5 ChatGPT Mistakes Recruiters Make and How to Fix Them

There are right ways to use ChatGPT throughout your recruiting process—and there are wrong ways. This guide covers five possible mistakes you could be making with ChatGPT and provides actionable steps to address them. Some mistakes could prevent you from reaching ChatGPT’s full potential and some could even put your organization at risk.

Since ChatGPT’s launch, recruiters have hit the ground running to understand how the generative AI tool can benefit a full workload. From researching keywords for a Boolean search string to drafting candidate outreach emails, ChatGPT has become a convenient tool to simplify processes that swallow precious time throughout the week. But the tool is still in its infancy, and as recruiters continue to experiment, unintentional mistakes along the way can happen.

Some mistakes may prevent you from getting the full benefits of ChatGPT, while others could put your organization at risk. Ahead, we’ve rounded up five common ways we’ve seen recruiters misuse ChatGPT and offer simple fixes to each.

Mistake 1: You’re exposing sensitive information

Your chat history is a treasure trove of training material for ChatGPT. As a Language Learning Model (LLM), ChatGPT improves its outputs by processing text from three sources: publicly available information online, information licensed from third parties, and information that ChatGPT’s trainers or users (that’s you) input. ChatGPT stores your inputs for learning purposes, so if sensitive information makes its way in, you risk exposure to the public and to cybercriminals.

For example, pretend you ask ChatGPT to help you edit a coding assignment for a software engineer, and you accidentally include a string of sensitive code that the hiring manager provided you. ChatGPT theoretically could use that code to train its model and it might even be used in responses to other users.

And ChatGPT is not immune from data leaks or account hijacking. OpenAI, the company behind ChatGPT, has had a slew of breaches, including one where over 100,000 account credentials were compromised and sold on the dark web.

What you can do:

You can disable whether your chat history can be used to train ChatGPT by toggling the option in the “Data controls” section under Settings. Your conversations will still be saved for 30 days before they’re deleted from the system, so this isn’t a foolproof method of keeping sensitive information safe.

OpenAI is actively addressing how it protects user information, but ultimately the responsibility lies with you. Gain a clear understanding of what is considered sensitive information at your organization. Commonly, it can include intellectual property, personally identifiable information, or financial information.

Mistake 2: You’re not taking bias into account

Since ChatGPT is partially trained with publicly available information from the internet, it’s loaded with varying perspectives and opinions that reflect society’s biases.

“Studies have shown that existing hiring practices are often biased,” says Sam Shaddox, General Counsel and Chief Privacy Officer at SeekOut. “To offer an example, reviews have shown the propensity of many employers to unintentionally overlook names associated with historically underrepresented populations. LLMs trained on that data will often reflect or even magnify that underlying bias.”

A Bloomberg investigation illustrates how this can happen: Reporters fed ChatGPT equally qualified resumes with demographically distinct names and asked it to rank them. After 1,000 attempts, ChatGPT favored names from some demographics over others “to an extent that would fail benchmarks used to assess job discrimination against protected groups.”

What you can do:

OpenAI has said that ChatGPT is not free from bias and stereotypes and encourages users to carefully review its content. Sam agrees that recruiters should scrutinize outputs for bias and emphasizes the need for employers to offer the appropriate training for success.

“It’s important for employers to ensure they are both identifying and correcting any underlying bias,” he says. “Knowledge is power, and responsible deployment of AI will create more neutral hiring processes.”

Mistake 3: You forget to use ChatGPT

To get the most out of ChatGPT, you’ll need to use it regularly— but that’s easier said than done. Whether your day gets too busy, you’re comfortable in your existing routine, or you’re only familiar with one or two ways to use it, there are plenty of reasons you might forget to reach for the tool.

What you can do:

Identify your learning goals. What parts of the recruiting process could you use help with? Create intention and focus by understanding what you want to get out of ChatGPT, like finding hidden talent with adjacent skills or honing your craft with stronger writing.

Follow recruiting influencers who are using ChatGPT. When you have chatter in your feed about specific ways that ChatGPT is used in recruiting, you’ll gain inspiration (and a periodic reminder).

Schedule ChatGPT into your existing routine. For example, if you’re spending the next few weeks researching diverse organizations where you could source candidates, incorporate ChatGPT into that knowledge-gathering process.

Share tips and tricks with teammates. If your network is also trying to use ChatGPT regularly, keep each other accountable: talk about new ways you used ChatGPT in your weekly standups, ask colleagues how they would use the tool in a certain scenario, etc.

Mistake 4: You over-rely on outputs

It’s possible to get a little too comfortable with AI tools. With ChatGPT, it can be easy to blindly copy and paste an output. This behavior is called automation bias, and it’s the tendency to over-rely on automated tools to make decisions because we place too much trust into AI systems.

ChatGPT generates responses by analyzing your input and predicting the most appropriate words that follow. It does this based on the patterns it’s learned from the text that’s been used to train it. ChatGPT is, in part, trained on text from the internet, which we know is rife with inaccuracies.

What you can do:

Always review ChatGPT’s outputs. ChatGPT shouldn’t be the final say of your work. Instead, view it as a brainstorm partner that can help you get to the finish line a little faster.

Your role is to iterate, identify inconsistencies, correct mistakes, refine the language, and add personality where appropriate. And consider ChatGPT as a starting point in any research process to inspire your next steps with reliable online resources.

Mistake 5: You give up too early

If ChatGPT doesn’t give you the content you want after the first or second attempt, don’t give up. ChatGPT’s strengths don’t always shine after the first or second output. As a collaborative tool, ChatGPT works best when you go back and forth with it and use multiple directives to craft the content you want. Common mistakes with an initial prompt include asking for more than one task at a time, not providing enough context, or even loading a prompt with too much information.

What you can do:

Iterate, iterate, iterate. When you break up a prompt into multiple, clear directives that build off one another, you increase the chances of getting more accurate results.

Here’s an example that shows how you could iterate your prompts by splitting up tasks.

Prompt 1: Write a 200-word email to a potential candidate for a frontend software engineer role at my company, ACME. The first paragraph of this email should summarize the role and provide an overview of our company. The second paragraph should outline the benefits and potential growth opportunities. Be friendly and add a sense of urgency.

Follow-up prompt 2: Now rewrite this email in the form of a 300-character LinkedIn connection request.

Follow-up prompt 3: Now write five subject lines for the email. They should be effective in getting people to open the email.

Follow-up prompt 4: Rewrite these with a friendlier tone.

Download this guide to see if you’ve been guilty of any of these mistakes and learn how to undo them before they become bad habits.