184 documents sit in the data room: supply agreements, framework agreements, amendments, some of them only as scans. The client wants to buy the supplier, and signing is three weeks away. Before the team goes into the negotiation, the partner needs four answers: Where does a change of control trigger a termination right? Where is liability unlimited? Which contracts end in the next twelve months? And where does foreign law apply?
Two associates can get it done, with a weekend and an Excel list. The real work only starts afterwards: assessing the hits, drafting questions for the seller, adjusting the purchase agreement.
This is where AI pays off in contract work: it takes over the finding and sorting, so more time is left for the assessing.
- Clause analysis is not about one contract but about many: Which of our 200 supply agreements have a change-of-control clause? Where is liability not capped? Which contracts renew automatically?
- This is exactly what AI is good at: it reads every contract in full, looks for the same clause types and enters the result in a table with a verbatim quote and its source.
- The standard comes from you: a playbook defines for each clause what is standard, what is still acceptable and what is not. The AI assigns, people decide.
- Whether a clause is valid, for example under the law on standard terms, is not something the AI assesses. A “not found” must be checked as well, because then the statutory rule often applies.
- Anyone choosing clause analysis software should check six points, above all: can every row of the results table be verified against the original with one click?
What clause analysis is and how it differs from contract review
In a classic contract review, one contract is on the table, and the question is: What is in it, what is missing, what is risky? What AI can do here and where its limits are, we describe in a separate article.
Clause analysis turns this around. It is about a few questions across many contracts: the same clause types, searched for in every document of a portfolio, entered in one consistent overview. Typical occasions:
- Due diligence: change of control, exclusivity, liability, non-compete clauses in the target company’s data room
- Contract management: terms, notice periods and automatic renewals in your own portfolio
- Changes in law or the market: Which contracts contain a price adjustment clause, a force majeure provision, a clause on data transfers to third countries?
- Standardisation: How far do the signed contracts deviate from your own template?
By hand, this is legwork that grows with every contract. For an AI, the 184th contract is read as quickly as the first.
The clauses people ask about most
| Clause type | Typical question | What matters |
|---|---|---|
| Change of control | Is there a termination or consent right in the event of a change of control? | Also search for indirect wording (“change in the shareholder structure”) and assignment bans |
| Liability | Is liability capped, and at what amount? | Exceptions (intent, cardinal obligations) and references to the other party’s standard terms |
| Term and termination | When does the contract end, does it renew automatically, with what notice? | Dates in amendments take precedence over the original contract |
| Exclusivity and competition | Does the contract tie you to a partner, a territory, a product? | Duration and scope of the commitment |
| Jurisdiction and governing law | Which law applies, which court or arbitral tribunal has jurisdiction? | Make deviations from the standard visible quickly |
| Price adjustment | May the price be adjusted, according to which index or procedure? | Unilateral adjustment rights |
| Confidentiality | How long does the obligation last, and what is excluded? | Survival after the end of the contract |
The list is a starting point. Which clauses matter depends on the mandate, and that is exactly what belongs in the playbook.
The playbook: the standard comes from you
An AI does not know which liability cap your client accepts or which notice period is usual in your industry. Without a standard, it can only find clauses, not classify them. That is why good clause analysis works with a playbook: a short written definition for each clause type.
| Clause | Standard | Still acceptable | Not acceptable |
|---|---|---|---|
| Liability | capped at the contract value | capped at twice the contract value | unlimited or not regulated |
| Change of control | no provision | duty to inform | termination right of the other party |
| Jurisdiction | client’s registered office | another place in Germany | abroad without an arbitration clause |
With such a playbook, the AI assigns every clause it finds to a column and flags the deviations. That is a suggestion, not an assessment: whether a deviation is acceptable in the individual case is for the lawyer to decide. The gain is that she no longer reads 184 contracts, but the 23 that stand out.
How a clause analysis works in five steps
- Define the questions.Which clause types, which detailed questions, which playbook? The more precise the question, the more useful the table. “Is liability capped, and if so, at what amount?” is better than “Check liability”.
- Complete the portfolio.Amendments, annexes and referenced standard terms belong in it. Scanned documents must be legible. An analysis is only as complete as the data room.
- Extract clauses, with sources.One entry per contract and clause type: verbatim quote, document, section. Where nothing is found, it says “not found”, not a guess.
- Match against the playbook.The AI assigns and flags deviations. Unclear cases are marked as such instead of being forced into a category.
