AI Screening Recruiter Setup Guide [Starter Guide]

Screen 100+ Candidates in 36 Hours (Not 3 Weeks)

Time Investment: 1-2 hours setup | Monthly Cost: $81-127 | Time Saved: 50+ hours per hiring cycle

Overview

What This Does

This system automates the first round of candidate screening using an AI voice agent that:

• Sends candidates a link to call an AI screener
• Conducts 10-minute verbal phone screens
• Asks consistent, structured questions
• Records transcripts and audio
• Collects structured data for easy review

The Results

Traditional Approach:

• 61 hours of manual work
• 3-4 weeks of calendar time
• Decision fatigue by candidate #73
• Inconsistent evaluation criteria

AI-Powered Approach:

• 10 hours total (1-2 hrs setup + 6 hrs review + 2 hrs decisions)
• 1.5 days of calendar time
• Fresh judgment for every candidate
• Perfectly consistent screening

When to Use This

✅ Great for:

• High-volume hiring (50+ applications)
• Remote/distributed roles
• Initial screening for basic qualifications
• Roles with clear requirements
• When speed matters

❌ Not ideal for:

• Executive/C-suite hiring
• Highly sensitive roles
• Very small candidate pools (<10 people)

Prerequisites

Required Accounts & Tools

Total Monthly Cost: ~$81 base + overage for heavy usage

💡 Real Usage Example: For 100 screening calls, actual cost was $22 (Creator plan) + $46 overage = $68 for ElevenLabs. Total system cost: ~$127. Creator works well for occasional hiring, upgrade to Turbo ($44/mo) for continuous recruiting.

Skills Needed

• ⭐ Basic: Copy/paste, follow instructions
• No coding required

Time Required

• Initial Setup: 1-2 hours
• Per Candidate Review: 3-5 minutes
• Ongoing Maintenance: ~30 mins/week

Architecture & Flow

System Overview

[Job Applications in Notion]
   ↓
[You Review & Approve for Screening]
   ↓
[Click "Send Email" Button in Notion]
   ↓
[Zapier Sends Email with Screening Link]
   ↓
[Candidate Clicks Link & Calls AI Agent]
   ↓ (10-min phone screen)
[AI Records Conversation & Collects Data]
   ↓
[You Review in ElevenLabs Analysis Dashboard]
   ↓
[Make Hiring Decision]


Data Flow

1. Applications come in → Captured in Notion
2. You do sanity review → Filter obvious rejections
3. Approved candidates → Click button to send screening link
4. Zapier triggers → Sends personalized email with link
5. Candidate completes AI call → Data collected in ElevenLabs
6. You review in ElevenLabs → Listen to recordings, read transcripts
7. Make decision → Move forward or reject

Key Decision Points

The AI Handles:

• ✅ Conducting consistent screening calls
• ✅ Basic qualification checks
• ✅ Communication skills assessment
• ✅ Salary/availability confirmation
• ✅ Recording & transcription

You Handle:

• ✅ Initial application sanity check
• ✅ Cultural fit evaluation (from recordings)
• ✅ Final hire decisions
• ✅ Technical deep-dives (if needed)
• ✅ Next-round scheduling

Step 1: ElevenLabs AI Agent Setup

1.1 Create Your ElevenLabs Account

1. Go to elevenlabs.io
2. Sign up and select a plan

Choosing Your Plan

Based on real-world usage for 100 screening calls:

• Based on actual usage: 100 calls (~10 min each) used 153,063 surplus credits beyond Creator’s 201,000 limit.

Recommendation:

• Start with Creator ($22/mo) if hiring for 1-2 roles
• Upgrade to Turbo ($44/mo) if screening 100+ candidates monthly
• Overage is reasonable: ~$0.46 per call

💡 Pro Tip: You can upgrade mid-month if you hit limits. Start with Creator and monitor usage.

