Many internet search engines and browsers offer an Artificial Intelligence (AI) search function. The answers seem palatable, convenient and accurate, but are they? I tested the AI search function to see for myself.
Internet search engines scour the web looking for tags, themes, keywords, etc. that match the intent of the search query. The AI searches seem more nuanced depending on the algorithm’s maturity. This should provide a more accurate result. However, the Internet is huge, unregulated, and contains a high volume of both accurate and inaccurate information.
One of the things I have learnt about AI is to try to define boundaries around preferred data sources. For instance, you can start your query with “according to OSH-Wiki, only……”, or according only to Safe Work Australia publications….” By identifying a reputable source and limiting the search to it, one can provide a credible, verifiable response. But there is a caveat. Both the sources mentioned above offer publicly available data. There are many reputable data sources available on the Internet, such as research publications and some newspapers and magazines, that are restricted and therefore not accessible to public AI searches.
Most recent content on SafetyAtWorkBlog is restricted to subscribers, so public AI search tools are limited. Examples of this are below.
I asked one internet search engine the following question:
“Using SafetyAtWorkBlog articles as the ONLY data source, what is SafetyAtWorkBlog’s position on job stress?”
The response was
“SafetyAtWorkBlog emphasizes the significant impact of job stress on workplace safety and employee well-being. The blog highlights that over 75% of people experience burnout, which can lead to various workplace challenges, including psychosocial hazards and insecure work conditions. It advocates for a holistic approach to workplace safety, recognising the importance of addressing both physical and emotional aspects of work to prevent harm and promote a healthier
work environment.”
It is largely correct, but the AI does not have access to the full articles due to the subscription restrictions. It only accesses the Open Access articles or the first paragraphs, which are accessible on the website. The blog does highlight the 75% figure, but that likely originates from the first paragraph of this article from February 24, 2025. The full article and other blog articles argue that the burnout percentage reflects a fundamental misunderstanding of work-related mental health and that it uses “burnout” colloquially as shorthand for workplace stress. The AI summary grants the 75% styatement more credibility than it deserves.
The reference in the original AI answer points to the blog’s homepage rather than a specific article.
The AI answer does include a further information icon which says:
“This summary was generated by AI from multiple online sources. Find the source links used for this summary under “Based on sources”. Learn more about Bing search results how Bing delivers search results“
I couldn’t find a “Based on sources” button or tab on the search results page. I would like to think that my source restrictions within my question have been effective, but this (generic?) information statement mentions “multiple sources”. Hopefully, that means multiple SafetyAtWorkBlog articles, but who knows for sure?
The response is constrained by the blog’s subscription requirement, but if the data site were open source and could be contributed to by anyone, I suspect the accuracy and, perhaps, the fairness of the response would be even more questionable.
As an academic exercise, I posed the same question but listed OSH-Wiki as the only source:
“Using OSH-Wiki as the ONLY data source, what is OSH-Wiki’s position on job stress?”
The AI Search Assist response was:
“OSH-Wiki recognizes work-related stress as a significant challenge in occupational health and safety, emphasizing the need for stress prevention activities across all organizations. It defines stress as a negative psychological state influenced by the interaction between an individual and their work environment, highlighting the importance of addressing both job demands and resources to manage stress effectively.”
Its listed references were OSH-Wiki and Wikipedia, despite my attempt to include a restriction in the question.
I tried again, but with Safe Work Australia:
“Using Safe Work Australia as the ONLY data source, what is Safe Work Australia’s position on job stress?”
The response was from both SWA and a source called Anitechgroup.com (?!).
“Safe Work Australia’s position on job stress is reflected in its efforts to collect and analyze data on work-related injuries and fatalities. They have launched an interactive data website that provides insights into workplace safety, including job stress-related incidents. The data indicates that reducing work-related incidents could significantly contribute to a healthier and more productive workplace, potentially increasing the Australian economy by $28.6 billion annually and creating additional jobs. Safe Work Australia emphasizes the importance of using data to identify and communicate workplace stories, thereby safeguarding workers and preventing further harm.”
“Job stress-related incidents” are just one part of the collective estimate of economic costs.
The best I can say about the AI search response is that it offers the gist of SafetyAtWorkBlog’s position on job stress, but it remains inaccurate. The Search Assist function on reputable, publicly accessible sites provides the clearest position statement from OSH-Wiki. It seems I need to keep learning and experimenting with AI, and probably improve the way I frame my Internet search questions, as I suspect we all do.
Kevin Jones
