Authors: Edward F Barry Rebecca Starr
This article is third in a series exploring artificial intelligence (AI) in human resources. Following the introductory article on the human aspect of HR, we turned our attention to risk. The second article examines the fundamental risks of generative AI and best practices. In this third article, we drill down to specific applications of AI in HR and put the best practices to the test with a ChatGPT experiment. We invite you to share your experiences with AI in HR for possible inclusion in future series installments.
Experimentation can save your job
We continue to hear from HR professionals concerned about generative AI's impact on job security. One healthcare organization chief human resources officer (CHRO) shared, "AI is awesome, but I'm thrilled to be retiring at the end of the year." Until then, she appreciates our AI101 approach, adding, "I need this simplified; I have so much else going on with no time to learn AI."
Our response to those who fear AI will replace them: Your fear is legitimate unless you learn how to use AI. An often-cited adage is that it's not AI that will take your job — it's the person who knows how to use AI.
Impact of experimentation on attitudes and perceptions of AI
Experimentation is key to learning to use generative AI effectively and minimize risk. A 2023 Gallagher survey of organizational communicators found that one in three are experimenting with AI. HR leaders represented 23% of survey respondents — the second largest segment following internal communications leaders at 49%.
According to the Gallagher survey, those using or experimenting with AI see its benefits and felt more positive overall about the value of AI — 55% compared to 45%.
Those who don't use AI or know whether their organization is using it expressed a significantly more negative outlook about generative AI's impact and potential. Among this group, 92% cited feelings of resignation, terror and denial — a whopping 26% more than the average.
AI optimists outpace pessimists
A 2024 MetLife survey* found that 29% of employees responding expressed more optimism about how AI will impact their jobs than pessimism (16%). Further, the cohort of AI optimists is growing faster, up 36% from 2023, than that of pessimists, up 22% from 2023. The survey confirms Gallagher's finding of the consistent correlation between a stated understanding of AI via experimentation and optimism around AI.
Our colleague, Drew Munn, future impact advisor in the Communication Consulting practice who focuses on the employee experience, recently wrote that even among those who agree AI is a communications game-changer, "only a few are reaping the rewards and recognizing its true value… the key to successful experimentation is knowledge." You're singing our song, Drew! Experimentation is the best way to get comfortable with using AI.
Putting ChatGPT to the test
In the first half of our risk discussion, we offered tips on creating prompts to return the best results. We also warned of faulty facts and bias. To inform your experimentation, we thought it would be fun to do our own experiment and use AI to ask about using AI in HR. We chose ChatGPT for this experiment, but the insights we share here apply to any generative AI tool.
Here's a look at ChatGPT's grasp of the query:
We asked: "What are good prompts for using ChatGPT in Human Resources?"
ChatGPT's preface to suggested prompts: "When using ChatGPT in Human Resources (HR) contexts, the prompts should focus on tasks such as recruitment, onboarding, training, employee engagement, policy clarification, and general HR inquiries."
Our assessment: We like the list of HR functions ChatGPT identified because they're generally safe areas to introduce generative AI. We're surprised benefits isn't on the list. Perhaps ChatGPT knows that a good chatbot can manage most benefit inquiries or —more likely — hasn't yet learned benefits is an imperative component of a total rewards offering to attract and retain employees. Likewise, it didn't identify payroll or tracking time worked, but that's ok.
Compliance requirements put both HR functions in a high-risk category, which at this stage of AI development is inappropriate for AI experimentation. Finally, ChatGPT excluded workforce planning, culture transformation and other highly strategic HR functions from its list. Perhaps the AI tool agrees such functions demand a human touch.
