How to screen resumes with AI (and still send humane replies).
Job listings drown you in resumes. This is how to filter them without turning candidates into a spreadsheet: point an AI at inbound Gmail applications, let it score each against a job spec you set once, and get personalized drafts for both the advance-to-next-round and polite-decline replies.
- Every applicant gets a personalized reply, not a form email
- Consistent ranking against a job spec you write once
- Bias controls baked into the prompt — you can inspect and edit
- Drafts only; every send stays approval-gated
The math that makes hiring bad
A well-known senior role attracts 200-400 applications in the first two weeks it's posted. A recruiter or hiring manager has, generously, 3-5 minutes per resume for a first-pass screen. Do the math: 200 applications × 4 minutes is 13 hours of pure resume triage. That work almost never gets done at that pace, so what actually happens is a scattershot review — the first 30 read carefully, the next 100 skimmed for keywords, the last 70 barely opened.
The two costs of that pattern show up in different places. Good candidates get missed because their resume was in the 'barely opened' bucket. And every candidate — good, bad, mid — gets a slower or less considered reply than they deserve, which shapes how your brand reads on Glassdoor and how many good people apply next time.
AI screening flips this. Every application is read consistently against the same job spec, ranked, and paired with a drafted personalized reply — advance or decline. Recruiters review the ranked list rather than the raw inbox; drafts are edited-and-sent rather than written from scratch. Time per application collapses from 4 minutes of triage to 30 seconds of review, and consistency stops being a function of what time of day the resume happened to land.
The workflow works because it inverts the bottleneck. Manual screening bottlenecks on reading; AI screening bottlenecks on judgment. Recruiters make the actual hiring calls; AI does the reading.
How it works
- 1
1. Write the job spec in Notion
One page: role summary, must-haves, nice-to-haves, red flags. Claude uses this as the rubric for every applicant.
- 2
2. Point the template at your applications Gmail label
Any label works — 'Applications', 'Careers', per-role labels. Pilotran only processes messages under that label.
- 3
3. Review the ranked list and drafts
Ranked list lands in Notion. Advance and decline drafts land in Gmail. You send.
What a ranked candidate looks like
Sample: incoming application for a Senior Software Engineer role. Claude scored against the job spec, ranked, and drafted the personalized advance reply.
- Roles that consistently attract 100+ applicants per posting
- Any role where a written spec cleanly encodes requirements (engineering, design, marketing)
- Teams where recruiter time is the bottleneck for hiring velocity
- Organizations that want application-response times to hit under 48 hours consistently
- Executive or leadership hires — nuance matters more than throughput; do these manually
- Roles where the spec is intentionally vague ('we'll know when we see it') — AI needs a rubric to score against
- Highly regulated hiring (financial services, healthcare) with strict compliance requirements around automated screening — check policy first
- Very small hiring volume (5 applications per role) — the setup effort exceeds the savings
Notes from the build
The job spec IS the rubric
Every applicant is scored against the exact spec you wrote once in Notion. If the ranking feels off, the spec is off. Iterate on the spec, not the prompt.
Bias controls are inspectable
The prompt tells Claude to ignore name, gender, school prestige (unless directly required), and demographic signals. You can inspect and edit that prompt line. Nothing is hidden.
Drafts, always drafts
Every advance and decline reply is a Gmail draft your recruiter approves. The AI never sends. This is not a policy — it's a hard limit in the template.
Ranking is a suggestion, not a filter
The bottom 30% of applications aren't filtered out. Your recruiter still sees them in a 'lower rank' section of the queue. You keep the final say on who gets a real read.
Job Application Screener
Every 30 minutes, screen application emails against your job description. Filed to Notion; top scorers get a courtesy draft reply.
See the templateFrequently asked
Can it introduce bias?
AI resume screening carries known bias risks. This template uses spec-based ranking, gives you the rubric, and every decision is drafted — never auto-sent. You are still the hiring authority.
Does it read attached PDFs?
Yes. Gmail-attached PDFs, DOCXs, and links to LinkedIn/portfolio pages all get read.