Company Information: the problem with AI-generated business data

When it comes to company data, the structural limits of AI chatbots are particularly evident: outdated information, inaccurate data, and hallucinations are very common when relying on web data scraping.

Large Language Models typically do not have access to official data from Chambers of Commerce, the Revenue Agency, or other official sources, and they generate their responses based on probability.

Company data search: the limits of LLM chatbots

In just a few years, AI-based chatbots (ChatGPT, Google Gemini, Perplexity AI, Claude AI, Microsoft Copilot, etc.) have revolutionized the way users search for information online, forcefully entering the workflow of both small and large companies.

These tools, capable of quickly generating responses to even complex questions, are widely used in customer service and communication campaigns, but also in lead generation — and therefore in data collection.

It is precisely on the quality of data that the limits of LLMs become most evident: regardless of whether they draw information from proprietary systems or the web, AI-based chatbots essentially work through data scraping (extracting data from websites, documents, etc.) and text generation based on probability.

These characteristics expose them to several issues in terms of information accuracy: outdated training data, for example, will lead to non-updated and unreliable information, while the search for the “most probable answer” can result in the well-known LLM hallucinations — that is, plausible but completely fabricated answers.

Therefore, when it comes to searching for information that can affect business decisions and the quality of company databases — such as data on company turnover, ownership, or registered office — chatbots cannot be relied upon.

Business Intelligence vs. chatbots: what are the differences?

Some company-related data, such as VAT number or PEC address, may be easily available on the web and therefore within reach of any non-specialized chatbot. However, when looking for official and updated company information to enrich databases, feed statistics, and automate workflows, it is not advisable to rely on LLM-generated answers.

Chatbots, in fact, obtain their information from blogs, news articles, and other unofficial sources, and they do not have access to data provided by Chambers of Commerce, the Revenue Agency, or other accredited sources.

Conversely, business intelligence platforms and those specializing in company data supply acquire their information through official registers and databases, ensuring accurate and always up-to-date data. Access to these types of registers is also the foundation of data enrichment services, which allow cross-referencing data from various sources to profile users and provide increasingly specific and detailed reports.

Company data generated by AI: what are the risks?

Any LLM can effectively summarize market trends, analyze emerging sectors, and provide insights into competitors, but it can also give inaccurate or completely false information about a company’s ownership or turnover — potentially misleading sellers, investors, and other users.

Because of how they work, LLMs tend to “fill in the blanks” probabilistically: when a chatbot cannot find the requested information, it will attempt to provide the most plausible answer — completely inventing numbers, names, and even entire companies.

In addition to the risk of receiving false information, using chatbots to verify company data exposes a structural limitation: since they have no access to official registers or documents, these tools cannot provide data on corporate structure, actual ownership, or the possible presence of protests or alerts.

It is therefore clear that the risks are very high: by relying on such uncertain data, basic operations like record enrichment can easily result in systems corrupted by LLM errors and hallucinations.

How to use chatbots for company data verification

The fact that chatbots have limitations in data quality does not mean they cannot be effectively applied in company data verification. The key is to provide them with certified data from official sources and “force” them to work with those.

To avoid the dangers of web data scraping while still leveraging the enormous potential of integrating AI into business systems, it is necessary to design a mechanism in which different operations are handled by the right “agents.” It is clear that a chatbot alone cannot provide secure access to company data.

To obtain reliable and always updated company information, as mentioned, it is necessary to refer to official registers — a process that can be easily automated through the integration of APIs that make hundreds of certified and updated company data points available in real time.

This is where chatbots for company data verification come into play: for some time now, AI-based assistants have been able to use the open MCP (Model Context Protocol) to connect to countless external data sources and tools, including Business Information APIs.

This makes it possible to query APIs directly through chatbots — that is, to obtain certified, updated, and real-time company information simply by interacting with an AI assistant using natural language.