An AI meeting tool can produce a readable summary of a long board meeting within minutes. The output is clean and structured. It lists attendees, headline topics, and a handful of bullet-point actions. For informal catch-ups, this is useful. For a board, a trustee meeting, a pension committee, or a regulated sub-committee, it is something else. It is a draft. A draft, however polished, is not a record.
That distinction is the subject of this post. We build ācta, which uses AI to draft minutes, so we make no blanket anti-AI argument. The argument is narrower: AI is a good tool for drafting minutes. It cannot replace the governance step that turns a draft into a record. Treating the two as one is how organizations end up with files that look like minutes but carry no weight.
How Do AI Summaries Differ from Governance-Grade Minutes?
An AI summary is a draft that no one has confirmed. Governance-grade minutes are a record. A named person reviews the draft, corrects it, and approves it, and the approved version is locked so any later change is detectable. AI speeds up the drafting. The review, approval, and lock make the document a record.
A typical meeting AI produces one of three shapes.
The first is a transcript: a near-verbatim rendering of the audio, with speaker labels of varying reliability. Useful for search, almost useless as a governance document, because nobody reads a long transcript to find a decision.
The second is an extractive summary: the model pulls sentences it judges important and stitches them together. The result looks coherent but is heavily shaped by whatever the model decided to keep. Context is lost. Nuance is lost. A crucial qualification on a decision ("subject to the CFO confirming the cash position") can quietly drop out.
The third is a generative summary: the model writes new prose describing what happened. This reads best and carries the most risk. Errors can enter the document without any flag. Attendees get misattributed. A point raised by one director is recorded as group consensus. The output is fluent enough that nobody notices until a dispute forces a careful read months later.
All three outputs share one defining property: no human has yet confirmed them. That is not a detail. It is the entire governance gap.
A Summary Is Not a Record
A summary is a description of what a piece of software thinks happened. A record is a document that a named, accountable person has read, corrected, and confirmed as accurate. The two are different artifacts, even when they contain the same words.
Corporate law draws this line in many places. In the United Kingdom, section 248 of the Companies Act 2006 requires a company to keep minutes of directors' meetings for at least ten years. Section 249 treats minutes authenticated by the chair of that meeting or of the next directors' meeting as evidence of the proceedings. Other jurisdictions set their own rules. Check the rules that apply to your organization with its counsel. The point holds everywhere: a signed minute is a record. An unreviewed AI output in a shared drive is not.
In general, a draft that nobody has approved is a starting point. It is not a decision record. Whatever your organization is required to produce, produce it from the confirmed version, never from the provisional draft.
Where AI Genuinely Adds Value
Here the argument cuts the other way. The bottleneck in meeting minutes has never been the format. It has been getting the first draft written while memories are fresh, context is intact, and the secretary is not also trying to join the conversation they are documenting.
A common practice is to circulate draft minutes soon after the meeting. Any later and memory erodes, attendees lose the thread, and the document that emerges is a reconstruction rather than a record. A clerk taking longhand notes while also contributing to a discussion often struggles to meet that standard. The time rarely exists.
This is where AI earns its place. A transcription and drafting pipeline can deliver a structured first draft soon after the meeting ends. Attendees, agenda items, decisions, and candidate action items are laid out. The clerk's job shifts from transcribing to reviewing: checking accuracy, filling gaps, correcting attribution, and confirming that the decisions as recorded match what was agreed. That is a different task, and a faster one.
AI is at its best when it compresses the time between meeting and draft. It is at its worst when it is treated as the final step rather than the first step.
There is a second benefit. AI is consistent in a way people are not. It does not skip the attendance section because the meeting was short. It does not forget the date of the next meeting. For a committee that meets monthly and wants a uniform record across years, that consistency has value.
The Step That Cannot Be Automated Away
The review and confirmation step is not paperwork. It is the moment at which a document becomes a record. It has three components that each do something AI cannot do.
1. A named human takes responsibility
When the chair signs minutes, they attest that the record is accurate. If the record is later challenged, the chair answers for it. No AI system can carry that accountability, because no AI system has a legal identity, a professional reputation, or a duty to the organization.
2. Discrepancies get surfaced and resolved
The review step is where an attendee reads the draft and says, "That's not quite what was agreed. We said six months conditional on sign-off from legal, not six months outright." These corrections are the point. They are the signal that real attention was paid. A document that goes from AI straight to archive, without this friction, preserves whatever errors the model made and buries them under the appearance of formality.
3. The record is locked
Confirmed minutes should not be silently editable. Once approved, the document is fixed. If a correction is needed, it goes through an amendment process at the next meeting and is itself minuted. This discipline keeps the record defensible years later. An AI output held in a generic document store, where anyone with write access can change a date or an attendee without trace, fails this test.
