AI Resume Scores Barely Predict Who Passes the Interview
For a year, AI graded every resume and ran the first phone screen for everyone who applied to a LatamCent role: 3,601 applications, 467 shortlisted. The surprise in the data is how little a strong resume score says about how a candidate actually performs on the call.
- 61% / 63%shortlist rate for Strong Yes vs. Yes resumes: the grade barely predicts the interview
- 77%shortlist rate for resumes scoring 95+, the one band that stands out
- 13%of applicants shortlisted before a recruiter stepped in
- ~2 minfrom passing the resume check to the first AI call
Key findings
- A strong resume barely predicts a strong interview. Resumes the AI rated Strong Yes were shortlisted after the call 61% of the time; those rated only Yes, 63%. Only resumes scoring 95 or higher stood out, at 77%. The resume grade decides who gets a call, not who passes it.
- 13% of applicants reached a shortlist with no recruiter time spent first. In the ten weeks with full tracking, 3,601 applications became 467 shortlisted candidates. The first AI call went out about two minutes after a candidate passed the resume check.
- The phone screen is where the funnel narrows, not the resume check. Of 2,242 candidates the AI phoned, 792 completed a rated interview. Most of the drop is reaching people, not rejecting them.
- The AI reached 4 in 10 candidates it phoned, every month. 4,542 of 11,354 candidates had a conversation of over 60 seconds, and the monthly rate never left the 37% to 44% band.
- The AI ran 261 hours of first-round conversations over the year. Every one was a first screen that no recruiter had to schedule, run or write up.
How our AI candidate screening pipeline works
LatamCent places Latin American professionals with US companies, mostly in B2B SaaS. Our busiest roles draw hundreds of applicants, and some more than a thousand, and the slowest part of recruiting used to be the first-round phone screen: finding a time, making the call, writing up notes.
Over the past year we handed that first round to AI. The pipeline below is what every applicant goes through today. This report covers steps 2 and 3, and how many candidates come out at step 4.
- ApplicationThe candidate applies to an open role on our careers page.Candidate
- AI resume checkAn AI model rates the resume against the role, from Strong Yes to Strong No.AI
- AI voice interviewCandidates who pass get a phone call within minutes. The AI introduces itself as an AI, and retries automatically if they miss the call.AI
- Recruiter shortlistA recruiter reviews everyone the AI rated Yes or Strong Yes, interviews them and presents the best to the client.Recruiter
The tools were switched on one at a time, so the numbers come from two windows. Call volume and reach cover the full year from November 8, 2025, when the AI placed its first call. Resume and interview verdicts were only saved consistently from July 22, 2026, so the funnel and verdict sections cover ten weeks. Each figure says which window it uses. A companion report breaks down reach by month, the best time of day to call, and answer rate by role.
How many applicants make the shortlist after AI screening?
From July 22, 2026, every resume verdict and every interview rating was saved, so for those ten weeks we can follow candidates from application to shortlist.
Of 3,601 applications, 467 candidates (13%) reached a recruiter's shortlist. The AI resume checker passed 2,287 (64%). The AI interviewer phoned 2,242 candidates, including 34 the resume checker had rated No, and completed a rated interview with 792. Of those, 467 were rated Yes or Strong Yes. The biggest drop is between phoning a candidate and completing the interview: reaching people, not rejecting them, is where most candidates fall out.
The AI screening funnel, from application to shortlist
- Applications screened by the AI resume checker3,601
- Passed the resume checkRated Yes or Strong Yes2,287 64% of applicants
- Phoned by the AI interviewer2,242 62% of applicants
- Completed the voice interviewThe AI gave the call an overall rating792 22% of applicants
- Shortlisted for a recruiterRated Yes or Strong Yes on the call467 13% of applicants
Does a strong resume predict a strong interview?
The AI grades every resume from 0 to 100 and sends the Yes and Strong Yes verdicts on to a voice interview. The obvious question is whether a higher resume grade means a better interview. Mostly, it doesn't.
Candidates whose resume scored a Strong Yes were shortlisted after the call 61% of the time. Those rated only Yes were shortlisted 63% of the time, a hair higher, not lower. Interview completion was identical for both, at 38%. The resume verdict barely moves the odds of passing the conversation.
The grade carries signal in one place: the very top. Resumes scoring 95 or higher were shortlisted 77% of the time, while everything from 70 to 94 sits in a flat 56% to 65% band. A near-perfect resume means something. The difference between a good resume and a very good one mostly doesn't survive the phone screen.
