We are testing AI meeting assistants against the same synthetic project-sync meeting. This page is the public log of what we can verify so far: setup friction, upload/import blockers, and the quality checks a human should run before sending AI-written follow-up notes.
The first scored output is not from a dedicated meeting assistant. In a no-login ChatGPT run on June 27, the synthetic transcript produced a usable follow-up plan scoring 4.34/5. That makes the beginner takeaway practical: if you already have a consent-safe transcript, a paste-only AI prompt may be easier to verify than a meeting bot that needs OAuth, calendar access, uploads, or a paid export path.
Dedicated-tool evidence is still mostly setup-friction evidence: Notta reached a Free dashboard and upload modal with a neutral test persona, Otter login was recovered but import remained unclear, and tl;dv exposed only Google/Microsoft/SSO login methods in the scheduled test environment.
For a search-friendly summary of free-plan, Notion, and Microsoft Teams boundaries, see our free AI meeting-notes limits guide.
Our synthetic project-sync meeting includes defined decisions, owners, due dates, unresolved questions, and must-not-invent constraints. We score tools against an answer key, not against how polished a summary sounds.
Status: No-login prompt baseline produced a usable follow-up plan
Setup: ~3 min setup, ~9 min run, ~12 min human QA
Score: 4.34/5
A prompt-only no-login ChatGPT run on June 27 turned the synthetic transcript into a strong decisions/actions/open-items recap. It is not a dedicated meeting bot and has only copy/export reuse, but it beat the current dedicated-tool evidence because it produced inspectable output without OAuth, calendar access, payment, or real meeting data.
Status: Free dashboard reached; upload blocked in scheduled browser
Setup: ~6 min recon
Score: Not scored
Neutral email signup worked and the Free dashboard/upload modal was visible, but the local fixture upload could not be completed through the scheduled file picker. The created account needs password reset/storage before reuse.
Status: Account recovered; import path unclear
Setup: ~15 min recon
Score: Not scored
Login to the existing neutral test account was verified, but the scheduled environment did not expose a reliable way to import the synthetic transcript or audio fixture.
Status: OAuth/SSO-only signup observed
Setup: ~3 min recon
Score: Not scored
The visible signup options were Google, Microsoft, and SSO only. We did not use personal OAuth, create an account, connect a calendar, or upload fixture data.
Status: Public-doc availability checked
Setup: Pending
Score: Not scored
Free upload appears promising enough for a later recon, but official export/download limits look Pro-gated, so it is a partial candidate rather than a recommendation.
Status: Blocked for uploaded fixture
Setup: N/A
Score: Not scored
Official docs point to live meeting capture rather than prerecorded upload, which does not fit this synthetic uploaded-fixture test without a safe live-meeting setup.
Status: Blocked on Free for uploaded fixture
Setup: N/A
Score: Not scored
The Free path requires live assistant/calendar-style testing; uploaded audio/video needs paid credits or a paid plan, so no no-spend benchmark output exists yet.
Status: Blocked for uploaded fixture
Setup: N/A
Score: Not scored
The product is designed around a live desktop meeting workflow, not prerecorded upload/import, so it is not a fit for this first uploaded-fixture pass.
A safe follow-up plan separates decisions, action items, unresolved questions, and risks. Before sending it, compare every owner, due date, price, launch date, approval, and policy claim against the transcript or your own notes.
The failure mode to watch for is false certainty: an AI note taker may phrase “we should check this” as if the team agreed to do it, or drop an important caveat because it makes the summary less tidy.
No affiliate links are used on this page. No payment method, personal account, OAuth login, calendar connection, real meeting data, or paid trial was used for these runs. The ChatGPT score is for a paste-only prompt baseline, not a dedicated meeting assistant or transcription product.
Last updated: June 27, 2026. Next evidence goal: add a second safe paste-only baseline or capture a dedicated meeting-tool output through a no-payment, no-personal-OAuth path before recommending any meeting-notes product.