Overview

Lecture URL: https://skill.re/learn/recruiting/common-ai-errors-in-recruiting-hallucinations-misinterpretations-omissions.php

TRANSCRIPT: Common AI Errors in Recruiting: Hallucinations, Misinterpretations, Omissions

Course: AI for Recruiters - Professional Credential

Module: Level 2: Hands-On Foundations

Section: Chapter 9 -- Reviewing AI Output Critically

Theme: reviewing-ai-output-critically

Lecture: 9.1

Duration: 60 min

Format: Workshop + Hands-On

Audience: Recruiters beginning to use AI tools

Prerequisites: L1 Certification

What you will learn: You'll understand the most common errors AI makes in recruiting contexts:

hallucinated details, misread qualifications, missed information. By the end, you'll know how to

prevent these errors and design prompts that catch them.

INTRODUCTION

AI is helpful and powerful. AI is also fallible and sometimes confidently wrong. When AI operates

in recruiting, those errors have real consequences -- they affect hiring decisions.

The best recruiting professionals aren't the ones who trust AI blindly or reject AI entirely. They're

the ones who understand exactly where AI breaks down, design prompts and processes to prevent those

breakdowns, and catch errors before they influence hiring.

Today, we're studying the most common errors so you can prevent them.

THE THREE TYPES OF AI ERRORS

AI makes three types of errors in recruiting: hallucinations, misinterpretations, and omissions.

ERROR TYPE 1: HALLUCINATIONS (MAKING UP DETAILS)

Hallucinations are when AI generates information that wasn't provided and isn't true.

Examples:

• "This candidate has 10 years of experience" (notes say "several years")
  - "Led a team of 5 engineers" (notes don't mention leadership)
  - "Passionate about machine learning" (candidate never mentioned this)

Why it happens:

AI is trained to be helpful and complete. If you ask a question and the information isn't directly

stated, AI might infer or generalize based on patterns. "How many years of experience?" If notes

don't specify, AI might guess based on context clues.

How to catch it:

Check every specific fact against source material. If summary says "8 years" and notes say

"several years," that's hallucination.

How to prevent it:

• Instruct AI to mark claims as explicit (directly stated) or inferred (based on context)
  - "If this information is not explicitly stated in the notes, say so. Do not infer or guess."
  - "Only include information that's directly stated or clearly implied."

REAL EXAMPLE:

Prompt: "Summarize this interview for an engineer role."

AI output: "This engineer has 15 years of experience, with deep expertise in machine learning,

and led a team of 10 people."

Actual notes: "Has worked in backend engineering. Mentioned machine learning came up in recent

project. Has worked independently most of their career."

Prevention: "Only include information that's directly stated or you can see evidence for in the

notes. Mark any inferred information."

ERROR TYPE 2: MISINTERPRETATIONS (GETTING THE MEANING WRONG)

Misinterpretations are when AI understands the words but misses the meaning.

Examples:

• Interview note: "Candidate mentioned they might be interested in work-life balance." AI

interprets: "Not serious about the role."

• Interview note: "Candidate is quiet in group settings." AI interprets: "Poor communication skills."
  - Interview note: "Candidate asked to understand our compensation structure." AI interprets:

"Motivated primarily by money."

Why it happens:

AI assigns meaning based on patterns in its training data, not based on the full context. "Quiet"

correlates with various traits in its training data, so it might connect it to shyness or

disengagement.

How to catch it:

Compare AI's interpretations to your actual impression from the interview. Does the interpretation

match what you observed?

How to prevent it:

• "Do not infer motivation or personality from communication style."
  - "Quiet does not mean disengaged. Provide evidence for any claim about engagement or motivation."
  - "Mark any interpretation-based claims as 'inferred' and provide the evidence."

REAL EXAMPLE:

Prompt: "Summarize this interview."

AI output: "Candidate seems disengaged. Asked few questions."

Actual notes: "Candidate is introverted. Asked thoughtful questions when they did engage."

Prevention: "Distinguish between observation (quiet in meeting) and interpretation (disengaged).

Don't assume communication style means engagement level."

ERROR TYPE 3: OMISSIONS (MISSING KEY INFORMATION)

Omissions are when AI leaves out information that's actually important for the decision.

Examples:

• Technical interview is reviewed, but soft skills from the team round are omitted
  - Concern is flagged but the context that explains it is left out
  - Candidate's growth story (junior -> senior) is summarized as just a job title change
  - Specific evidence for a skill is dropped in favor of a general rating

Why it happens:

AI is given limited token budget or word limit. When forced to choose what to include, it might

leave out nuance or details that are actually important for decision-making.

How to catch it:

Compare the summary to your memory of the interview. What did the interviewer(s) emphasize that

didn't make it into the summary?

