n8n for Recruiting Automation

Primary intent: use n8n as of early 2026 as a workflow engine for recruiting operations: triggers (webhook, schedule, polling) drive nodes that transform, branch, and call HTTP APIs, so rows, messages, and tasks move on rules you can read off a graph.

n8n is not a replacement for your ATS or for judgment on candidates. It is the plumbing between systems once the policy, field mapping, and review gates are clear. Teams win when they automate one measured loop (time saved, error count, reply quality) before they chain five vendors.

Self-hosted and cloud deployments both exist; secrets, logs, and subprocessors still need the same security conversation as any iPaaS (see OAuth and API security). Pair n8n with LLM steps only after prompts and output shapes are stable in chat (see prompt chain).

If you are choosing between paste-first AI and automation, read How it compares to similar tools below, then follow Practical steps with a read-only path before any node writes to mail or the ATS.

Side-by-side tool notes: ChatGPT, Claude, Gemini, Cursor. Full tools directory.

If you are new to workflow automation for TA, ship one flow that reads more than it writes, measure errors for two weeks, then widen scope. Feature lists change; the table below is about recruiting-shaped jobs, not benchmark scores.

Where to start (opinionated): if your problem is only better wording in a doc, stay in Gemini or ChatGPT until the template is boring. If the problem is the same ten steps every Monday, pilot n8n on a read-only Slack digest first. If engineering already standardises on Git for runbooks, draft the node list in Cursor, then implement in n8n with code review on credentials. When Zapier or Make already passes security with the same vendors, switching is a procurement question, not a moral one.

A 15-minute first workflow (read-heavy, no candidate mail)

1. Create an n8n account (cloud or self-hosted per IT) in a workspace TA ops can access, not a personal trial tied to one recruiter's inbox.

2. Pick one trigger you can replay safely: for example Manual trigger with a sample JSON row, or a Google Sheet watch on a sandbox tab with fake candidates.

3. Add three nodes only: trigger → Set (normalise field names) → Slack (or Email to yourself) with a short template that lists role, source, and timestamp. Do not attach full CV text until legal agrees.

4. Run ten test executions. Count duplicates, empty emails, and wrong stage values. Fix the Set node rules until the noise drops.

5. Add an IF node that stops the path when must-have fields are missing, and route misses to a dead-letter channel so they do not vanish.

Optional: add an LLM step later

After the shape of the JSON is stable, insert an OpenAI, Anthropic, or Google Gemini node that only reads already-validated fields. Log model name and prompt version on each run for audit (see structured output).

Second prompt: workflow red-team (before go-live)

You are a TA automation reviewer. Below is a plain-language outline of an n8n workflow (nodes and purpose). List the top ten failure modes: duplicate sends, PII leakage, wrong stage writes, missing rollback, token expiry, rate limits, and GDPR issues. For each, mark SEVERITY High/Med/Low and one mitigation. Do not invent vendor features; if an assumption is missing, write UNKNOWN.

WORKFLOW OUTLINE:
[paste]