← Back to AI Legal Lab
Insight
Generative AI for Legal WorkContract ReviewLegal Outsourcing

Drafting and Polishing Meeting Minutes: A Standard Use of Generative AI

Hello, I'm Noriaki Asato, Representative Attorney at LegalAgent.

Preparing meeting minutes from meeting notes takes surprisingly long in practice. Picture a meeting with a business partner to discuss contract terms at which the legal team is present. Various issues are discussed on the spot, such as the scope of liability for damages and the monetary cap, the acceptance inspection period, and whether subcontracting is permitted. Yet when you review your notes afterward, the scribbles and inconsistent wording often stand out, and it is frequently hard to grasp immediately "what was agreed, what remains undecided, and who will handle what by when."

Minutes of contract negotiations are an important record that will later be referred to for confirming terms and as the basis for internal approvals. Whether agreed matters and undecided matters are recorded accurately determines how much rework follows in practice. Note that the minutes discussed here refer to records of everyday meetings and contract negotiations, and their position differs from that of shareholders' meeting and board meeting minutes, whose required contents are strictly prescribed under the Companies Act.

Generative AI, which is good at producing text (here I have in mind services used for writing, such as ChatGPT, Claude, and Gemini), is well suited to organizing such scribbled notes into a structured draft. Because it lets you prepare a first draft more quickly than writing from scratch, it is one of the easiest uses to incorporate into daily work.

Reading Agreed Matters and Action Items from the Notes

Notes taken during a meeting are not always consistent in level of detail or wording. Make the points that were agreed and the points that were taken back for further consideration explicit, so that even people who did not attend can understand them.

If you give the notes to a generative AI and instruct it to organize them in the format of minutes, it becomes easier to create a skeleton that separates agreed matters from action items. For example, with notes from a contract negotiation, you can have it present the content organized into agreed matters, undecided pending matters, and future action items, such as "agreed in the direction of limiting the scope of damages to direct damages," "the other side will take the proposal of a 14-day acceptance inspection period back for internal consideration," and "both sides will confirm internally by the next meeting whether subcontracting is permitted." Note that limiting the scope of damages (such as limiting it to direct damages) and setting a monetary cap on damages are separate issues, and the 14-day acceptance inspection period proposal that the other side took back must also be recorded, accurately distinguished, as an unsettled pending matter. By dividing the work so that you skip writing from scratch and a person checks the facts and finalizes the draft the AI has prepared, you reduce the burden of composing the text and can spend your time on checking it against the original notes.

A Prompt for Requesting a Draft of the Minutes

Organizing minutes according to standardized items makes the content easier to grasp when reviewing it later. Especially in meetings involving legal, it is important to clearly separate "confirmed decisions" from "matters taken back or pending." In negotiations, confirmed matters and matters requiring internal confirmation tend to be mixed together, so if both are written in the same section, there is a risk of misreading terms not yet agreed as confirmed matters.

It is also important to have the person in charge and the deadline stated for future action items. The basic form of the instruction is as follows.

Please organize the following meeting notes into the form of minutes, using the headings below.

- Date, time, and place  - Attendees  - Agenda  - Decisions  - Pending matters and matters taken back (state who will confirm them and by when)  - Action items (state the person in charge and the deadline; write "To be confirmed" for anything unclear)

Do not fill in anything that is not written as fact, and mark anything unclear as "To be confirmed." Where the speaker can be identified, note the speaker alongside. Paste the meeting notes on the next line. [Paste the meeting notes here]

State expressly "do not fill in what is not written" to instruct the AI not to add things based on guesswork. Even with that instruction, however, errors can occur, so checking the output against the source is essential.

For contract negotiations covering multiple issues, you can also create a table by issue. Showing which items are undecided makes it easier to use for internal reporting and preparation for the next negotiation.

Please organize the following contract negotiation notes by contract term issue. Output them as a table with the columns: Issue / Our position / Counterparty's position / Current outcome (agreed, pending, or continued discussion) / Next action.
- Do not fill in anything not written in the notes by guesswork, and mark anything unclear as "Not stated in notes"
- Transcribe numbers such as amounts, numbers of days, and deadlines exactly as they appear in the notes

Paste the contract negotiation notes on the next line. [Paste the contract negotiation notes here (with company names and personal names redacted)]

Dealing with Filled-In Facts and Misread Numbers

What requires the most care is that, even when the instruction prohibits filling in, the AI may invent a history or agreement that was not written in order to produce a natural flow of text. If the notes merely say "damages: discussion to continue," but this is rewritten as "agreed to limit to direct damages," it will cause serious confusion in later contract negotiations. It is essential always to confirm that decisions, the persons responsible, and deadlines have a basis in the original notes, and to treat anything unclear as "To be confirmed" rather than filling it in by guesswork.

Next is the mix-up of numbers and proper nouns. Differences in figures such as "14 days" and "40 days," or "tax-inclusive" and "tax-exclusive," significantly change the meaning even if the surrounding context reads smoothly. Check amounts and deadlines one by one against the original notes or audio recordings that have been properly obtained and may be used. If the records conflict with one another or with attendees' understanding, do not turn the matter into a decision while it remains unconfirmed; confirm with the people involved. Also, so as not to round off expressions of intention such as "we will consider it positively" into definitive agreements, you need to record the nuance of what was said clearly separately from the decisions.

Establishing a Common Internal Format

To make the preparation of minutes with generative AI take root across the organization, it is effective to prepare a standardized format and instructions within the company. If the format differs from author to author, extra effort is required when searching past history or comparing terms.

In the common instructions, specify that decisions and pending matters be separated, and that the confirmed person in charge and due date be written. If you also share the procedure of marking unclear items as "To be confirmed" and checking the numbers against the source materials after output, the checks can be carried over even when the person in charge changes.

When entering meeting notes, it is a prerequisite not only to redact company names and personal names but also to confirm that you have obtained approval for the use in accordance with the company's information management rules. In addition, treat text output by the AI strictly as an unconfirmed draft, and share and store it as a finalized record only after a person has confirmed the facts. If you save materials before they have been checked, label them so that they can be recognized as unconfirmed drafts.

Related articles

Articles connected to this topic.

Insight / 2026.10.03 Internal Use and Copyright: What to Check When Sharing Articles, Preparing Training Materials, and Using AI Summaries Insight / 2026.10.02 How to Draft a Data Provision Agreement: Scope of Use, AI Training, and Treatment on Termination Insight / 2026.08.02 Searching Internal Documents with RAG: How Copyright Risk Changes with the Type of Material

Services connected to this topic

Legal outsourcing Ongoing legal team support for contract review and legal operations. Generative AI legal consulting Terms, privacy, copyright, AI governance, and internal AI use rules.
View AI Legal Lab articles