What the category covers
AI contract review is software that performs the preparatory work of reviewing a contract. It reads the document, extracts the clauses that matter, compares them against a standard set in advance, and marks where the text departs from what the firm considers acceptable. Its output is not a position; it is organised material — clauses, flags and drafting suggestions, each tied to a specific point in the contract.
That definition is deliberately narrow. It excludes the program that merely highlights keywords, and it excludes the practice of pasting the file into a general-purpose chatbot. The first is too little to save meaningful time; the second is problematic at a law firm for an entirely different reason, which this article returns to further down.
The real pain is not a lack of expertise
An experienced lawyer spots a badly drafted limitation of liability in half a minute. The problem is not that they could not — it is that they have to do it across dozens of contracts a week, largely on the same points. Repetitive reading, filtering out risky clauses and aligning text to the firm’s own template is not a professional challenge; it takes time and attention away from what is.
At a small or mid-sized firm this bites hardest, because there is no dedicated preparation capacity behind it: everything rides on the partner and a few colleagues, and there is nobody to delegate the first read to. When partner time is the only bottleneck, growth is limited by turnaround, not by expertise. AI contract review is an answer to that bottleneck — and, honestly assessed, only to that.
What AI automates today, and what it does not
What the machine does reliably
First, clause extraction: identifying the parties, the obligations, the term, the liability, the termination provisions and the GDPR clauses, then laying them out in a structure you can scan. Then comparison: measuring those clauses against the firm’s standard positions and marking the deviations — and, just as importantly, flagging the protective clause that is missing, which slips past a human reader far more easily than a badly worded paragraph does. Finally, drafting: redlines and standard clauses in the firm’s voice, plus question and answer on the document itself.
What it does not take over
The legal assessment. Whether this risk is acceptable for this client in this commercial situation, where it is worth conceding in a negotiation, and what follows from all of it. Nor does it take over the client relationship: trust, the right question and an understanding of the situation are not automatable sub-tasks. The software prepares; the acting attorney reviews, decides and finalises — and the responsibility stays with them throughout.
This is not written out of caution. Anyone who starts from the assumption that the tool will tell them the answer will also use it badly: skimming the output rather than checking it. Anyone who starts from the assumption that they are receiving prepared material for an expert to work on gains real time from it.
Why clause-level citation matters
A flag on its own is worth little. If the system says the limitation of liability is disproportionate but does not show which paragraph it inferred that from, the lawyer has to read the contract again — spending exactly the time the tool was supposed to save. Citation is therefore not a convenience feature but the precondition for the whole workflow: every statement has to be traceable to a specific point in the document.
There is a second benefit. Citation is what makes it visible when the system is wrong. With unverifiable output the error goes unnoticed, and does damage precisely where it least should; with cited output it surfaces at the first check. What makes a tool trustworthy is not that it never errs, but how quickly its errors become visible.
What a playbook is, and why a generic checklist is worth less
A playbook is the firm’s standard positions written down. It is not a collection of legislation and not a generic risk list: it records which limitation of liability this firm accepts, which notice period it considers acceptable, which confidentiality wording it expects, and which clause simply has to appear every time.
A generic checklist flags the same things for everyone. So it will often raise something that is not a risk at this particular firm, while staying silent about something that is. A system tuned to a playbook measures the contract against the firm, not against a market average. That has a price: the playbook has to be assembled, which is a few working sessions of shared thinking. In return, this is the point at which a general tool becomes the firm’s tool.
Where the data question enters
So far this has been the professional side — yet at a law firm a different question usually decides the matter: where does the uploaded file go? Privileged client material cannot be handled like an internal memo, and that narrows the field of eligible tools before anyone looks at a feature list. We worked through that question in a separate piece.
How AI Szerződéselemző implements it
The system works in six steps. A Hungarian-language contract is uploaded as PDF or DOCX; clauses are extracted for the parties, the obligations, the term, the liability, the termination provisions and the GDPR clauses; risk is flagged against the firm’s own playbook, missing protective clauses included; redlines and standard clauses are offered in the firm’s voice; and cited question and answer runs on the document. The sixth step is deliberately not performed by the software: the acting attorney reviews and finalises.
All of it runs as private AI: on private EU hardware with no external AI API — or entirely on the firm’s own server. The system works on Hungarian-language contracts, tuned to a Hungarian legal corpus, and a rollout starts with one or two contract types, typically a non-disclosure agreement or a services agreement. That narrowing is not modesty but method: on a single type it is measurable what the system finds and what it misses.
This article is general information, not legal advice. AI Szerződéselemző is software; the legal decision and the responsibility remain with the acting attorney.