Voice-interview shortlist rate by resume grade
| Resume grade | Interviewed | Shortlisted | Shortlist rate |
|---|---|---|---|
| Strong Yes verdictScore 86 to 100 | 813 | 493 | |
| Yes verdictScore 64 to 85 | 219 | 137 | |
| Score 95 to 100 | 146 | 112 | |
| Score 90 to 94 | 481 | 271 | |
| Score 85 to 89 | 200 | 115 | |
| Score 80 to 84 | 117 | 76 | |
| Score 70 to 79 | 88 | 56 |
This is why we don't let the resume check make the final call on anyone. It decides who gets a conversation. The conversation, and then a recruiter, decide who moves forward.
How fast is AI candidate screening?
The main thing AI changed for us is time. A candidate who passes the resume check now gets a call in about two minutes, while the application is still fresh in their mind. Over the year the AI ran 261 hours of candidate conversations, first-round calls that no recruiter had to schedule or sit through.
The slow part of recruiting used to be getting to that first conversation: finding a time, placing the call, writing it up. Handing it to AI took that off the recruiters' calendars and out of the candidate's wait.
How to use AI screening calls in your own hiring
If you're thinking about adding AI to your own first-round screening, or hiring through someone who already does, these are the lessons we'd pass on.
- Plan for 6 in 10 candidates not answering
A 40% reach rate held for a full year. Size your applicant pool for it, and don't read silence as lack of interest.
- Don't shortlist from the resume alone
Strong Yes verdicts ran 50% on resumes and 27% on voice interviews. Spoken communication is where most of the filtering happens.
- Keep a person at the end
The AI narrows 3,601 applications to 467. A recruiter still interviews each of them before a client sees a name.
Method and limitations
All figures come from LatamCent's own screening records: applications in our applicant tracking system, matched to call logs from the AI voice interviewer. The funnel and reach figures were pulled on October 2, 2026 and cover the year to October 1, 2026. The resume-grade analysis in Section 3 uses a later pull, on October 24, 2026, that runs through October 23, so its interview counts are larger. No AI calls were placed before November 8, 2025.
- Phoned
- An application that received at least one AI call. ~10,400 different people is a count of distinct phone numbers.
- Reached
- At least one call to the candidate lasted over 60 seconds. Used for the full year because ratings weren't saved before July.
- Completed interview
- A call the AI gave an overall rating to (July 22 onward). Stricter than reached.
- Shortlisted
- Rated Yes or Strong Yes on the voice interview.
- Conversation hours
- The total length of all calls over 60 seconds.
A few limits to keep in mind:
- This is one company's applicant pool: Latin American candidates applying to remote roles at US companies, mostly in B2B SaaS. Reach rates in other markets or for other kinds of roles may differ.
- Reached measures whether a candidate talked to the AI, not whether they were a good fit. We haven't yet linked shortlist verdicts to final hires. Our separate retention data by function covers what happens after placement.
- Ratings saved between July 3 and July 21, 2026 included scores for voicemails by mistake, so the funnel starts on July 22.
- The shortlist-rate-by-resume-grade figures in Section 3 match each candidate's resume grade to their own interview outcome, but the base is only candidates who completed a rated interview, not all graded resumes. Candidates graded in the last few days may still be waiting for a call.
- Between September 21 and 25, 2026, results for 172 calls weren't returned by the voice provider, so those calls weren't scored or retried. September figures are slightly understated as a result.
- 137 of the 22,259 calls had no application linked and are left out of the per-candidate figures.
Cite or reuse this research
This data is free to republish under a CC BY 4.0 license. Quote the figures, download the numbers, or embed the chart below in your own article, report or presentation. All we ask is a link back to this page as the source.
LatamCent. (2026). AI Resume Scores Barely Predict Who Passes the Interview. LatamCent Research & Insights. https://latamcent.com/research-insights/ai-candidate-screening-data/
Download the data
Download the aggregate figures (CSV), the screening funnel and the shortlist rate by resume grade, ready to drop into your own analysis.
Embed the key chart
Copy the code below the chart and paste it where you want the chart to appear. It renders live and links back here.
Frequently asked questions
In LatamCent's data, 40% of candidates the AI phoned were reached for a real conversation (a call longer than 60 seconds). That's 4,542 of 11,354 candidates between November 2025 and October 2026, and the rate stayed between 37% and 44% every month.
Yes. The AI interviewer tells candidates it's an AI at the start of the call.
No. In LatamCent's process the AI handles the first resume review and the first-round phone screen. Recruiters then review the shortlist, interview the candidates and present them to the client. The AI removes the scheduling and first-call workload, not the recruiter's judgment.
Once a candidate passed the AI resume check, the first AI call went out in about two minutes. Candidates who missed that call were retried automatically, and one in five of the candidates reached only answered on a retry.
Yes. You're welcome to quote the figures and use the charts with a link back to this page as the source. The suggested citation is in the Cite this research section.
Get a Shortlist That's Already Screened
Every candidate we present has passed the AI resume check, the AI voice interview and a recruiter interview. Tell us what you're hiring for and we'll walk you through how the search would run.
