Widespread AI Web Search Usage Raises Corporate Risks
More than half of internet users now rely on artificial intelligence (AI) for web searches. However, a new in-depth study reveals that many popular AI search tools deliver data with inconsistent accuracy, creating potential hazards for businesses in areas such as regulatory compliance, legal matters, and financial planning.
Discrepancy Between Trust and Accuracy in Generative AI
Generative AI (GenAI) technologies have undoubtedly enhanced efficiency across many sectors. Yet, the research conducted by consumer watchdog Which? underscores a troubling gap: users often place high trust in AI outputs despite frequent inaccuracies in the information provided.
The survey involved 4,189 UK adults in September 2025, revealing that approximately one-third consider AI search more critical than traditional web search methods. This suggests that employees are likely using AI tools for business-related research, amplifying corporate exposure to risk if data is flawed.
Performance Variances Among Leading AI Tools
The study evaluated six major AI platforms—ChatGPT, Google Gemini (standard and AI Overviews), Microsoft Copilot, Meta AI, and Perplexity—across 40 queries related to finance, law, and consumer rights. Accuracy scores varied notably:
- Perplexity led with a 71% accuracy rate.
- Google Gemini AI Overviews followed closely at 70%.
- ChatGPT scored 64%, ranking second-lowest despite its dominant market presence.
- Meta AI scored lowest at 55%.
This divergence highlights the flawed assumption that popularity equates to reliability in GenAI outputs.
Critical Errors in Financial and Legal Advice
The investigation uncovered troubling inaccuracies that could have serious consequences. For example, ChatGPT and Microsoft Copilot failed to detect a deliberate error in a prompt about the UK’s £25,000 annual ISA allowance, offering guidance that risked breaching HM Revenue & Customs (HMRC) regulations. Conversely, Google Gemini, Meta, and Perplexity correctly flagged the mistake.
Legal queries also revealed systemic issues. AI tools often generalized regional laws, overlooking significant jurisdictional differences within the UK, such as between Scotland and England and Wales. Such oversights can expose businesses to compliance failures.
Moreover, AI platforms rarely recommended consulting qualified professionals for complex legal or financial issues. One notable example involved Google Gemini advising a user to withhold payment in a builder dispute—a recommendation that legal experts warned could jeopardize contract enforcement.
Transparency and Source Verification Challenges
Another major concern is the opacity surrounding AI-generated sources. The research found that AI often cited vague or outdated sources, including forum posts with questionable validity. In one instance, ChatGPT and Perplexity directed users to commercial tax-refund services instead of the official and free HMRC resources, potentially incurring unnecessary costs.
Such algorithmic biases could lead to inefficient spending or engagement with vendors lacking proper corporate due diligence, further increasing operational risks.
Industry Responses and the Role of Verification
Tech giants acknowledge these limitations. A Microsoft representative clarified that Copilot synthesizes content from various web sources rather than providing definitive answers and emphasized the importance of user verification. OpenAI highlighted ongoing efforts to enhance accuracy, noting that their latest GPT-5 model represents their most advanced step forward.
Recommendations for Mitigating AI Risks in Business
Experts advise against banning AI tools outright, citing that prohibition often drives usage underground. Instead, firms should develop comprehensive governance policies to manage AI’s integration responsibly:
- Specify Prompts Clearly: Employees must craft precise queries that include relevant context, such as jurisdictional details, to minimize misinterpretation.
- Verify Sources Rigorously: Users should demand transparent citations and cross-check information, especially for high-stakes topics, possibly using multiple AI platforms to confirm accuracy.
- Incorporate Human Expertise: AI outputs should be treated as supplemental opinions. Final decisions involving legal, financial, or medical matters require validation by qualified professionals.
As AI tools continue evolving, their accuracy for web search is improving incrementally. However, premature overreliance without proper oversight could lead to costly compliance failures and operational setbacks.
Related Reading: How Levi Strauss is leveraging AI for its direct-to-consumer business strategy
Related Articles
Study Finds Labeling Ads as AI-Generated Reduces Clicks by 31 Percent
China Approves DeepSeek’s Purchase of NVIDIA H200 AI Chips Under Specific Conditions
Snapchat Launches ‘Topic Chats’ to Boost Public Conversations on Trending Interests
Converge Bio Secures $25M Series A to Advance AI-Powered Drug Discovery
