AI meeting notes compared: when “I will share it” becomes “done”

Our pick for this meeting: Claude

Claude separated immediate actions from unresolved work and kept Taehun attached to the conditional customer announcement draft. It was the shortest answer, yet retained the details needed to decide who does what next. Concision was useful here because it did not erase the conditions.

ChatGPT gave more explanatory context. Gemini put the announcement draft in its task table, making it easy to find. Those are real strengths. Before sharing either output, however, we would restore the missing draft owner in ChatGPT and remove the unsupported completion label in Gemini. We judged the handoff, not the writing length.

The test: a deadline is not always a delivery promise

Our fictional Korean team met on Thursday, September 17, 2026, at 10 a.m. Minho promised an FAQ screen draft by 3 p.m. the next day. Taehun promised 12 frequent questions by 6 p.m. that day. Seoyeon promised a mobile feasibility and effort estimate by Monday morning, not finished implementation.

The same notes contained a tentative September 25 launch target, an unassigned review of seven old help articles, a dark-mode suggestion and an announcement draft that could start only after the launch date was settled. We fixed ten factual checks before collecting responses. Each service received the same Korean request to turn the notes into shareable minutes, with no imposed word count or response format.

ChatGPT kept the context but lost the draft owner

ChatGPT clearly distinguished the launch target from a confirmed date. It retained mobile in scope, excluded the search filter and kept the instruction to review old help articles before deciding whether to remove them. That context would help someone joining the project.

Its customer-announcement summary retained the start condition and the unassigned sender but omitted Taehun as the possible draft author. It also grouped several matters under decisions to make at the next meeting, although the original explicitly put only dark mode on that decision schedule. These might be sensible agenda suggestions; the minutes should distinguish them from commitments already made.

Gemini created a useful task table—and an unsupported “done”

Gemini retained Taehun’s announcement draft with the correct condition: start after the launch date is confirmed. Its development task remained a feasibility and effort check. Both details made the action table practical.

But Jisu’s row for sharing the notes was marked 완료, meaning complete. The input said she would share them after the meeting; it did not say she had done so. Copying that cell into a task tracker could close an outstanding task. Gemini also omitted the total of 48 help articles while retaining the seven under review, the deletion restriction and the unassigned reviewer. That loss of context is less consequential than inventing completion.

Claude preserved the conditional assignment

Claude placed the announcement under pending discussion. Its sentence kept the possible draft author, the condition for starting and the still-unassigned sender together. It did not mark Jisu’s sharing task complete.

It ended by recommending that missing owners be assigned before the next meeting. The wording made this a recommendation, not an existing agreement. Against our ten prewritten checks, Claude retained all ten in this response. This is a result for one synthetic memo, not a general ranking of the services.

The handoff WekeyLab AI would share

This is our edited handoff, translated from the same Korean test. Names and dates stay attached to the original fictional team. Times below are local to that Korean meeting; they are not rescheduled US deadlines.

Help-center update | September 17, 2026, 10 a.m. Scope: mobile included; search filter excluded. Launch: September 25 target only, to be reconsidered after development review. Jisu: share meeting notes after the meeting; completion unconfirmed. Taehun: send 12 frequent questions by September 17, 6 p.m. Minho: share FAQ screen draft by September 18, 3 p.m. Seoyeon: check mobile feasibility and effort by September 21 morning—not complete development. Conditional: Taehun can draft the customer announcement after the launch date is confirmed. Sending owner unassigned. Pending: review seven old articles out of 48 before deciding on removal. Reviewer and review deadline unassigned; do not delete now. Dark mode: proposal, inclusion to be decided next meeting. Next meeting: Tuesday, September 22, 11 a.m.

A prompt you can reuse in your own workflow

For the comparison we deliberately used a simple request. At work, add explicit rules for status and uncertainty. The prompt below is our improved version, not a second tested run. After generating minutes, check names, converted dates and completion labels against the source before sharing.

The distinction matters beyond meeting notes: “can draft” is not “will send,” and “will review” is not “will deliver.” Our related comparison of replies to weekend work requests examines the same boundary between a proposed task and an accepted commitment.

Turn these notes into meeting minutes and action items. Separate confirmed decisions, proposals and conditional work. Leave missing owners or deadlines unassigned. Do not mark future commitments complete. Put your own recommended next steps separately from what the meeting actually agreed.