Prompts for specific HR functions
Next, we reviewed ChatGPT's examples of prompts for each referenced HR function against the list of best practices outlined in our last post and our "risk lens." The chart shows ChatGPT's HR prompts organized from strong to poor (and risky), with our thinking behind the ranking.
| HR Function | Strong Prompts | Why We Think So |
| Onboarding | What are some best practices for onboarding new employees remotely? |
Clear and specific with appropriate context. Improve by adding organization-specific details, such as the industry. |
| Training | What are some innovative methods for delivering employee training programs? | |
| Engagement | What are some effective strategies for improving employee morale and engagement? | |
| Engagement | How can we foster a culture of collaboration and teamwork within our organization? | |
| HR Function | So-So Prompts | Why We Think So |
| Recruitment | Can you provide tips for writing an effective job description for a [specific role]?" |
Lack of context and/or invites a yes or no answer. While these might generate meaningful responses, providing more context — for example, paste the referenced process in the prompt and begin prompts with a directive (provide tips for…) will deliver better results. |
| Onboarding | Can you suggest ways to make our onboarding process more engaging and effective? | |
| General HR Inquiry | Can you provide tips for conducting effective performance reviews? | |
| Recruitment | How can I optimize our recruitment process to reduce time-to-hire? |
Lack of context. Strong prompts, but without including the company's associated processes, the generic response may lack value for your organization. |
| Onboarding | How can we ensure that new hires feel welcomed and integrated into the team during onboarding?" | |
| Policy | What is our company's policy regarding [insert specific issue]? |
Strong only if you use a licensed version of a generative AI tool that overlays your organization's information. If using a free/public version, the information could put you at risk because it doesn't know your policies. |
| What are the guidelines for handling workplace conflicts in our organization? | ||
| General HR Inquiry | How can we address employee performance issues in a constructive manner? | Lacks specificity. No risk, but the response likely will be generic and brief. Consider prompting for "10 ways to …" |
| HR Function | Poor Prompts | Why We Think So |
| Training and Development | Can you recommend resources for developing leadership skills among our employees? | Too generic. Specify the type(s) of resources, such as training tools, thought leadership and employees (front-line managers, C-suite). |
| Policy | Can you clarify the procedure for requesting time off? |
Subjective concepts. "Clarify" invites risk because you're asking the machine to judge the procedure. Instead, ask, "How do you request time off," or "What is the procedure for…?" For "effectiveness," the tool needs to know the specifics of your training initiative and objectives. |
| Training & Development | How can we assess the effectiveness of our training initiatives? | |
| Recruitment | What are some strategies for attracting diverse candidates during the recruitment process?" | Inquiries related to diversity risk a high potential for bias. Counter risk by including "without introducing bias" into your prompt. |
| General HR Inquiry | What are the legal requirements for terminating an employee's contract? | Super risky! Legal requirements vary by state and municipality. Legal advice should come from a human expert. |
ChatGPT's grasp of risk
ChatGPT ended its response to our query with this reminder: "Remember to tailor the prompts to your organization's specific needs and context. Additionally, ensure that any sensitive or confidential information is handled appropriately and in compliance with privacy regulations."
Kudos to ChatGPT for this response, especially regarding sensitive or confidential information. As we know, enormous risk is associated with mishandling employee data. Once you share information with a generative AI tool, there is no taking it back. Never provide identifiable information about your employees. If your employee profile is relevant to your query, use only aggregated and de-identified data.
Apply these takeaways to your experimentation
It's your turn. Consider which HR function might benefit from offloading some of the strategy or communications-related work, or would benefit your overall HR operation. Experiment by creating a prompt for a generative AI tool to produce the information you would typically have to generate on your own. Follow best practice tips in our last post.
Now, review the response using our assessment of ChatGPT's prompts as a guide. Play around with the prompt to change the wording to add more specificity or context. Alter the tone of the inquiry to see how it changes the response. Approach the exercise as if conversing with a colleague, building on your initial prompt and giving feedback for incorrect information.
Tell us about your experience
Tell us how it went. What is your biggest takeaway? What advice would you offer to a colleague beginning to experiment with generative AI? We won't share your personal information without your permission, but we love sharing examples from HR professionals.
Recommended resources
- Artificial Intelligence Risk Management, Gallagher
- AI in HR: 2024 Guide to Opportunities and Applications in HR, Academy to Innovate HR
- Ways AI Is Changing HR Departments, Business Week Daily
- Biases in LLM Generative AI: A Guide to Responsible Implementation, Forbes
- Why It's Impossible to Build an Unbiased AI Language Model, MIT Technology Review
- Best practices for prompt engineering with the OpenAI API, OpenAI Help Center