Attempts to automate this step away almost always reduce to "trust the model". That is an odd position to take for any decision important enough to be minuted in the first place.
What Are Common Mistakes When Using AI for Minutes?
Most failures come from the process around the tool, not from the AI itself. These five mistakes recur.
- Filing the AI output as the minutes. The draft goes to the archive with no named approver.
- Reviewing against the draft instead of the discussion. The reviewer checks that the text reads well, not that it matches what was agreed.
- Skipping attribution checks. A point raised by one person is recorded as the view of the whole group.
- Dropping qualifications. A condition on a decision falls out of the text and nobody notices.
- Leaving the approved version editable. A later change leaves no trace, so the record cannot be defended.
A Worked Example of the Review Step
Suppose the board discusses a vendor contract. The motion carried is: "Approve the contract, subject to sign-off from legal." The AI draft says: "The board approved the vendor contract." The sentence reads well. It is also wrong, because the condition is gone.
The reviewer compares the draft with the recording and restores the condition. The chair approves the corrected text. From that point the approved version is the record, and the first draft is only a step along the way. Without the review, the file would state an unconditional approval that the board never gave.
Why Do Unconfirmed AI Summaries Fall Short as a Record?
The weakness of an unconfirmed AI summary becomes concrete when you imagine it being asked for later. These hypothetical cases show the pattern.
- A charity is asked to show how trustees considered a significant transaction. The file it holds is an AI summary that nobody approved. It is hard to treat that file as a reliable account.
- A company faces a claim that a decision was taken outside proper procedure. A signed minute would serve as evidence. An unsigned AI draft is only prose.
- A housing provider is audited on a repairs policy. The minutes are AI outputs with no confirmation step on file. The auditor has to ask how anyone knows the text is correct.
- A pension trustee board relies on an AI summary to document an investment decision. Years later, a beneficiary challenges it. The trustees cannot show who approved the text, or whether it changed after the meeting.
In none of these cases is the AI tool the problem. The problem is the missing step between draft and record. A process that skips that step lowers the quality of the record without anyone choosing to.
How ācta Uses AI
ācta is built around a specific view of how AI should sit inside a governance workflow. AI drafts. People confirm. The system locks.
AI handles the drafting. A recording goes in, a transcript is produced, and a structured draft of the minutes comes out: attendees, agenda items, decisions, action owners, due dates, and deferred items. Segments marked off the record are removed before any AI model sees the transcript.
People handle the governance. The draft is explicitly a draft. It goes to the chair or minute-taker for review and edits. Nothing is final until a named user with the right authority approves it.
Approval locks the minutes and seals them with a SHA-256 hash, so a later change is detectable. Approvals and changes to minutes go into a tamper-evident audit trail. That is what tamper-evident means in practice: the approved version stays recoverable, and any difference from it shows.
The rule we follow is simple. AI drafts. Humans confirm. The system locks. Each step is accountable, and no step tries to do another's job.
What Should a Governance-Grade Minutes Tool Include?
If your organization has governance obligations, a few practical questions cut through the marketing on almost any meeting tool.
- Is there an explicit confirmation step, with a named approver, that changes the document's status from draft to record?
- After confirmation, is the record locked against silent modification, with changes captured in an audit trail?
- Does the tool distinguish between attendees, apologies, and absentees, and does it let a human correct attribution errors before sign-off?
- Can you show, on demand, who approved a specific set of minutes and when?
- Can your organization control how long audio and transcripts are kept, and can you export records if you change tools?
- Does the vendor explain how the AI component is used, and can you review the full transcript behind the draft?
A tool that answers yes to all six is doing the job. A tool that answers yes to only the first two or three is a note-taker with ambitions. There is nothing wrong with a note-taker. Just do not ask it to stand in for a record.
The Honest Position
The honest position on AI and minutes is that both things can be true at once. AI has changed how fast a good draft can be produced. That is a real advance for any organization that takes its meeting record seriously. It has not changed the part of the process where a person accepts responsibility for the accuracy of the record.
If you run a board, a charity or nonprofit, a pension scheme, a company, or any organization where the minutes might one day be read by an auditor, a regulator, or a judge, the test is not whether your tool uses AI. The test is whether your process ends with a confirmed, locked, attributable record. If it does, AI saves time. If it does not, AI is a fast way to produce something that looks like a record without being one.
Ask one question of any minutes in your archive. Could you prove, today, that this document is what was approved, unchanged, by the person who approved it? If the answer is no, the tool that produced it is not the one you need.
ācta follows this sequence. AI drafts the minutes, a named person reviews and approves them, and approval seals the record with a SHA-256 hash.
Sources
- Companies Act 2006, section 248: Minutes of directors' meetings. legislation.gov.uk, October 6, 2026.
- Companies Act 2006, section 249: Minutes as evidence. legislation.gov.uk, October 6, 2026.