John Davie, Founder & CEO of Buyers Edge Platform and CEO of CollectivIQ, leverages data and AI to drive smarter enterprise decision-making.getty​With what feels like increasing frequency, the leading AI companies are announcing new models that are supposed to be more powerful and reliable than their predecessors. So why is user trust declining? In short, many people don’t believe they can rely on AI outputs, despite how they’re marketed. After years of rampant hallucinations, inaccuracies and public concerns about the companies building them, people are hesitant to put their trust in the models.​There’s plenty of data that reflects this skepticism. Stanford's 2026 AI Index Report pinpointed a decline in transparency across the leading foundation model companies. Hallucination rates were also concerning, with some top models reaching as high as 94%, according to the study. At the same time, The New York Times reported that Google’s AI Overviews produced incorrect answers roughly one in 10 times. And one workplace study found that only 9% of employees trust AI systems to support complex decision-making. ​Accelerated AI adoption has exposed a foundational flaw in both how these systems were built and how we’re using them. Some businesses are treating their AI models like a single source of truth, even though they were never designed to function that way. Ask several different LLMs to perform the same task, and you’ll see that you’ll get different outputs. Each model was built on different code and trained on different data sets, which means that inconsistencies are inevitable. ​This becomes a serious issue when businesses move from experimenting with AI to using it in business-critical functions. It’s already happening, with companies leveraging AI models for financial analysis, legal review, operations optimization, enterprise decision-making and more, all areas where accuracy and nuance are critical.​That’s where the trust problem comes to life. We can all agree AI is useful, but most organizations still lack reliable ways to validate outputs without sacrificing the speed and efficiency that made the technology appealing in the first place. In the rush to implement AI across the workplace, many companies skipped over that step, but it’s not too late to correct course. We already know that high-stakes decisions require review, verification and multiple perspectives. AI outputs should be treated no differently.​I predict the next phase of enterprise AI adoption will depend less on finding one perfect model and more on building systems of consensus. Instead of treating a single LLM as the definitive authority, organizations should compare outputs across multiple models, identify where they agree or disagree and synthesize the strongest conclusions.​If multiple models independently converge on the same answer, trust is likely to increase. On the other hand, divergence signals the need for additional oversight, stronger source material or further review. In both instances, consensus creates confidence.​This approach also helps to reduce dependence on the biases, blind spots and shifting priorities of any one AI vendor. No individual model is entirely objective, comprehensive or consistently accurate, and users deserve the flexibility to leverage and learn from different LLMs without breaking the bank, especially as the industry rapidly advances and vendor superiority is still very unclear.​We’re already seeing that over-reliance on the latest and greatest AI model doesn’t create confidence or increase accuracy, and simply adopting bigger and better models won’t solve the trust problem. Instead, companies should implement systems and processes that prioritize transparency, validation and informed oversight, allowing users to benefit from AI without sacrificing confidence in the results.​​​Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?