AI in due diligence: what works and what is marketing
The traceability test: if an answer does not cite the document and page, it cannot be used in diligence.
The useful question about AI in a data room is not whether it has AI. It is whether an answer can be traced to a source you can put in front of counsel.
What genuinely works
Cross-document synthesis. Asking one question across 4,000 documents and getting an answer that cites each source. This is the capability that changes the work, because it replaces the part of diligence that is reading everything to find the three things that matter.
Clause extraction at scale. Pulling every change of control, exclusivity or termination clause into one table, with links back to the source.
Duplicate and version detection. Finding that a contract appears three times in different folders in two versions.
Q&A drafting with citations. A first-draft answer, with the source, that a human then approves.
What is mostly marketing
Per-document summaries. Useful occasionally, but a summary of one file at a time does not reduce the reading; it adds a layer.
Risk scores without sources. A number with no citation cannot be relied on, defended, or disclosed.
Chat over a single document. A demo feature. The hard problem is the whole set.
The traceability test
Ask the provider one question: does every AI answer cite the document and the page it came from. If it does not, the answer cannot be used in diligence, because nobody can check it and no adviser will stake an opinion on it.
Then ask where inference runs. If document text is sent to an offshore model provider, you have a cross-border disclosure question on every query, whatever the storage region says.
Governance is the other half
An AI feature reading confidential deal documents needs risk assessment, human oversight of outputs, and monitoring in production. ISO/IEC 42001 is the external audit of exactly that. As of September 2026 Linkro is the only data room provider we are aware of that holds it. Based on our review of publicly listed certifications, last checked September 2026.
Linkro's AI is opt-in, runs on Australian infrastructure, and cites the document and page behind every answer. Features still in build are marked coming soon.
Frequently asked
- Does AI actually help with due diligence?
- Yes, for cross-document synthesis, clause extraction at scale, duplicate and version detection, and drafting Q&A answers with citations. Per-document summaries and unsourced risk scores add little.
- What is the traceability test?
- Ask whether every AI answer cites the document and page it came from. An answer without a citation cannot be checked, defended or relied on in diligence.
- Is it safe to run AI over deal documents?
- It depends on where inference runs and how the AI is governed. Ask whether document text is sent to an offshore model provider, and whether the provider holds ISO/IEC 42001 for AI management.
- Is Linkro's AI optional?
- Yes. AI features are opt-in and security is identical whether they are on or off. Inference runs on Australian infrastructure.
Questions on anything in this article? Email sales@linkro.com.au.
Ready to experience the future of dealmaking?
One AI-native platform. Enterprise security by default. Turn on intelligence when you're ready.