Mitigating Business Data Accuracy Threats

More than half of internet users are now leveraging AI for web searches, yet the persistent challenge of data accuracy in these tools introduces new dimensions of business risk.

While generative AI (GenAI) promises substantial efficiency advantages, a recent investigation reveals a concerning gap between user confidence and technical precision. This discrepancy poses specific risks related to corporate compliance, legal defensibility, and meticulous financial forecasting.

For the C-suite, widespread adoption of GenAI parallels a classic ‘shadow IT’ predicament. A survey conducted in September 2025 with over 4,000 UK respondents indicated that roughly one third considered AI more crucial than traditional web browsing. With this level of personal integration, employees are highly likely using these platforms for business-related inquiries.

The investigation, spearheaded by consumer advocacy group Which?, underscores that uncritical dependence on these platforms could have considerable repercussions. Approximately half of AI users expressed ‘reasonable’ or ‘great’ trust in the information returned. However, a granular analysis of AI model responses shows this trust is often unfounded.

The Accuracy Gap in AI Web Searching

The study scrutinized six prominent AI tools – ChatGPT, Google Gemini (inclusive of standard and “AI Overviews”), Microsoft Copilot, Meta AI, and Perplexity – across a range of 40 common queries. The questions covered areas critical to business such as finance, law, and consumer rights.

Perplexity achieved the pinnacle score at 71 percent, closely trailed by Google Gemini AI Overviews at 70 percent. Contrarily, Meta’s offering achieved the lowest score at 55 percent. ChatGPT, despite its widespread usage, attained a middling 64 percent, making it the second lowest performer among the sampled tools. This highlights the dangers of equating market dominance with reliable information generation in the GenAI domain.

More concerning was the finding across all tools of regularly misunderstood information or incomplete advice, each creating distinct business operation risks. For legal and finance departments, the type of these errors raised particularly significant concerns.

In a financial test, questions were posed about a fictitious £25,000 annual ISA investment limit, when UK law stipulates different amounts. Both ChatGPT and Copilot took the amount as truth and provided advice based on the error, which potentially risked breaching regulatory rules.

Gemini, Meta, and Perplexity accurately spotted the amount issue demonstrating the variance across platforms. For businesses, this points to a need for precise “human-in-the-loop” monitoring of any AI-driven process to safeguard precision. This is particularly critical where automated systems integrate within existing business units.

For legal teams, AI struggles to understand regional legal differences presents a critical business risk. The research found these tools regularly failed to differentiate specific legal statutes across different UK regions, such as that of Scotland compared to England and Wales. Misinterpreting specific legal standards for regional application creates significant risk.

Beyond regulatory errors, a secondary ethical evaluation was run across all the models when facing high-stakes queries. Across legal and finance-based questions, the models rarely suggested consulting with a registered professional for legal guidance. When asked about a payment hold with a construction firm, Gemini advised pausing owed payments; which legal experts said could weaken the claim of the user.

This problematic “overconfident advice” creates immediate organizational exposure. Allowing AI to deliver preliminary compliance checks or automated contract review without a verified process potentially leads towards regulatory missteps – particularly where employees are untrained to spot the technical errors.

Source Transparency Issues

A pivotal concern for enterprise data governance is the lineage of information used by these tools. The test revealed opacity within several tools, with vague, non-existent, or poorly vetted references, commonly older forum threads. This lack of rigor creates potential inefficiencies and vendor-based exposure.

Concerning tax codes, AI was prone to algorithmic bias. In this, ChatGPT and Perplexity guided users toward costly tax-refund services instead of linking to freely accessible HMRC resources. This bias toward third-party services may unintentionally trigger overspending.

This means, from a business procurement standard, reliance on these same AI search tools can skew vendor engagement. Companies might unintentionally engage in services that fall short on standard due diligence, therefore creating unnecessary expense and further financial risks.

Major tech providers acknowledge these limitations, putting the onus back on end users and on businesses, by extension to verify results.

A Microsoft rep emphasized their tool works as a synthesiser rather than an authority: “Copilot streamlines information using web data, placing a higher emphasis on user verification of content and output precision.” The statement reinforced the responsibility for third-party output and accuracy.

OpenAI addressed the concern too, saying: “Improving accuracy is an ongoing industry-wide effort. With GPT-5, this model incorporates enhanced intelligence and overall precision.”

Mitigating AI Business Risk: Policy and Workflow

The suggested course of action for organizational leaders is not to ban AI, however to incorporate robust governance frameworks. Doing so will bring AI out of the shadows, while insuring the maximum accuracy through rigorous process.

• Enforce Specificity in Prompts: The investigation exposed limitations regarding AI’s prompt analysis. Corporate training should focus on precise queries to get stronger, less risky data. When researching, employees must specify details, such as “legal rules for England and Wales” rather than leaving the AI to infer.

• Mandate Source Verification: Single-source reliance creates risks. Employees must ask to see primary sources and manually conduct verification. High-stakes scenarios dictate “double sourcing” or verifying findings across multiple AI and non-AI-based resources. Gemini AI Overviews helps facilitate some transparency by allowing users to check the originating web links.

• Operationalize the “Second Opinion”: At its current stage, GenAI insights are one perspective. Finance, law, and medical concerns require human intervention, considering GenAI lacks capacity for nuance. Professional human advice must always be at the forefront.

As AI continues to improve, the key is to embrace change with caution. Businesses can maximize AI efficiency while preventing compliance failure, however only with a smart verification process.