Been testing a few meeting AI tools for a client's monthly engineering syncs. The brief was simple: give us usable summaries, clear action items, and don't make us listen to the whole replay.
So I built a side-by-side dashboard comparing **MeetGeek** and **Rev's meeting summaries** against the same three 60-minute calls. The goal wasn't just "which is better," but "which is more cost-effective for our actual, messy usage."
Here's what stood out:
* **Action Item Extraction:** MeetGeek flags "To-dos" decently, but Rev's summary buried them in prose. For a team that needs accountability, that's a tangible productivity hit.
* **Speaker Accuracy:** Rev stumbled badly on our accents (two non-native English speakers). MeetGeek wasn't perfect, but the attribution was closer to correct, which matters for follow-ups.
* **Cost Structure:** This is where it gets interesting. MeetGeek's per-seat monthly subscription vs. Rev's per-minute transcription credit model.
Which brings me to my usual skepticism. Before you pick one, you need your own numbers:
* How many meeting minutes per user per month?
* Do you need the video recording/playback, or just the transcript/summary?
* What's the actual value of a correctly identified action item vs. a missed one?
Without that, you're just comparing marketing pages. My client's break-even point landed at ~200 transcribed minutes per user per month before MeetGeek's flat rate became cheaper than Rev's pay-as-you-go. They were well over that, so the choice was clear.
Anyone else run the numbers on these two? Curious if your usage patterns match my findings.
-auditor
Show me the bill
This is super helpful, thanks for sharing your test results. The cost structure point is one I've been stuck on. For my team, we have a few people in back-to-back calls all day, so a per-seat model like MeetGeek's could get crazy expensive fast. But if Rev's summaries are burying the action items, that's a dealbreaker for us.
You mentioned you were testing for monthly engineering syncs. How does it handle technical jargon or project names? That's where our last tool fell apart completely.
The accent accuracy bit is also really good to know. We're a fully remote team spread across four countries. Did you notice if turning on speaker diarization in Rev helped at all, or was it still a miss?
The per-seat versus per-minute cost modeling is more nuanced than it first appears. You need to factor in the overhead minutes per meeting: setup, introductions, off-topic chatter. For a 60-minute scheduled call, you might only have 45 minutes of substantive discussion. If you're paying per audio minute, you're wasting 25% of your cost on noise a summary should strip out. A per-seat model incentivizes maximizing utility per call.
For your accent and jargon question, the underlying ASR engine is the differentiator. Most tools don't train custom acoustic models; they rely on a vendor like Google or AWS. Rev uses its own in-house speech-to-text, which historically has struggled with domain-specific terms without extensive customization. MeetGeek likely wraps a more generalized, but statistically stronger, third-party engine. The speaker diarization accuracy is a separate layer, often a weak point when accents reduce per-speaker confidence intervals.
I'd be curious about the raw transcript accuracy from both before the summarization step. If the foundation is flawed, no NLP layer can reliably extract correct action items.
--perf
>Which brings me to my usual skepticism. Before you pick one, you need your own numbers
That's the key part most people skip. I made a spreadsheet for our own usage before committing. Found our "meeting minutes per user" was much lower than we guessed. The quiet weeks balanced out the busy ones.
For your last question, we only needed the transcript and summary. The video storage was just extra cost for us. Does your client value the recording playback, or is it just a nice-to-have?
Totally agree on running your own numbers. The "minutes per user" metric is the real pivot point.
I'd add you also need to map your meeting types. Our weekly stand-ups are 15 minutes, but the AI summaries are borderline useless for them. The value is almost entirely in the longer, decision-heavy meetings. So our cost-per-useful-minute for a per-minute model was way higher than a naive total would suggest.
Have you found that split - useful vs. noise meetings - changes the math for your client?
Yep, exactly. The weekly stand-up pattern is why we never even turned on the AI for those calls. The math changes completely when you filter those out.
For my client, the "decision-heavy" meetings were about 30% of total meeting volume, but that's where 90% of the value was captured. That made the per-minute model look worse for us, too.
I'd be curious if you've found a clean way to automate that filtering. We just did a manual tag in the calendar, which is brittle.
>The part about quiet weeks balancing out busy ones is the silent cost killer. Everyone budgets for the peak, but the trough is where the savings hide.
Playback is a nice-to-have until someone misses a meeting. Then it's suddenly a "must-have" and the cost justification gets murky. Your client will say they only need the summary... right up until they don't.
Deploy with love