Supplier discovery in today’s age on the surface seems easy, however we don’t think LLM’s are providing procurement teams with the accurate knowledge that they need. Artificial intelligence has transformed how businesses interact with information, but not all AI models are designed for every use case. LLM’s are an impressive general-purpose tool capable of answering questions, generating content, and simulating human-like conversations. However, when it comes to supplier discovery, it falls short.
While they can provide broad insights, they lack the procurement-specific intelligence required for accurate and reliable supplier identification. Procurement teams need tools that not only surface relevant suppliers but also evaluate them using structured, verified data—something that LLM’s cannot guarantee. Recent AI benchmarking tests, such as SimpleQA, revealed that even OpenAI’s most advanced models have hallucination rates as high as 61.8%, underscoring the risk of relying on them for fact-based decision-making.
In this post, we’ll explore why relying on LLM’s for supplier discovery can be risky and why dedicated procurement intelligence platforms offer a better approach.
Design Principles
As mentioned, the LLM’s we all know and use are brilliant general-purpose AI tools that provide conversational responses based on a vast dataset. The use cases are broad, and there’s no denying the value that they have. However, there’s no point pretending that they specialise in procurement-specific insights.
Generating responses based on probabilistic predictions rather than verified datasets means LLM’s lack the precision and transparency that procurement professionals require. While they can provide general guidance or surface broad supplier-related information, they cannot replace dedicated supplier discovery tools that leverage real-time data, contextual ranking, and support with wider procurement functions such as due diligence.
For procurement teams seeking actionable insights rather than generalised responses, solutions designed explicitly for supplier intelligence offer a more reliable and effective alternative.
Data Reliability & Transparency
OpenAI recently tested their models and found GPT-4.5 has a hallucination rate of 37.1%, while GPT-4o has a staggering 61.8% hallucination rate. These numbers are from a test called SimpleQA designed to measure hallucinations using challenging questions with exact-fact-based answers. There’s no denying that these numbers are alarming. Furthermore, it can often be difficult to know whether an answer is true or a hallucination, meaning further research is required to validate any answer given.
Additionally, general-purpose LLM’s generate responses based on a mixture of sources and may not always provide traceable or verifiable data. In a procurement context, this creates at best a waste of time and at worst, serious risks — especially when evaluating suppliers where accuracy and credibility are non-negotiable.
You might get a list of suppliers with impressive-sounding capabilities, certifications, or locations, only to find out later that the information was outdated, incomplete, or entirely fabricated. Without transparency into source data and the ability to verify claims in real-time, relying on LLMs for supplier discovery becomes less of a smart shortcut and more of a gamble.
Geographic filtering for Supplier Discovery
One of the biggest shortcomings of general-purpose LLMs in supplier discovery is their inability to accurately filter results based on geographic constraints. For procurement teams, supplier selection often depends heavily on location; whether for compliance with trade agreements, minimising shipping costs or ensuring proximity to a production facility. But LLMs don’t have structured access to business registries, trade zones or map data so they can’t reliably narrow results to a specific geography.
They’re also prone to misinterpreting place names (e.g., confusing Georgia the country with Georgia the U.S. state) and often surface outdated or globally scattered results instead of local, relevant ones. Without real-time, structured access to location-aware business data, LLMs tend to generate a vague, unfiltered list of suppliers — many of which may not even operate in the desired region.
Automated Supplier Due Diligence
A critical step in supplier discovery is due diligence. It’s not just important to find a supplier but it’s also essential to vet them. This means scoring vendors based on key metrics like firmographics, quality certifications, sustainability practices and capabilities. General-purpose LLMs aren’t built for this. They don’t access third-party data sources and they can’t dynamically enrich supplier profiles. What you’re left with is generic, surface-level information instead of an informed, risk-adjusted supplier score.
Without real-time enrichment and metric-based ranking, LLMs can’t distinguish between a well-audited, proven supplier and a risky unknown — which makes them a shaky foundation for any serious procurement workflow.
While the rise of large language models has undoubtedly changed how we access and process information, not all AI is created equal — especially in complex, high-stakes domains like procurement. Supplier discovery isn’t just about generating names; it’s about finding trusted, verified partners who meet precise business, regulatory and logistical requirements.
General-purpose LLMs, for all their versatility, fall short in delivering the accuracy, depth, and contextual intelligence that procurement teams demand. Their tendency to hallucinate facts, lack of integration with real-time or structured datasets, and inability to handle critical tasks like geographic filtering or due diligence make them an unreliable tool for this use case.
Where LLM’s do excel is supporting in areas like:
✅ Suggesting sourcing strategies
✅ Providing supplier vetting checklists
✅ Recommending marketplaces based on your needs
Procurement professionals don’t need probabilistic guesses for something as critical as choosing the right supplier — they need clarity, traceability, and confidence in the data behind every supplier decision.
That’s why purpose-built procurement intelligence platforms, designed with domain-specific logic and powered by enriched, verifiable data sources, are essential. Rather than trying to retrofit general AI to meet procurement’s needs, it’s time to embrace solutions that are engineered from the ground up to support supplier discovery, evaluation, and decision-making at an enterprise level.