1.2 Create New Conversational AI Agent

Navigate to Agents Platform

1. In ElevenLabs dashboard, click “Creative Platform” dropdown (top left)
2. Switch to “Agents Platform”
3. You’ll see the Agents sidebar with “Configure” section

Create Your Agent

1. In left sidebar under “Configure”, click “Agents”
2. Click the “+” button
3. Select “Blank Agent” (top option)
   • Full control over behavior
   • Templates have predefined behaviors

Name Your Agent

1. Enter Agent Name: e.g., “Candidate Screening Agent”
2. ⚠️ CRITICAL: Do NOT enable “Chat only” toggle
   • Leave OFF (default)
   • Disables audio if enabled
3. Click “Create Agent”

‍

1.3 Configure Agent Settings

System Prompt (Most Important!)

Step 1: Use AI to Generate Initial Prompt

1. Click System prompt field
2. Describe your agent:

You are an interviewer screening candidates for [Role] at [Company].
Ask 3 focused questions about their experience.
Keep calls to 10 minutes.

Step 2: Add Your Methodology

At the END, add your guardrails:

# Guardrails
Avoid generic questions. Focus on behavioral questions requiring
specific examples. Do not exceed three main questions. Do not offer
feedback or engage in casual conversation beyond screening scope.

# Tools
You have access to the candidate's resume (in text format).


Step 3: Refine

Generated prompt includes:

• # Personality: Agent character
• # Environment: Context (phone screen for X role at Y company)
• # Tone: Professional, encouraging, efficient
• # Goal: Screening objectives
• # Guardrails: Rules to follow
• # Tools: Available resources

💡 Use external LLM (Claude/ChatGPT) to refine, then paste back.

Critical Elements Checklist:

• [ ] Agent role and purpose
• [ ] Company name and position
• [ ] Time limit (10 minutes)
• [ ] Specific questions to ask
• [ ] Red/green flags to note
• [ ] Communication style
• [ ] Guardrails (what NOT to do)

First Message

What the agent says when candidate connects:

Better for recruiting:

Hi! Thank you for applying to [Company]. I'm here to learn more about
your background for the [Position] role. This should take about 10
minutes. Ready to get started?


Settings:

• Interruptible toggle: Keep ON
• Natural conversation flow
• Candidate can interrupt

Voice Selection

Current recommendation: Eric - Smooth, Trustworthy

How to choose:

1. Click “Voices” section
2. Browse available voices
3. Test 2-3 options

Recommended for professional screening:

• Eric: Smooth, trustworthy
• Rachel: Professional female, clear
• Adam: Authoritative male
• Natasha: Warm, conversational

Selection criteria:

• Clear articulation
• Professional but warm
• Matches company culture

Language & LLM

Language: English (default)

LLM: Gemini 2.5 Flash ⭐ RECOMMENDED

Why Gemini 2.5 Flash:

• Does NOT get stuck in loops (critical for phone)
• Fast response times
• Handles interruptions well
• Best for structured prompts
• Most reliable for voice

Advanced Settings

Automatic Speech Recognition:

• Enable chat mode: OFF
• Use Scribe: OFF (not production-ready)
• Audio format: PCM 16000 Hz (default)
• Keywords: Blank (add only if needed)

Conversational Behavior:

• Eagerness: Normal (balanced turn-taking)
• Take turn after silence: 7 seconds
• End after silence: -1 (Disabled - no premature endings)
• Max conversation duration: 600 seconds (10 minutes)

1.4 Configure Data Collection (CRITICAL!)

This is how you’ll identify and track candidates in the Analysis dashboard.

1. Click on “Analysis” tab
2. Find “Data collection” section on the right
3. Click “+ Add data point”

Essential data points to add:

1. ID (Number)

• Type: Number
• Purpose: Match to your Notion database
• Prompt: “What is your unique application ID?”

2. Name (Text)

• Type: Text
• Purpose: Candidate identification
• Prompt: “What is your full name?”

