In a survey of 400 business users of generative AI conducted by Eltes, approximately 80% had experienced hallucinations, highlighting the impact of inaccurate information both inside and outside the company. The survey also showed the extent to which fact-checking and other measures were implemented in response.
As the use of generative AI expands rapidly, Eltes surveyed 400 business professionals who frequently use generative AI in their work and found that 81.1% had experienced hallucinations. Many respondents cited conducting fact-checks on their own initiative as a reason for identifying them.
A total of 33.4% said that accepting hallucinated information as accurate had affected their work. The impact frequently extended to materials for external use and to internal decision-making and reporting. The most common response when hallucinations occurred, cited by 53.1%, was to point out the inaccuracies and issue another instruction.
A total of 55.5% said they took measures when using generative AI, such as instructing it in advance to provide supporting evidence or reference URLs. The largest group, at 38%, said they conduct fact-checks routinely. Meanwhile, 13.5% said they conducted little or no checking, and many in this group said they took no particular measures.
The survey concluded by emphasising the importance of measures against hallucinations and indicating that comprehensive controls combining the development of guidelines, log visibility and training are necessary.