How to Automate AI Resume Screening: Auto-Qualify Candidates + Schedule Interviews — n8n Workflow with n8n
Every résumé that lands in your inbox is a decision waiting to be made — and a founder or ops lead making it manually is burning 15 minutes per candidate on work an LLM does in four seconds. This work
Every résumé that lands in your inbox is a decision waiting to be made — and a founder or ops lead making it manually is burning 15 minutes per candidate on work an LLM does in four seconds. This workflow reads each incoming CV, scores it against your role criteria, drops the structured result into Notion, and sends approved candidates a Calendly link automatically. No copy-paste, no spreadsheet triage, no "I'll get to the inbox tonight." Here is how it works and how to build it in n8n.
The problem: hiring throughput dies in the inbox
The bottleneck in early-stage hiring is never the interview — it's the funnel before it. A single LinkedIn or job-board posting generates 80–300 applications. Each one needs to be opened, read, matched against your must-haves, ranked, logged somewhere, and then either rejected or moved forward. That's hours of unstructured work that lands on whoever has the least leverage to say no.
The failure modes are predictable: good candidates sit unreviewed for days and take another offer; screening criteria drift because a tired human applies them inconsistently at 11pm; and nothing gets logged, so you can't answer "how many senior backend applicants did we get last month?" The result is a hiring process that feels busy but converts poorly, and a founder who becomes the rate-limiter for the whole company's growth.
The solution: an autonomous screening pipeline
The template turns your inbox into the trigger for a fully automated pipeline. The moment a résumé arrives as an email attachment, n8n picks it up, extracts the text from the PDF, and hands it to an AI model with your role's scoring rubric. The model returns a structured verdict — a fit score, a short rationale, and a qualify/reject decision.
From there the workflow branches. Qualified candidates are written to a Notion database and immediately sent a Calendly scheduling link so they can book an interview without a single back-and-forth email. Rejected candidates are still logged — every application, scored and searchable, becomes a hiring dataset instead of a lost thread. The human only ever looks at pre-qualified, pre-ranked candidates who have already booked time.
Step-by-step: building it in n8n
The workflow is roughly nine nodes. Here's the spine and the configuration that matters.
1. Email Trigger (IMAP Email node). Point it at a dedicated hiring inbox — jobs@yourcompany.com — with "Download Attachments" enabled so the PDF arrives as binary data. Set the mailbox to poll every minute. Using a dedicated address keeps the trigger clean; you don't want the workflow firing on every internal email.
2. Extract From File node. Set the operation to "Extract from PDF" and reference the incoming binary property (usually attachment_0). This pulls the raw résumé text into a JSON field. For image-based or scanned PDFs, swap in an OCR step or route those to a manual-review branch — plain text extraction returns empty on scanned documents.
3. AI scoring (OpenAI / Anthropic Chat node or the AI Agent node). This is the brain. Send the extracted text with a system prompt that pins down your rubric: required skills, minimum years, deal-breakers, and the output schema. Force structured output — ask for JSON like {"score": 0-100, "qualified": true/false, "reasoning": "...", "top_skills": []} — and enable JSON mode or a Structured Output Parser so downstream nodes get clean fields instead of prose. Set temperature to 0.2 for consistent scoring; a rubric applied at high temperature drifts exactly like a tired human does.
4. IF node. Branch on {{ $json.qualified }} or a threshold like {{ $json.score >= 75 }}. True goes to the qualify path; false goes straight to logging.
5. Notion node (Create Database Page). On both branches, write a row: candidate name, email, score, reasoning, top skills, decision, and a timestamp. Map the AI's structured fields directly to Notion properties. Use a Select property for the decision and a Number property for the score so you can sort and filter your pipeline later.
6. Calendly / Send Email node. On the qualified branch, email the candidate a Calendly link (or use Calendly's API to generate a single-use scheduling link). Personalize the subject with the candidate's name pulled from the parsed data. This is the step that removes the scheduling ping-pong entirely.
Wire an Error Trigger workflow alongside it so a malformed PDF or an LLM timeout notifies you on Slack instead of failing silently. Test end-to-end by emailing yourself three real résumés — one strong, one weak, one edge case — and confirm the scores, Notion rows, and Calendly send all fire correctly before you point a live job posting at it.
The benefits: speed, consistency, and a real dataset
The obvious win is time. A pipeline that scores and routes 200 applications runs in the background for pennies in API cost, versus a full day of human triage. But the compounding wins matter more.
Consistency: the same rubric is applied to application #1 and application #200 at identical rigor. No fatigue, no drift, no unconscious bias creeping in on candidate 150.
Speed-to-contact: strong candidates get a Calendly link within a minute of applying, while your competitors are still letting résumés pile up. In a tight market, being first to the interview is a genuine edge.
A hiring dataset: because every candidate lands in Notion with a score and rationale, you can answer real questions — conversion by source, score distribution, how many qualified applicants you're actually getting — and tune your job posting based on data instead of vibes.
Common pitfalls to avoid
Over-trusting the score. The AI qualifies, it doesn't hire. Keep the human in the loop for the actual interview and offer. Treat the score as a ranking and filtering tool, not a verdict — and periodically spot-check rejected candidates to confirm your rubric isn't screening out good people.
Weak prompts produce weak scoring. "Rate this résumé 1-10" gives you noise. Spell out must-haves, nice-to-haves, and explicit deal-breakers, and ask the model to justify its score against each. A vague rubric is the single biggest cause of bad output here.
Scanned PDFs break extraction. The Extract From File node returns empty text on image-based résumés, which then score as unqualified for the wrong reason. Add a check: if extracted text length is under ~100 characters, route to a manual-review branch instead of scoring it.
No error handling. Without an Error Trigger, a single malformed attachment can stall the queue or drop a candidate silently. Build the failure path before you go live.
Compliance and privacy. You're processing personal data. Store only what you need, be mindful of GDPR/EEO obligations, and don't let the model make final rejection decisions in regulated hiring contexts — keep the audit trail in Notion so every decision is reviewable.
Build it once, test it against real résumés, and your hiring funnel screens itself while you focus on the candidates who are already on your calendar.
Ja construimos isso pra voce
Nao comece do zero. O AI Resume Screening: Auto-Qualify Candidates + Schedule Interviews — n8n Workflow e um workflow n8n pronto para instalar — conecta suas ferramentas em minutos, sem codigo.
Instalar por $49.0 →
