AI Recruiting Academy › Module 5: Putting AI to work
The short answer
Pick one repetitive handoff and write it as a sentence: “when this happens, do that”. Check whether your ATS can already do it. If not, connect your apps with a no code automation tool such as Zapier, Make or n8n.
Add AI only to the steps that need reading or writing, such as summarizing an application or drafting a reply. Keep a person on every step that rejects someone, makes an offer, or sends a candidate something nobody has checked. Automate the paperwork, not the judgment.
Which recruiting tasks are worth automating?
The ones that happen often, follow the same steps every time, and do not need judgment. Scheduling is the classic example. GoodTime’s 2026 hiring report, a survey of more than 500 talent leaders by a scheduling software company, found recruiters spend 38% of their time scheduling interviews.
| Task | When this happens | Do this | Person checks |
|---|---|---|---|
| New applicant alert | An application arrives for a priority role | Post a short AI summary to the recruiter’s chat channel | Recruiter reads and decides |
| Interview booking | A candidate moves to the interview stage | Send a booking link and add the interview to calendars | Only exceptions |
| Reminders | An interview is booked | Remind the candidate and interviewer a day before | None |
| Scorecard chasing | An interview ends | Nudge the interviewer until the scorecard is in | None |
| Status updates | A candidate has been in a stage for seven days | Alert the recruiter to update the candidate | Recruiter writes or approves the message |
| Offer paperwork | An offer is approved | Fill the offer template from ATS fields | Recruiter checks every figure before sending |
| Rejections | A recruiter moves a candidate to rejected | Queue a polite rejection email | The decision itself is always a person’s |
What is the difference between a rule, an AI step and an agent?
A rule does exactly what you told it: when the stage changes, send this email. It is cheap, fast and predictable.
An AI step sits inside a rule and does something that needs language: summarize this resume, pull the start date out of this email, draft a reply in our tone. It can be wrong, so its output should go to a person or a low risk place first.
An agent is given a goal and chooses its own steps, as covered in Lesson 1. Most recruiting automation does not need one. Start with rules, add AI steps where words are involved, and reach for an agent only when the path truly changes case by case.
Which tools connect recruiting apps?
Check your ATS first. Many have built in automations for stage changes, emails and reminders. Most also offer webhooks, which send a message to another app the moment something happens. Greenhouse, Lever and Ashby all document theirs.
When the ATS cannot do it alone, a no code automation tool connects the pieces:
| Tool | Good for | Cost to start (checked September 2026) |
|---|---|---|
| Zapier | The most app connections and the easiest setup. Also offers agents and an MCP connection for AI assistants | Free plan with 100 tasks a month and two step automations; paid from $19.99 a month |
| Make | Visual, branching workflows at a lower price per run | Free plan, then credit based paid plans |
| n8n | Technical teams that want full control or to host it themselves | Self hosted community edition, or cloud plans from €20 a month billed yearly |
How do you build your first automation?
1. Map the process on paper. Write every step of today’s process, who does it and how long it takes. You will often find a step that can simply be dropped.
2. Pick one trigger and one action. “When a candidate reaches the phone screen stage, send the booking link.” Small wins build trust.
3. Use the ATS feature if it exists. A built in automation is easier to maintain than an outside one.
4. Test with a fake candidate. Create a test record with your own email and run it through every path, including the unhappy ones.
5. Add AI carefully. Send AI output to a recruiter first. Use the prompt method from Module 1 and tell the AI to write “not stated” rather than guess.
6. Name an owner. Every automation needs one person who gets the error alerts and fixes it when an app changes.
7. Keep a list. A simple sheet of what runs, what triggers it and who owns it saves hours when something breaks.
Where must a person stay in the loop?
Anywhere the result seriously affects a candidate. In Europe this is law: GDPR Article 22 gives people the right not to be subject to decisions based solely on automated processing that significantly affect them, and the right to ask for a human to step in. Rejecting an application automatically can fall under that rule. New York City’s rules on automated hiring tools, covered in the resume screening lesson, add audits and notices.
The safe pattern everywhere: let automation prepare the rejection, the offer or the sensitive reply, and let a person press send.
What usually breaks?
| Problem | How to prevent it |
|---|---|
| Someone renames a stage or field and the automation stops | Turn on error alerts and write the dependency in your list |
| Duplicate candidates get two emails | Add a check for an existing record before sending |
| The automation fails silently | Send a weekly count of runs to the owner so a zero stands out |
| The AI step invents a detail | Only let AI fill fields from text it was given, and route output to a person |
| You hit a plan limit mid month | Check task usage weekly for the first month |
| Emails from automations land in spam | Set up email authentication, covered in the outreach lesson |
Quick answers
Do I need to code to automate recruiting?
No. Zapier and Make are built for people who do not code. n8n is friendlier to technical users but still visual.
Can an AI assistant build automations for me?
Increasingly, yes. Zapier’s MCP connection lets assistants such as Claude and ChatGPT take actions across thousands of apps, the approach explained in Module 1, Lesson 5. Review what it sets up before you switch it on.
What should I automate first?
Interview reminders or booking links. They save time every day, are easy to test, and rarely go wrong in a way that hurts a candidate.
Is it safe to send candidate data through automation tools?
Only through tools your company has approved, and only the fields each step needs. Module 4, Lesson 5 covers what candidate data can go into AI tools.