Home
Hire tech talent with AI

6 Myths About AI in Recruiting

Recruiters often believe in some myths that impact their usage of AI.

Let's debunk some common misconceptions.

Why Read This Guide?

AI has evolved rapidly, leaving little time for people to fully understand its fundamentals.

Yet, there is pressure to implement AI in recruiting, causing confusion and uncertainty.

This guide aims to demystify common myths around AI in recruiting, empowering recruiters to make informed decisions.

6 Myths covered in this guide

1

AI makes biased decisions

2

AI is keywords driven

3

Newer models are always better

4

Our private data will leak to the model providers

5

AI's output can't be trusted

6

AI products are all alike

Myth 1: AI makes biased decisions

Fact: LLMs are not trained on any specific recruiting data to select or reject candidates. They are trained on general language and knowledge and can be instructed to use logics as defined by you

1

You do have a lot of control

You have the freedom to define your own rules in English

2

Careful prompting can reduce biases

Give clear instructions to AI to avoid biases of any kind

3

Omit critical info to avoid biases further

While sending data to AI, redact personal info like gender, race, religion, country etc.


Myth 2: AI is keywords driven

Facts:

LLMs look at words to sense "meaning"

You can use different words and LLMs will still understand since they work based on your intent, not keywords

But you can insist on keywords too

If you do want specific hard skills and keywords, you can still ask AI to look for them. It's just not mandatory.

Myth 3: Newer models are always better

Fact: Most models have reached a point which is good enough for a lot of recruiting tasks

Size ≠ Effectiveness

Smaller models can excel in specific tasks

Task specific optimization

Well-defined goals trump model size

Engineering matters

Beyond the model and prompts, a lot depends on how the feature is designed and implemented

Myth 4: Our private data will leak to the model providers

Facts

1

LLMs only learn during their training

Providers like OpenAI allow opting out of your data being used to train their next versions

2

Model providers scrub any personal data you send automatically

However you should avoid sending personal information to the models for absolute control

3

Your data is remembered only during "a conversation"

A conversation stores your messages to give contextual results. Starting a new conversation starts everything from scratch

Myth 5: AI's output can't be trusted

It's true that LLMs are trained to be "creative" by default. But

1

Your prompts can avoid this to a big extent

Giving specific instructions and setting "temperature" to 0 will reduce errors

2

Use good prompting techniques

Prompting techniques like "Chain of Thoughts" avoid hallucination.

3

Never rely on AI blindly

Never auto reject candidates based on AI. Always do manual testing of AI's results randomly and regularly.

Myth 6: AI products are all alike

It's not just about "a model and a prompt". Different AI products have

Difference in goals

Each AI product may have specific design goals

Difference in AI implementation

The way data is stored, data is processed, prompts are written and AI components are combined, creates a big difference in the end product

Difference in Product Engineering

The way data is collected, stored, processed and how UX is designed changes the results

AI beliefs

Each AI vendor may have different thoughts on AI vs Humans, Safety vs Speed, Quality vs Quantity which can greatly impact your results.

This guide is shared with the "Gen AI Recruiters" community

Make your recruiting career "AI proof". Join the community.

Initiative by


Hire skilled tech talent. Without the unproductive work.

Made with