Only one in four (26%) users click on links provided by AI chatbots, like ChatGPT, as sources for their messages to verify the information, according to a study by the Association of Online Publishers, as reported by Press Gazette.
The overwhelming majority of users trust AI, despite being aware that it may hallucinate facts. An AI hallucination is when a computer model makes up false, misleading, or entirely fabricated information and presents it with total confidence.
In Google's AI-assisted search mode, where its AI Gemini provides answers to searched terms, the willingness to click on a source increases to 34% according to the study. When search results are offered without AI assistance, this increases to 41%.
The study, which surveyed people in the United Kingdom, shows that if users recognise the source as trustworthy, 90% will believe what the chatbot says. When the source is recognised as untrustworthy, this figure drops to 10%. A source that is neither trustworthy nor untrustworthy makes 25% of users believe what the chatbot says.
High or low trust in a source cited by a chatbot often leads to 'stop here' behaviour, where the user retrieves information from AI without further interaction. This behaviour is most seen with highly trusted (35%) and highly untrustworthy (32%) sources.
AI companies have increasingly signed partnership agreements with publishers to give their Large Language Models (LLMs) access to articles.
Despite most AI interfaces featuring warnings about the chatbot sharing potential mistakes and false information, 37% of respondents did not realise that AI can make errors. This awareness is highest among younger users and lowest among older users, with almost half of 45-54-year-olds not being aware of so-called AI hallucinations.