3. Custom fields (optional based on your needs)

• Email
• Phone
• Current Company
• Years of Experience
• Availability Date

Why this matters:

• Easy candidate identification in Analysis tab
• Filter and search by ID or name
• Cross-reference with your Notion database
• No confusion about which call belongs to which candidate

💡 Pro Tip: Keep data points minimal (3-5). You’ll ask for these AT THE START of the call, so don’t overwhelm candidates.

Example conversation flow:

Agent: "Before we begin, I need to confirm a few details.
       What is your unique application ID?"
Candidate: "It's 47"
Agent: "Great, and your full name?"
Candidate: "Sarah Chen"
Agent: "Perfect, let's get started..."


1.5 Widget Settings

1. Go to “Widget” tab
2. ⚠️ CRITICAL: Ensure “Chat (text-only) mode” is OFF
3. Keep defaults:
   • Send text while on call: ON
   • Feedback collection: ON

Get Your Shareable Link:

• Find “Embed code” section
• Copy the agent-id value
• Format: https://elevenlabs.io/convai/conversation?agentId=YOUR_AGENT_ID

1.6 Security Settings

1. Go to “Security” tab
2. Overrides section: Ensure “Text only” is OFF
3. Authentication: OFF (for easy candidate access)
4. Allowlist: Leave empty initially
5. Guardrails: None (initially)

1.7 Test & Deploy

1. Click “Preview” to test internally
2. Have 3+ team members test full conversations
3. Click “Publish” → “Publish to a new branch” → Name it “test”
4. Go to “Branches” → Deploy test branch (100% traffic)
5. Test again with deployed version
6. When ready: Merge test into Main
7. Get production link: “Publish” → “Copy shareable link”

Final Checklist:

• [ ] Agent configured (prompt, voice, LLM)
• [ ] Data collection points added (ID, Name, etc.)
• [ ] Widget voice mode enabled
• [ ] Security text-only disabled
• [ ] Advanced settings optimized
• [ ] 3+ people tested successfully
• [ ] Production link copied

Step 2: Notion Database & Zapier Automation

2.1 Create Notion Database

1. Create new Notion page
2. Type /database → “Table - Inline”
3. Name: “Candidate Pipeline”

Essential Fields

Status options:

• 🟡 New Application
• 🔵 Cleared for Screening
• 🟣 Screening Email Sent
• 🟢 Completed (optional - for manual tracking)
• 🔴 Rejected

Add “Send Email” Button

1. Open any record
2. “…” menu → “Add to…” → “Button”
3. Configure:
   • When: Button is clicked
   • Do: Send webhook
   • URL: (Add in Step 2.3)
   • Content: Check ONLY page_id
4. Name: “Send Email”

2.2 Create Zapier Workflow

1. Go to zapier.com
2. Sign up for Team Plan ($50/mo)
3. Click “Create Zap”
4. Name: “AI Screening Email Sender”

2.3 Step 1: Webhook Trigger

1. Trigger: “Webhooks by Zapier” → “Catch Hook”
2. Copy webhook URL
3. Paste into Notion button URL field
4. Test: Click button → Verify page_id received

2.4 Step 2: Retrieve Notion Page

1. Action: “Notion” → “Retrieve a Page”
2. Connect Notion account
3. Page ID: Select “1. Data Properties Page Id Formula String”
4. Test: Should show full candidate data

‍

2.5 Step 3: Send Email (Gmail)

1. Action: “Gmail” → “Send Email”
2. Connect Gmail account

To: Map to “2. Properties Email Email”

Subject:

Next steps for your [Position] application at [Company]


Body:

Hi {{First Name}},

You've been selected from 500+ applications for our screening round.

Complete your 10-minute screening call here:
[YOUR ELEVENLABS LINK]

When prompted, provide your unique ID: {{ID}}

Reply to this email once complete.

