AI chatbots like ChatGPT, Claude, and Perplexity can break down complex financial concepts in seconds, but a new study shows they struggle to recognize when users are vulnerable and may even amplify financial harm. Researchers from the University of Canberra tested how these three popular AI models handled financial queries from people facing real-world challenges like cost-of-living crises, unequal household finances, and pressure to invest in risky assets. The findings reveal a troubling gap between what looks like sound advice and what actually serves vulnerable users.

What Did the Researchers Actually Test?

The study evaluated ChatGPT, Claude, and Perplexity by running five hypothetical financial scenarios through each model multiple times to test consistency. Researchers created realistic situations representing different life stages and economic vulnerabilities:

• Recent Graduate: A 22-year-old university graduate trying to save for a home deposit while facing a cost-of-living crisis.
• Pregnant Mother: A pregnant woman planning for maternity leave with a partner who does not share finances.
• Single Parent: A single parent of two children with modest income who was encouraged by a cousin to invest in cryptocurrency.

To eliminate bias from identifying information, researchers removed names, locations, race, and income details before testing. Each scenario ran five times in separate browsers with cleared caches to ensure outputs were not influenced by retained data.

How Did Each AI Model Perform?

The three models showed distinctly different approaches to financial guidance, each with significant limitations. ChatGPT delivered highly detailed and practical advice but failed to recognize vulnerability signals embedded in the prompts. Instead of tailoring recommendations to the user's specific challenges, it relied only on explicitly stated information and ignored whether the situation suggested a need for additional support.

Perplexity took the most conservative stance, frequently urging users to seek professional financial advice. However, this caution came at a cost: the model produced the least detailed responses, offering limited actionable guidance. Claude offered comprehensive recommendations but leaned heavily toward self-guided financial planning, which may not suit users who need personalized support.

The most alarming finding involved how the models handled vulnerable users. In the case of the recent graduate, AI recommendations focused on saving for a deposit but ignored how high living costs would make that goal harder, even though explicit details about financial strain were provided in the prompt. For the pregnant mother, both ChatGPT and Perplexity assumed the partner would help with household expenses after birth, despite the prompt explicitly stating the partners did not share finances. This reflected social stereotypes associating mothers with parenting and fathers with financial provision.

Why Are These Blind Spots So Dangerous?

The single parent scenario revealed perhaps the most troubling gap. Although all three models advised caution about cryptocurrency, ChatGPT went on to describe in detail how to invest in crypto and recommended specific cryptocurrencies for beginners. A human financial advisor would immediately flag the combination of modest income, dependent children, high-risk investment, and anecdotal advice from a relative. Instead, the AI models failed to synthesize these warning signs into appropriate guidance.

This happens because AI models are trained on massive datasets created by humans, and those datasets carry human biases. When models learn from historical records reflecting systemic discrimination, they can replicate social stereotypes and make biased assumptions in their outputs. Research also shows that when AI explains its recommendations, humans are far more likely to trust the advice blindly, ignoring whether it is actually correct.

How to Use AI for Financial Guidance Safely

• Cross-Check Information: If you are already financially literate and know how to verify data, AI can serve as a useful brainstorming and fact-finding tool, but treat it as a starting point, not a final answer.
• Recognize Your Vulnerability: Be explicit about your financial constraints, family dynamics, and risk tolerance when asking AI for advice, and recognize that the model may not pick up on subtle vulnerabilities you don't state directly.
• Seek Human Expertise: For high-stakes financial decisions involving dependents, major life changes, or complex situations, consult a qualified human financial advisor who can understand your emotional, family, and personal context.
• Question Assumptions: When AI gives advice, ask yourself whether it reflects your actual situation or relies on common assumptions that may not apply to you.

What Needs to Change?

The researchers emphasized that financial planning is not just about numbers; it involves advice grounded in human values, emotional anxieties, family dynamics, personal experiences, and risk tolerance. AI has democratized access to instant financial information, but until these models can truly comprehend the complex, vulnerable, and emotional realities of human life, critical thinking remains the most valuable financial skill.

"Financial planning isn't just about numbers; it's about advice based on human values, emotional anxieties, family dynamics, personal experiences and risk tolerance," the researchers noted.

Bomikazi Zeka, Associate Professor in Finance, University of Canberra, and Raechel Johns, Professor of Marketing, University of Canberra

For AI tools to become a safe avenue for financial guidance, policymakers and financial regulators must build clear frameworks around AI-generated advice. Strict transparency standards are also needed to regulate how these models handle consumer data. Financial information is deeply sensitive, and consumers must have absolute clarity on how their inputs are stored, whether the AI retains a record of their finances, and who has access to that data.

The bottom line: ChatGPT, Claude, and Perplexity can help you understand financial concepts and explore options, but they cannot yet replace the human judgment, contextual awareness, and ethical responsibility that a qualified financial advisor brings to the table.