- Review what stands out and approve.A person checks the flagged rows and every “not found” row against the original, assesses them and approves the overview. Only then does it go to the client or into the due diligence report.
What the result must contain
The result of a clause analysis is a table: one row per contract, one column or block per clause type. To hold up in a report or a negotiation, every row needs four things:
- The verbatim quote, not a summary
- The source: document, section, for scans the page
- The classification against the playbook, recognisable as a suggestion
- The review status: who checked the row, and when?
Where the limits are
Assessing stays with people
Whether a limitation of liability in standard terms withstands judicial review, under §§ 305 et seq. of the German Civil Code for example, is a legal assessment in the individual case. The AI finds the clause and shows it. A person has to assess it.
“Not found” does not mean “no risk”
If a liability cap is missing, statutory liability applies, and it can go further than any agreed one. Empty cells therefore deserve as much attention as flagged ones.
What is not in the portfolio, nobody finds
Many clauses sit in annexes, framework agreements or the other party’s standard terms. If these documents are missing, no AI can find them either. A good report states which referenced documents are missing.
A change of control is not always called that
A keyword search misses paraphrases. Language models recognise them much better, but not without gaps. A spot check therefore belongs in every larger analysis.
Choosing clause analysis software: six questions
- Source:Does every row show the verbatim quote with document and section, and can it be verified against the original with one click?
- Check before output:Is it verified before display that the quote really appears in the document as shown?
- Your own playbook:Can you store your review criteria, or does the tool assess against someone else’s standard?
- Whole portfolio:Does the tool process hundreds of documents including scans and amendments, or only individual uploads?
- Results in your working environment:Can the overview be reused as an Excel or Word file?
- Confidentiality:Where is the data stored, where is the AI request processed, is the data used for training? For German law firms, § 203 of the German Criminal Code applies as well, more on this in our article on § 203 StGB and AI in law firm practice.
For an overview of providers, see our comparison AI tools for law firms 2026.
How aiuno handles clause analysis
At aiuno, contract review is handled by the Legal Crew, a skill with a fixed methodology. It works through contracts in a structured way, also contract by contract across an entire portfolio, and delivers a source-based result.
- A source for every statement: For every row, aiuno names the document, the section and the verbatim quote. Before an answer appears, aiuno checks that the document exists and that the quote really is there.
- Your standard: We set up your review criteria together with you, based on your requirements. The result only shows the match against them: complies, deviates, missing or to be reviewed.
- Results to keep working with: On request, aiuno creates the overview as an Excel or Word file, also in your house template.
- Confidentiality: Hosting in Germany, no training on your data, an architecture compatible with § 203 StGB and § 43a BRAO.
aiuno does not assess whether a clause is valid. Final responsibility stays with the lawyer. More on its use in law firms on our page AI for law firms.
Show us a portfolio, and we will show you the table.
In the demo we work with sample contracts. You ask the questions, we show the overview, and you check every source yourself.
Conclusion
And the partner’s four questions? In the end there is a table with 184 rows, each with quote and source. 23 rows are flagged, plus a handful of “not found” that the team rereads in the original. The weekend is no longer spent searching. The time goes into what only lawyers can do: assess, ask, negotiate.
- German Civil Code (BGB), §§ 305 to 310 (standard business terms)
- German Criminal Code (StGB), § 203 (violation of private secrets)
- Federal Lawyers’ Act (BRAO), § 43a (basic duties)
This article gives an overview and does not replace legal advice. As of 20 November 2026.
Frequently asked questions
What is the difference between clause analysis and contract review?
Contract review goes into depth on one contract. Clause analysis asks the same questions of many contracts and delivers one consistent overview, for example for due diligence or contract management.
Can AI check whether a clause is valid?
No, not reliably. Whether a clause withstands a review of standard terms, under §§ 305 et seq. of the German Civil Code for example, is a legal assessment in the individual case. The AI finds and assigns, the assessment stays with people.
How many contracts can an AI clause analysis process?
That depends on the tool. What matters is less the number than completeness: amendments, annexes and referenced standard terms must be in the portfolio, and scans must be legible.
Do I need a playbook?
Not for finding alone. For deciding which clause stands out, yes. Even one page with three columns per clause type (standard, still acceptable, not acceptable) makes the results much sharper.
What happens if the AI does not find a clause?
Then the overview says “not found”. These rows must be checked against the original, because a missing provision often means the statute applies, and that can be a risk of its own.