Across roughly 5.5 million captured conversations, one number shows that AI-generated scorecards have landed with recruiters: interviewers keep 77.4% of the scorecards Metaview drafts from the conversation.¹
A second number shows where the value comes from. Scorecards that start as a generated draft get submitted 50.3% of the time, against 28.6% for the ones interviewers write from scratch, in a same-denominator comparison of 93,502 generated and 26,498 manual scorecards.² That is a 21.7 point gap in how often a scorecard gets submitted at all.
That gap is the argument of this piece. The biggest thing a team can change about a scorecard is where it starts.
The comparison is observational. It shows association, and the causal question stays open, so treat the starting point as a lever to test on your own team.
It is a large gap, and it points at a workflow choice any team can make this quarter.
Teams keep the draft
Start at the document level. In a sample of 285,859 AI-generated scorecards, interviewers dismissed only 22.6% outright.¹ They rarely throw away the draft as a whole.
Keeping the draft is only the first step. Each generated scorecard is built from individual suggested fields: a competency rating, a piece of linked evidence, a summary line. Metaview drafts them from the conversation, and the interviewer decides what happens to each one before the scorecard is submitted.
When teams run that review loop, the payoff shows up as time saved. Nitin Moorjani, who runs Talent Operations at Automattic, put a number on it in the company's Metaview case study:
The most clear impact is the time saved. Recruiters save 20 minutes per interview from wrangling notes and submitting scorecards. Per month, that’s 53 hours saved in total.”
What starting from a draft changes
The association our scorecard completion study documents has two parts: the submission gap at the top of this piece, and a field-count gap to go with it.
Scorecards that began as generated drafts carried 7.85 completed fields on average. Scorecards interviewers wrote from scratch carried 2.61.³ The draft arrives fuller, and it gets submitted more often.
Our AI Notetaker captures every spoken word of the interview, so the draft arrives with each suggestion linked to what the candidate said. Reviewing it means reading and adjusting what the draft already contains, a lighter task than rebuilding the conversation from memory.
The interviewer fixes what’s wrong and fills what’s missing, and no rating gets made for them.
Why scorecards come back thin
Even when the draft is kept, plenty of scorecards come back sparse. Across a sample of created scorecards, interviewers filled a mean of 7.20 of the 15.17 fields a template carries.⁴ That is roughly half the template left empty, on average, which is the normal state of hiring documentation more than a sign anything broke.
The data records what got filled. It doesn’t record why the rest didn’t. So this section is a short list of ordinary suspects, and the data can’t tell us which one dominates.
- Templates outsize interviews. A single interview rarely produces strong evidence for every field a template lists, and the leftover fields are the easiest thing on the screen to skip.
- The review lands late. A draft opened after the debrief, or the next day, meets an interviewer whose decision already feels made. Filling the remaining fields reads as paperwork at that point.
- Suggestions arrive in bulk. Clearing a stack of prefilled fields takes a click each. Weighing each one takes attention, and the next call is already competing for it.
- Trust is per field. An interviewer who disagrees with one suggested rating may clear the rest rather than audit them one by one.
A suggested field is evidence to check, and an interviewer who clears a suggestion the conversation never supported is doing exactly what the review step exists for, so some empty fields are the review working.
A reviewed draft is better than an empty template, and better than accepting every suggestion. The number worth managing is whether suggestions get considered at all, and an empty-field count alone can’t tell you that.
What tracks with fuller scorecards
These are workflow choices associated with fuller, more complete scorecards. Three are worth testing on your own team, on top of making the draft the starting point.
Fit the template to the stage
Created scorecards in a platform sample carry an average of 15.17 fields.⁴ A recruiter screen, a technical deep dive, and a final panel don’t generate the same evidence, so a single template that size guarantees empty fields somewhere.
Trimming each stage’s template to what that stage can assess shrinks the surface a draft has to fill, and it makes every remaining suggestion easier to take seriously. Our structured interview guide covers how to pick those fields per stage, and the interview questions piece covers which prompts earn their place.
Review while the call is fresh
Submission timing in the data is heavily skewed. The median submitted scorecard lands 2.32 hours after the interview.⁵
The average is far longer, near a day and a half, pulled out by a long tail of scorecards that land days after the call. A draft reviewed in the quarter hour after the call gets checked against a fresh memory. The same draft opened next week meets a colder memory and probably gets less attention.
Booking the review into the interviewer’s calendar, as part of the interview rather than admin after it, is the cheapest experiment on this list.
Track what happens to suggestions
You can’t manage a review step you can’t see. Metaview Reports breaks scorecard behavior down by team, stage, and interviewer: coverage, submission, recommendation mix, and how long feedback takes to land.
If one department leaves most suggested fields empty while another accepts nearly everything as-is, those are two different conversations to have. The report tells you which one you’re walking into.
Put the scorecard where the team already works
Adoption also depends on where the reviewed scorecard lands, not only on how it gets filled. Metaview connects to your ATS, calendar, and video tools through its integrations, so a reviewed scorecard is submitted into the system your hiring team already works in instead of one more tab.
Where to start
Pull 90 days of your own numbers before changing anything: the share of interviews that get a scorecard at all, the submission rate on generated drafts, and how completely those scorecards come back. One of those will be obviously the weakest.
Then run the cheapest fix first, which is usually the one closest to the interview itself. If generated drafts are not already your default starting point, that is the change to make first. Everything else on this list is a refinement of it.
See where your scorecard adoption stalls.
Review coverage, submission, and how completely scorecards come back across your own hiring team in Metaview.
Frequently asked questions
Does Metaview score candidates on its own?
No. Metaview drafts the scorecard from the conversation and links each suggestion to what was said. Recruiters and hiring managers review the draft, edit it, and make every rating and recommendation themselves.
How many fields should a stage template have?
The right number is the one a single interview at that stage can actually support. A recruiter screen and a final panel assess different things, so their templates should not be the same size. When the same fields come back empty at a given stage again and again, that stage is being asked for evidence the conversation does not produce, and those fields belong on a later stage or off the template.
How do I compare submission rates between two teams fairly?
The comparison only means something when the two teams are otherwise alike. Submission rate moves with interview stage and template size, so a final-panel team and a phone-screen team can look different for reasons that have nothing to do with review habits. Hold the stage constant, use the same denominator for both teams, meaning submitted scorecards over the interviews that should have produced one, and look at a window long enough to smooth out a slow week. A team that starts from generated drafts and one that starts from blank templates are not a fair comparison either, since the starting point is the variable you are trying to read.
What data is this based on?
Aggregated, anonymized Metaview product data across roughly 5.5 million captured conversations, about 5.2 million of them candidate interviews. Sample sizes for each figure are in the footnotes. No individual, company, or candidate is identifiable.
¹ Scorecard-level kept rate: 285,859 AI-generated scorecards; 77.4% kept, 22.6% dismissed. ² Submission comparison: 93,502 AI-generated and 26,498 manual scorecards; 50.3% of generated versus 28.6% of manual submitted. A same-denominator comparison.
³ Completed-field comparison: a separate 80,000-scorecard sample; generated drafts carried a mean of 7.85 completed fields against 2.61 for manual ones. ⁴ Scorecard completeness: 13,373 created scorecards; a mean of 7.20 of 15.17 fields filled. ⁵ Submission timing: 811,298 submitted scorecards; median 2.32 hours from interview to submission, mean 36.16 hours.
Source: Metaview’s corpus of roughly 5.5 million captured conversations (about 5.2 million candidate interviews), 2026. All figures are aggregated and anonymized. Every comparison in this piece is observational.