Best,
[Your Name]
[Company]


Map fields:

• {{First Name}}: “2. Properties First Name Title Plain Text”
• {{ID}}: “2. Properties ID Unique Id Number”
• [YOUR ELEVENLABS LINK]: Your production link from Step 1

2.6 Step 4: Update Notion

1. Action: “Notion” → “Update Data Source Item”
2. Database: Candidate Pipeline
3. Page ID: “1. Data Properties Page Id Formula String”
4. Screening Email Sent: Set to True
5. Leave all other fields EMPTY

2.7 Turn On & Test

1. Click “Publish”
2. Test full flow:
   • Create test candidate in Notion
   • Add “Send Email” button
   • Click button
   • Verify email arrives
   • Click screening link
   • Complete test call
   • Verify Notion checkbox marked

Step 3: Daily Workflow & Review Process

3.1 Morning: Review New Applications

1. Open Notion “New Applications” view
2. Quick scan each application
3. For rejects: Change Status to “Rejected”
4. For good fits: Change Status to “Cleared for Screening”

Sanity check criteria:

• Basic qualifications met
• Resume shows relevant experience
• Not obviously overqualified/underqualified
• Location works (if relevant)

3.2 Midday: Send Screening Links

1. Open Notion view: Status = “Cleared for Screening”
2. For each approved candidate, click “Send Email” button
3. Zapier automatically:
   • Sends personalized email
   • Marks checkbox as sent
4. Candidate receives email within seconds

3.3 Afternoon/Evening: Candidates Complete Calls

• Candidates click link at their convenience
• AI conducts 10-minute screening
• Data collected in ElevenLabs
• No action needed from you

3.4 Review Completed Screenings

Where to review: ElevenLabs → Analysis tab

What you’ll see:

• List of all completed conversations
• Date, duration, status
• Expand each to see collected data (ID, Name, etc.)
• Access to full transcript
• Link to audio recording

Review process:

1. Quick filter: Look for recent calls (today/this week)
2. Identify candidate: Check ID or Name data point
3. Read AI summary: Look at conversation overview
4. Scan transcript: Read key questions and answers
5. Listen if borderline: Play recording for voice/communication assessment
6. Make decision: Strong fit, maybe, or reject

Time per candidate: 3-5 minutes average

• Clear yes/no: 2-3 min (just transcript)
• Borderline: 5-10 min (listen to recording)

3.5 Optional: Track Decisions in Notion

If you want to track decisions back in Notion (manual process):

1. Open candidate record
2. Add field: “Screening Result” (Select)
   • Strong Fit
   • Maybe
   • Rejected
3. Add field: “Next Step” (Select)
   • Send Assignment
   • Schedule Interview
   • Reject
4. Add field: “Notes” (Text)

Update after reviewing each candidate in ElevenLabs.

3.6 Optional: Set Up Notifications

Get alerted when calls complete:

Option A: Email notifications from ElevenLabs

1. Go to ElevenLabs Settings
2. Enable email notifications for completed conversations
3. Receive email with link to review

Option B: Slack notifications

1. Connect ElevenLabs to Slack (if available in integrations)
2. Get real-time alerts in a dedicated channel

Option C: Daily digest

• Check ElevenLabs Analysis once daily
• Review all new calls in batch
• More efficient for high volume

Troubleshooting

Webhook Not Triggering

• Verify webhook URL in Notion button
• Check page_id is selected
• Ensure Zap is ON

Email Not Sending

• Check Gmail connection
• Verify field mapping
• Check spam folder

Agent Sounds Robotic

• Try different voice
• Adjust stability to 85%

Agent Interrupts Too Much

• Set Eagerness to Low

Can’t Find Candidate in Analysis

• Check data collection points are configured
• Verify candidate provided ID during call
• Search by name or date

Cost Summary

ROI: 50+ hours saved × $100/hr = $5,000 value for ~$100 cost

Next Level: Want More Automation?

This hiring system is just the beginning. Apply this approach to:

• 📞 Customer discovery - AI user research interviews
• 🎯 Sales qualification - Automated lead screening
• 💬 Support triage - First-line inquiries
• 📊 Feedback collection - Scale insights gathering

Happy Building!

‍