More than half of us now use AI to conduct web searches, but the data accuracy of common tools remains low, creating new business risks.
Generative AI (GenAI) offers undeniable efficiency gains, but new research reveals that the disconnect between user trust and technical precision poses particular risks to companies’ compliance, legal standing, and financial planning.
For executives, implementing these tools represents a classic “shadow IT” challenge. According to a survey of 4,189 UK adults conducted in September 2025, around a third of users already believe that AI is more important than standard web search. If your employees rely on these tools for personal questions, they’re almost certainly using them for business research.
Research conducted by Which? suggests that untested reliance on these platforms can be costly. About half of AI users report having a “reasonable” or “very” degree of trust in the information they receive. But that trust is often misplaced when looking at the granularity of responses provided by AI models.
The accuracy gap when using AI to search the web
The study tested six key tools across 40 common questions spanning financial, legal, and consumer rights: ChatGPT, Google Gemini (both standard and “AI Overview”), Microsoft Copilot, Meta AI, and Perplexity.
Perplexity achieved the highest total score of 71%, closely followed by Google Gemini AI Overviews at 70%. In contrast, Meta had the lowest score at 55%. Despite its widespread adoption, ChatGPT had a total score of 64%, the second-lowest performing of the tools tested. This disconnect between market dominance and reliable production volumes highlights the dangers of equating popularity with performance in the GenAI space.
However, research has shown that all of these AI tools frequently misread information or provide incomplete advice that can pose serious business risks. The nature of these errors is of particular concern to finance and legal departments.
When asked how to invest their £25,000 a year ISA allowance, ChatGPT and Copilot were unable to identify a deliberate error in the prompt regarding statutory limits. Instead of revising the figures, they provided advice that could be in breach of HMRC rules.
While Gemini, Meta, and Perplexity have successfully identified errors, there is no consistency across platforms, and business processes involving AI require strict “human presence” protocols to ensure accuracy.
Using AI for web searches poses a clear business risk for legal teams as AI tends to generalize local regulations. Testing revealed that the tool often misunderstood that laws often differ between parts of the UK, such as Scotland and England and Wales.
Additionally, the study highlighted ethical gaps in the way these models handle high-stakes queries. Regarding legal and financial issues, the tool rarely advised users to consult registered experts. For example, when asked about disputes with builders, Gemini advised to withhold payments. Experts said this tactic could put users in breach of contract and weaken their legal position.
This “overconfident advice” creates operational risks. If employees rely on AI for proactive compliance checks or contract reviews without checking the jurisdiction and legal nuances, organizations can be exposed to regulatory risks.
Source transparency issues
A key concern for enterprise data governance is the lineage of information. Our research found that while AI search tools often have a strong commitment to transparency, they frequently cite sources that are obscure, non-existent, or of questionable accuracy, such as old forum threads. This opacity can lead to financial inefficiency.
In one tax code test, ChatGPT and Perplexity displayed links to premium tax refund companies rather than directing users to the free official HMRC tool. These third-party services often feature high fees.
In the context of business procurement, such algorithmic bias when using AI tools for web searches can lead to unnecessary vendor spending and engagement with service providers that pose a high risk due to not meeting corporate due diligence standards.
Major technology providers are aware of these limitations and are placing the burden of validation firmly on users, and ultimately on businesses.
A Microsoft spokesperson stressed that the company’s tools act as synthesizers, not as sources of authoritative information. “Copilot answers questions by distilling information from multiple web sources into a single answer,” the company said, adding that it “encourages people to verify the accuracy of content.”
In response to the findings, OpenAI said, “Improving accuracy is an industry-wide effort. We are making good progress, and our latest default model, GPT-5, is the smartest and most accurate we have built.”
Reduce AI business risk through policies and workflows
The way forward for business leaders is not to ban AI tools (which in many cases will push their usage further into the shadows and increase), but to implement robust governance frameworks to ensure the accuracy of the output when used for web searches.
- Force prompt specificity. The study notes that the AI is still learning to interpret the prompts. Corporate training should emphasize that ambiguous queries yield dangerous data. If your employees are researching regulations, they should specify the jurisdiction (e.g. “England and Wales legal regulations”) rather than assuming the tool will infer the context.
- Require source verification: Relying on a single output is operationally unsound. Employees should refer to the source and request manual verification. This study suggests that for high-risk topics, users should verify results across multiple AI tools or “double-source” information. Tools such as Google’s Gemini AI Overview, which allows users to directly review the web links presented to them, performed slightly better in scoring because they facilitate this validation process.
- Practice a “second opinion”: At this stage of technological maturity, GenAI’s work should be seen as just one opinion among many. When it comes to complex issues involving financial, legal, and medical data, AI lacks the ability to fully understand the nuances. Corporate policies should ensure that expert human advice remains the final arbiter of decisions that have real-world consequences.
AI tools are evolving and web searches are becoming more accurate over time, but once the research is complete, relying too much on them at this point may prove costly. For companies, the difference between improving operational efficiency through AI and risking non-compliance lies in the verification process.
See also: How Levi Strauss is leveraging AI for its DTC-first business model
Want to learn more about AI and big data from industry leaders? Check out the AI & Big Data Expo in Amsterdam, California, and London. This comprehensive event is part of TechEx and co-located with other major technology events including Cyber Security Expo. Click here for more information.
AI News is brought to you by TechForge Media. Learn about other upcoming enterprise technology events and webinars.
