• Key Insight: Learn how AI-driven HSA guidance enables personalized benefits education at scale.
• What’s at Stake: Low HSA literacy risks underutilized tax-advantaged savings and increased employer administrative costs.
• Supporting Data: 69% of employees unclear on HSAs; over 50% unaware HSA funds are investable.
• Source: Bullets generated by AI with editorial review

HSAs can be a complicated benefit for employees to navigate on their own and an intimidating subject to broach with benefit leaders, but AI can make education easy and accessible.

Sixty-nine percent of employees were unclear on HSA benefits and uses, according to data from the Plan Sponsor Council of America, despite half of employers offering them. As a result, a study from Fidelity found that over 50% of respondents weren’t even aware that HSA dollars could be invested. In an effort to bridge some of the disconnect, HSA administrator HealthEquity launched a new AI tool to help educate employees and support their leaders.

“It was clear that benefits administrators and HR teams, as well as their employees, had questions about HSA rules,” says Shuki Licht, head of innovation and AI technology at HealthEquity. “When employees don’t understand their HSAs, they underutilize them, missing out on triple tax advantages and long-term savings opportunities. And for employers, this translates to lower engagement with the benefits they’re investing in and higher administrative costs from repetitive employee questions.”

Read more: These AI tools are helping benefit leaders do their jobs faster and more efficiently 

HSAnswers is available right on HealthEquity’s site for workers in any industry for free. Employees can ask the AI questions in normal, conversational language just as they would ask their own benefits administrator. After a short series of follow-ups, the chatbot then provides instant, personalized responses based on their specific situation. 

For example, Licht says an employee could ask a question like, ‘I’m turning 65 and signing up for Medicare, can I still contribute to my HSA this year?’ and receive a response immediately, rather than waiting for HR to research and respond. The AI is also equipped to handle more complicated questions around topics such contribution limits, eligibility, qualified expenses and long-term investment strategies.

Unlike many AI chat tools on the market, HSAnswers doesn’t rely on public information — it’s built exclusively from HealthEquity’s own knowledge base that includes over 500 curated educational resources, ensuring that users only get reliable and contextually relevant answers whenever and wherever they need them. 

Read more: How this new AI tool is taking the stress out of open enrollment

“What’s important is that we’re meeting employees where they are,” Licht says. “When they have a question at 8 PM on a Sunday about whether a medical expense is qualified, they don’t have to wait until Monday to call benefits administration.” 

Benefit leaders need AI support too

While communicating with employees and addressing benefit concerns is a major part of a benefit leader’s job, it can easily consume the time they should be spending on other equally important aspects of their work. In fact, HR and benefit leaders are spending a significant portion of their workweek answering employee questions, according to data from workplace insights platform StatusHR, with some studies indicating they spend up to 10 to 15 hours per week on repetitive inquiries alone. 

Having employees engage with a reliable AI tool doesn’t just alleviate some of the administrative burden by eliminating the amount of employees seeking their help; they also have a use for the tech. 

Read more: These technology-driven healthcare benefits are improving access to care

“We’re even seeing usage from benefits administrators themselves who use it as a reference tool when their employees ask questions,” Licht says. “It ensures consistent, accurate information across the organization.” 

As new demands arise, HealthEquity is working on expanding capabilities so employees can not only get answers, but also complete other necessary tasks like replacing a lost card or uploading missing documents for reimbursements claims. 

“AI is going to be transformative for benefit education because it solves the core challenge of scale,” Licht says. “You have millions of employees with individual questions and situations, but limited human resources to provide personalized guidance. AI enables us to deliver that personalized education at scale.”

An embarrassing discovery has rocked Newfoundland and Labrador’s education establishment. The province’s flagship education reform document—which ironically advocates for “ethical” AI use in schools—appears to have been partially written by the very technology it seeks to regulate.

CBC News broke the story last Friday, revealing that “A Vision for the Future: Transforming and Modernizing Education” contains at least 15 fabricated academic citations that experts believe came from AI language models.

The 418-page roadmap took 18 months to complete and carries enormous weight for the province’s educational future. Released August 28, the document maps out a decade-long transformation plan for public schools and universities across Newfoundland and Labrador.

Co-chairs Anne Burke and Karen Goodnough, both Memorial University education professors, presented the report alongside Education Minister Bernard Davis. None anticipated the brewing scandal.

The Smoking Gun Citation

One fake reference particularly caught researchers’ attention. The report cites a 2008 National Film Board movie titled “Schoolyard Games”—a film that never existed, according to board officials.

The citation didn’t materialize from thin air. Aaron Tucker, a Memorial assistant professor specializing in Canadian AI history, traced its origins to a University of Victoria style guide. That academic document uses fictional examples to teach proper citation formatting.

The style guide explicitly warns readers on page one: “Many citations in this guide are fictitious.” These examples exist solely for educational demonstration, not as actual sources.

Yet someone—or something—lifted the fake citation wholesale and planted it in the education report as legitimate research.

Tucker spent considerable time hunting for multiple sources referenced in the document. His searches through Memorial University’s library, academic databases, and Google came up empty.

“The fabrication of sources at least begs the question: did this come from generative AI?” Tucker told CBC. “Whether that’s AI, I don’t know, but fabricating sources is a telltale sign of artificial intelligence.”

AI’s Citation Problem

Language models have struggled with source fabrication since their inception. These systems generate statistically plausible text based on training patterns, prioritizing believability over accuracy.

ChatGPT, Gemini, Claude, and similar models excel at producing convincing fiction. When their pattern recognition fails to align with reality, the result sounds authoritative while being completely wrong.

The fabricated citations slip past human reviewers because they appear professionally formatted and contextually appropriate. Academic and legal fields face particular vulnerability to this deception.

Even AI models equipped with web search capabilities can fabricate citations, select inappropriate sources, or misrepresent their content.

The Crushing Irony

The fake citations scandal becomes particularly awkward given the report’s own recommendations. Among its 110 policy suggestions, the document specifically urges the provincial government to “provide learners and educators with essential AI knowledge, including ethics, data privacy, and responsible technology use.”

Sarah Martin, a Memorial political science professor, invested days reviewing the document and uncovered multiple questionable citations.

“Around the references I cannot find, I can’t imagine another explanation,” she told CBC. “You’re like, ‘This has to be right, this can’t not be.’ This is a citation in a very important document for educational policy.”

Josh Lepawsky, former Memorial University Faculty Association president, resigned from the report’s advisory board in January, citing concerns about the process. His assessment proves prescient.

“Errors happen. Made-up citations are a totally different thing where you essentially demolish the trustworthiness of the material,” Lepawsky told CBC, describing the process as “deeply flawed.”

Damage Control Begins

Co-chair Karen Goodnough declined CBC’s interview request, writing via email: “We are investigating and checking references, so I cannot respond to this at the moment.”

The Department of Education and Early Childhood Development acknowledged the problem through spokesperson Lynn Robinson’s statement. The department recognizes “a small number of potential errors in citations” and promises updates to the online version “in the coming days to rectify any errors.”

The scandal raises fundamental questions about academic integrity in the AI age. If a major government report advocating for responsible AI use contains AI-generated fabrications, what does this mean for educational standards and public trust?

Written by Alius Noreika