Why ChatGPT is not suitable for M&A longlists

Building high-quality M&A longlists is labour-intensive work and one of the core tasks in the transaction process – on both the buy-side (identifying attractive targets) and the sell-side (identifying potential buyers). Many advisors and investors are currently experimenting with large language models (LLMs) such as ChatGPT, Claude or Grok to automate these manual, time-consuming steps. But LLMs quickly reach their limits here – as we show below.

A sensible alternative is ProxDeal: a data terminal with a highly parallelised AI search system.

Missing data sources – registers and financial metrics are crucial

M&A analyses rely on verified data such as commercial register (Handelsregister) entries, balance sheets, shareholder structures and transaction information. ChatGPT and comparable LLMs, however, draw primarily on superficial, first-order web content (the “surface-level internet”). The result is lists that are heavily shaped by marketing copy, SEO scores and online visibility.

Typically, LLM applications such as ChatGPT have:

• No register data: Events such as capital increases, changes of managing director or insolvencies are missing.
• No financial metrics: Revenue, EBIT or headcount cannot be filtered reliably.
• No shareholder data: Ownership structure, owners’ ages, institutional holdings or succession indicators remain invisible.

The reason is that this data has to be painstakingly extracted – in the case of shareholder data, for example, from scanned shareholder lists. Yet it is precisely these signals that are crucial for identifying hidden gems in the market.

ProxDeal, by contrast, offers unrivalled depth of data – particularly in the German market – because it covers not only websites and press releases but also all relevant registers and specialist registers.

No hidden gems – SMEs with little search engine presence remain invisible

Many of the most interesting targets or buyers are Mittelstand companies (Germany’s typically owner-managed mid-sized businesses) without an SEO-optimised web presence. These “hidden gems” often stand out for high profitability, niche products or distinctive market positions – but not for technically optimised websites. Some relevant targets have no website at all.

LLMs, however, prioritise content that ranks well on Google or other search engines, and therefore return companies that are visible but often unsuitable or already well known. The consequence:

• Blind spots among relevant targets – and therefore missed deals.
• Lesser-known buyers on the sell-side are not captured – even though it is precisely these buyers who often turn out to be ideal transaction partners, especially in the small-cap segment.

ProxDeal, on the other hand, is a neutral research tool that works independently of search engines and brings valuable but otherwise invisible companies to light.

No up-to-date data and missing signals

In the M&A process, timeliness matters. While specialised tools such as ProxDeal regularly incorporate insolvencies, changes in share capital and commercial register events, LLMs remain static:

• No daily-updated insolvency data
• No succession signals (e.g. based on shareholders’ ages or shareholder structure)
• No tailored updates from proprietary sources

The result is lists that are not only incomplete but potentially misleading. ProxDeal, by comparison, updates its data regularly – insolvent companies, for example, are filtered out daily.

Lack of tailored enrichment and filters

A professional longlisting tool lets you select companies by quantitative criteria: revenue, headcount, net profit, EBIT. It also enriches records with individual contact details to prepare outreach at decision-maker level.

ChatGPT cannot:

• Set reliable filters
• Verify or enrich contact details
• Carry out tailored segmentation by KPIs

LLMs do not interpret filter criteria as quantitative, deterministic rules, but merely as part of the prompt – in other words, as qualitative suggestions. This leaves no basis for professional buyer or target analysis.

ProxDeal, in contrast, enables highly granular filtering by hard KPIs as well as enrichment with valid contact details for targeted outreach. Selection is unambiguous and quantified – with no fuzziness or room for interpretation.

Hallucinations and lack of reliability

A well-known problem with LLMs is hallucination: the model invents data points, industry classifications or even entire companies and people. In a highly sensitive process such as M&A, where facts and validity are decisive, this is a significant risk.

ProxDeal, however, relies exclusively on information published by the companies themselves – with numerous guardrails and error checks for maximum reliability.

Data protection and compliance risks

Working with client mandate data demands the utmost care. Uploading confidential documents to generic AI tools carries significant data protection risks:

• No data locality: It is often unclear where data is processed and stored – especially with US-based providers.
• Training on customer data: In many cases, it cannot be ruled out that data will be used for model training.
• Publication of your data: In the worst case, shared content can even become discoverable via Google.

In M&A, EU hosting, anonymisation strategies and strict data protection policies are therefore indispensable. ProxDeal ensures the highest security standards through EU hosting, clear deletion policies, strict anonymisation and full GDPR compliance.

Why specialised M&A tools such as ProxDeal are indispensable

ChatGPT and other LLMs are powerful tools for tasks such as research and writing – but not for M&A longlists.

A widespread risk is what is known as the halo effect combined with selection bias:

• Halo effect: If the model returns seemingly “good hits” in the first few queries – such as well-known competitors or large buyers – this creates the impression that the system also delivers reliable results across the board.
• Selection bias: This impression is reinforced because you unconsciously focus on these positive examples and overlook gaps, false information or missing candidates.

In reality, the problem is structural: after the first “obvious results”, nothing substantial follows. Without access to the commercial register, the Company Register (Unternehmensregister) or insolvency notices, the model’s universe ends where SEO-optimised company websites stop. In concrete terms, this means:

• Hidden gems on the buy-side – highly profitable but barely visible mid-market companies – are not identified.
• Lesser-known buyers on the sell-side – often decisive in SME transactions – simply do not show up.

Instead, after the first wave of “obvious results”, the model either produces no further hits at all or starts to hallucinate: it invents company names, revenue figures or acquisition histories that do not actually exist. For M&A longlists, this is fatal – because completeness and reliability are what make all the difference.

If you want to build reliable longlists, you need highly specialised research software such as ProxDeal. With ProxDeal, you can integrate register and financial data, identify company shareholders, spot succession signals, enrich contact details, find transaction data, and work securely and in full compliance. This is the only way to uncover hidden opportunities in the Mittelstand and reliably find strategic buyers, financial investors or targets.

Frequently asked questions