Why You Should Verify Information Provided by ChatGPT
OpenAI introduced ChatGPT Search in October, aiming to enhance user experience by enabling web browsing and access to up-to-date news from partner sites and sources that haven’t blocked its web crawler. However, a recent review by Columbia University’s Tow Center for Digital Journalism raises concerns about the tool’s accuracy and reliability.
Testing ChatGPT’s Accuracy
The Tow Center conducted an experiment to evaluate how well ChatGPT represents content from publishers. They selected 10 articles from 20 publishers, including OpenAI partners, publishers involved in lawsuits against OpenAI, and those who either allowed or blocked OpenAI’s web crawler.
The test involved extracting 200 quotes from these articles and checking whether search engines like Google or Bing could correctly identify the sources. When these quotes were presented to ChatGPT, the goal was to determine if the AI could accurately attribute the sources, as promised by OpenAI.
Findings
The results revealed mixed accuracy:
- Some responses were entirely correct, others entirely incorrect, while some were partially accurate.
- In most cases, ChatGPT confidently provided answers—even when they were wrong. Only seven responses hinted at uncertainty.
- Misattribution was a recurring issue. For example, ChatGPT inaccurately attributed a quote from the Orlando Sentinel to a Time article.
Misattributions not only harm traffic for publishers but can also damage their brand reputation and audience trust.
Additional Issues
The review also found other concerning examples:
- Plagiarism: ChatGPT cited content from a site that had plagiarized an article from The New York Times, despite the NYT blocking OpenAI’s web crawler.
- Improper Syndication: It referenced a syndicated version of an MIT Tech Review article instead of the original, even though MIT Tech Review permits crawling.
The Larger Question
This research highlights an important debate: Does partnering with AI companies like OpenAI truly benefit publishers, or does it risk undermining their control and business models? The methodology and data for this study are publicly available on GitHub for further review.
Takeaway for Users
When using AI tools like ChatGPT, always verify the information. Click on provided footnotes or perform a quick search using trusted engines like Google. Taking these extra steps can help avoid misinformation and ensure accuracy.
By staying vigilant, users can mitigate the risks of “AI hallucinations” and maintain trust in the information they consume.
