The Hidden Side of Generative AI in Web Research: When ChatGPT Invents Law Through Hallucinations

Did you know that ChatGPT provides WRONG information in 60% of cases during web searches? 😱

For us, corporate lawyers, this isn’t just a number. It’s a direct threat to our professional credibility.

This is what the Columbia Journalism Review study reveals, which identifies 3 major risks that can be translated into our daily practice:

1️. Advice based on NON-EXISTENT precedents → ChatGPT confidently cites rulings or case law that never existed.

2️. Truncated citations of legislative texts → The study shows that these tools systematically fail to correctly cite their sources.

3️. Lack of updates on legislative information.

Example: A lawyer using ChatGPT for research on the right to be forgotten was recommended a procedure based on a poorly cited European directive, omitting updates to the GDPR. Result: incorrect procedure and major risk for the company.

This applies even to premium models which, although more expensive, do not guarantee better accuracy. On the contrary, they can provide incorrect responses with deceptive confidence, making it difficult to distinguish between accurate and false information. 💰↔️❌

Let’s clarify that this exclusively concerns the limitations of ChatGPT’s web search function. This tool, however, retains many strengths for other uses such as writing or summarizing. But for legal research specifically, vigilance is required.

My advices :

• Always verify the information provided by ChatGPT with reliable and recognized sources. 🔄✅
• Use ChatGPT as a complementary tool rather than as a single source of information. 🤝🔍
• Stay informed about the limitations and updates of AI tools for safer and more effective use. 📚🧠

In a world where speed is valued, let us not forget that our added value lies in the reliability of our legal advice. Technology must remain an auxiliary to our expertise, never a substitute for our professional vigilance where responsibility remains ours.

Source : https://www.cjr.org/tow_center/we-compared-eight-ai-search-engines-theyre-all-bad-at-citing-news.php