How to prevent it:

• Don't set tight word limits. Let the summary be as long as needed to include important

information.

• Specify what dimensions matter most: "Technical depth and team fit are most important. Include

evidence for both."

• "If something is missing that would help the hiring team decide, include it."

REAL EXAMPLE:

Prompt: "Summarize this interview in 200 words."

Output: Hits word limit, leaves out the team collaboration round entirely.

Prevention: "Summarize completely, prioritizing: (1) technical depth, (2) team fit, (3) concerns.

Length: as needed, not limited."

THE PREVENTION FRAMEWORK

To prevent errors, design your prompts with three safeguards:

1. EXPLICIT VS. INFERRED DISTINCTION

"Mark every claim as explicit (directly stated) or inferred (based on context). If inferred, provide

the reasoning."

2. EVIDENCE REQUIREMENT

"For any significant claim about skills, personality, or fit, cite the specific evidence from the

interview."

3. SCOPE CLARITY

"Summarize completely. Include all dimensions that matter for the decision. Do not omit information

to save words."

ANTI-PATTERNS

ANTI-PATTERN 1: IGNORING HALLUCINATIONS

Description: Accepting AI summaries without verifying specific facts.

Why it fails: Hallucinations compound. One false fact leads to bad decisions.

How to avoid: Spot-check every specific fact.

ANTI-PATTERN 2: ASSUMING GOOD INTERPRETATION

Description: Trusting AI's interpretation of soft skills or motivation without verifying.

Why it fails: AI's interpretations are often based on stereotypes, not evidence.

How to avoid: Mark interpretive claims and require evidence.

ANTI-PATTERN 3: ACCEPTING INCOMPLETE INFORMATION

Description: Using a summary that omits a whole dimension (soft skills, team fit) for a decision.

Why it fails: You're deciding without complete information.

How to avoid: Specify what dimensions must be included. Review for completeness.

PRACTICE PROMPTS

Exercise 1: Identify Hallucinations

Take an AI summary. For every specific fact (years of experience, team size, specific

achievements), verify against original source. Find any hallucinations.

Exercise 2: Catch Misinterpretations

Take an AI summary. Find any interpretive claims (about motivation, personality, engagement).

Verify that the interpretation matches the actual evidence.

Exercise 3: Check for Omissions

Review an AI summary. What important information from the interview is missing? Technical details?

Soft skills? Context?

Exercise 4: Improve Your Prompt

Take a prompt that produces errors (hallucinations, misinterpretations, omissions). Add

safeguards: explicit/inferred distinction, evidence requirement, scope clarity. Test the improved

prompt.

Exercise 5: Create Your Error Checklist

Based on errors you've found, create a checklist for reviewing AI outputs. What are the most

common errors you see? Build your checklist around those.

KEY TAKEAWAYS

1. AI makes three types of errors: hallucinations (making up details), misinterpretations

(getting meaning wrong), omissions (leaving out important information).

1. Hallucinations are common when information isn't directly stated. Prevent by requiring explicit

vs. inferred distinction.

1. Misinterpretations happen when AI assigns meaning based on patterns rather than context. Prevent

by requiring evidence and forbidding personality inference.

1. Omissions occur when AI prioritizes brevity over completeness. Prevent by specifying scope and

dimensions upfront.

1. Always verify specific facts in AI output against source material.
2. Mark interpretive claims and require evidence. "Disengaged" requires specific behavioral

evidence.

GLOSSARY

Hallucination: AI generating information that wasn't in the source and might not be true.

Misinterpretation: AI understanding words but assigning wrong meaning based on pattern-matching.

Omission: AI leaving out information that's important for decision-making.

Explicit vs. Inferred: Distinction between directly stated information and information inferred

from context.

SYNTHESIS AND APPLICATION

Understanding these three error types lets you design prompts and processes that prevent them. Once

you've prevented common errors, AI becomes genuinely trustworthy.

This week, identify one error type that most affects your hiring. Build a specific prevention

strategy for it. Test it.

REFLECTION EXERCISE

1. Which of these three error types have you experienced? When? How did you catch it?
2. If you reviewed your recent AI-assisted recruiting with this framework, what errors might you

find?

1. Which error type is most dangerous for your hiring (hallucinations, misinterpretations, or

omissions)? Why?

CLOSING REMARKS

Error awareness is the first step to error prevention. Once you know where AI fails, you can design

your processes to catch and correct those failures.

AI for Recruiters Certification Program

Level 2: Hands-On Foundations | Reviewing AI Output Critically | Lecture 9.1

A SkillsClinic initiative.

Duration: ~60 minutes | Word Count: ~2,480