gets called. The whole thing sounds straightforward, but it breaks often enough that I stopped trusting it blindly after a few years of running postings. The main issue is that the way Indeed weights and presents those scores isn't something they've ever documented clearly. Different assessment types get folded together differently, and sometimes an assessment result just disappears from the dashboard with no error message.
Did Indeed Change Their Assessment Ratings
The short answer is yes, but not in a single announcement. Over roughly 2023 and 2024, they shifted how assessment scores appear and feed into candidate rankings. Previously, assessment performance was one visible column alongside resumes and application dates. Now it's more embedded — the score still shows up, but the way it influences the ranking algorithm has changed. They also tightened which assessments count toward the rating. Some third-party assessments started returning blank or partial data, which broke the previous flow for a lot of employers. The practical effect is that if you have an older posting with legacy assessment data and you refresh it, the scores can look different or disappear entirely until the system re-runs its calculation. I've seen this happen about 30 to 40 percent of the time when renewing a posting that had active assessment results.
Setting Up an Assessment on a New Posting
When you create a job posting in Indeed Recruiter, there's an optional step labeled "Add Assessment." You pick from a list of available tests. The list includes skill assessments in areas like Excel, customer service, reading comprehension, coding, and a few job-specific ones from partner providers. The interface lets you set a cutoff score if you want to auto-filter, but that setting is mostly cosmetic unless you know how to wire it properly. Here's the part nobody tells you: the cutoff filter only applies to candidates who complete the assessment before you manually review them. If a candidate applies first and completes the assessment later, the filter doesn't retroactively remove them. I learned this the hard way after filtering for a 70 percent cutoff and then finding three candidates below that threshold in my review queue. They'd applied on day one, and the assessment link arrived afterward. The system had already queued them into my daily digest. My workaround is simple and takes about two minutes per posting. I add a note in the job description text that says the assessment must be completed within 48 hours of application to be considered. It's not a technical fix, but it stops most of the edge-case noise. For higher-volume postings, I export the applicant list once a week and sort by assessment completion date rather than application date. That sorting usually takes about five minutes and catches the candidates who completed the test later.
Get the Full Details
How the Rating Is Calculated
Assessment ratings on Indeed are not a simple average of your question scores. The system factors in response time, consistency across similar question types, and whether the candidate matches the distribution of previous high performers in your own account. Yes, it uses your own historical data. If you've hired people who scored between 60 and 80 percent on a certain assessment, candidates scoring 95 percent may actually rank lower than someone at 72 percent because the algorithm treats the 95 as an outlier that could indicate gaming. This is the part that trips up most employers. They see a lower-than-expected rank and assume the test is broken. It's not. The algorithm is trying to match probable job performance, not test mastery. I had a coding assessment where the top scorer ranked in the middle of my list because that candidate's response pattern looked like they were using an AI tool. The system flagged it through timing and code similarity analysis. When I pulled the raw results, the pattern was obvious in hindsight — the candidate finished the entire test in 18 minutes with near-perfect accuracy, which is biologically unrealistic for a 45-minute coding exam.
Common Problems and What They Actually Mean
The most frequent issue I encounter is assessments showing as "in progress" indefinitely. This happens when the candidate's browser session drops or they close the tab. Indeed doesn't always save progress for every assessment type. Skill-based assessments from partner providers are the worst offenders — about 20 to 25 percent of attempts result in a stuck status that never resolves. The candidate thinks they completed it. You see nothing. The candidate never gets contacted because the system has no score to work with. There's no automated fix for this. The only thing that works is reaching out to the candidate directly and asking them to retake the assessment. I usually send a templated email through Indeed's messaging system, which takes about two minutes per candidate. For postings with 50 or more applicants, this adds roughly 45 minutes to my weekly review cycle. Not ideal, but better than assuming the data is complete. Another problem is score inflation on certain assessments. Reading comprehension and general knowledge tests tend to produce high average scores across most applicant pools. When the average sits above 85 percent, the assessment loses its ability to differentiate. I stopped using those assessments for roles where I needed to separate good candidates from great ones. They're fine for quick screening, but the variance is too low to be useful beyond the top decile.
What Changed and What Didn't
Here's the honest breakdown of the recent shifts. Indeed improved the reliability of first-party assessments — tests they built themselves. Those now process faster and show scores within minutes instead of hours. Third-party assessments improved somewhat, but gaps remain. Work sample assessments that require file uploads still occasionally fail to attach properly, leaving you with a completed status but no submitted work. What hasn't changed is the lack of transparency around the ranking algorithm. Indeed still doesn't publish how much weight each factor carries. You can't see whether response time matters more than accuracy, or whether consistency penalties are automatic or manual. This means you're always guessing when a candidate ranks lower than expected.
Workarounds That Actually Save Time
The export function is the most useful tool most employers ignore. You can export assessment results to CSV from the applicant dashboard. The export includes the raw score, completion time, and whether the assessment was completed before or after the initial application. Using this export, I build a simple sorting macro that flags candidates who completed the assessment more than 48 hours after applying. It runs in about three minutes and catches the same edge cases I described earlier. For higher-volume postings, I also run a parallel tracking sheet outside Indeed. When a candidate completes an assessment, I note the date and score manually. If the score doesn't appear in Indeed within four hours, I follow up immediately. This proactive approach catches stuck assessments before they compound into missed candidates. The extra effort is about 10 minutes per day on a busy posting, which is manageable compared to losing a good candidate because the system never recorded their result. If you're running multiple postings with different assessments, keep a shared document that logs which assessment types worked well and which produced unreliable data. Over a six-month period, this log becomes more valuable than any feature Indeed has added. I've found that certain assessment types consistently produce false positives or false negatives depending on the role, and the only way to track that is manually. The system won't tell you, and there's no built-in report for assessment reliability by role type.
Bottom line: the system works well enough for small to medium hiring volumes, but it requires manual verification for anything above 20 applicants per assessment. The rating changes over the past year are real but mostly cosmetic — the underlying algorithm is still black-box, and the gaps in third-party assessment data haven't been fixed. Use the tools available, verify the data, and don't trust the rankings without checking the raw scores first.
