Can AI assistants help teams working in Romanian capture better meeting notes? We tested three to find out.
“Meetings are long. Memory is short.”
We have all been there. You wrap up a 90‑minute meeting, after a long day of context‑switching, and someone suddenly asks: “So, who’s taking care of the meeting notes?”
Like many of us, juggling technical tasks and team check-ins, I often found myself in meetings trying to stay present, while quietly worrying about who’d remember all the important points. And when the meeting was in Romanian, but the notes had to be shared in English later… well, let’s just say my notes didn’t always survive the translation.
That recurring frustration sparked an internal research project at ISD: Can AI‑powered meeting assistants make this easier?
Why meeting minutes matter
Meeting minutes aren’t just paperwork; they’re an anchor for decision‑making and accountability. They help:
- Team members who were present confirmed what they agreed to.
- Colleagues who missed the meeting, maybe they had a conflicting call or a personal emergency, to catch up quickly without scheduling another recap.
- Project managers track progress and follow up on owners and deadlines.
When meeting minutes are delayed or unclear, important tasks often get overlooked, duplicated work creeps in, or deadlines slip. Multiply that by several meetings a week and you’ll have serious inefficiency.
Our experiment
I volunteered to lead a hands‑on test of three AI meeting assistants that explicitly claim Romanian speech recognition:
Meeting type: Up to 1h meeting held every 2 weeks (intensively using technical subjects and vocabulary).
Number of speakers: The group ranged from 4 to 6 people.
Scoring grid: After every meeting, every colleague had to come up with feedback on generated meeting minutes text. A score from 1 to 5 was used to appreciate the results obtained.
Time budget. The whole test series took around 2.5 months.
We ran all three in Microsoft Teams on their free plans, inviting each bot to:
- Round 1: A typical stand‑up conducted mainly in Romanian.
- Round 2: An equivalent meeting in English.
- Round 3: Like Round 2, except that the meeting is entirely held online
Participants evaluated how accurately the generated summaries reflected the discussion.
- Language understanding & summary accuracy
| Language | Ranking | Observations |
| Romanian | 1. MeetGeek
2. Fellow 3. Fireflies |
MeetGeek captured context best; Fireflies drifted off‑topic. |
| English | 1. MeetGeek
2. Fellow 3. Fireflies |
All tools improved, but MeetGeek stayed ahead. |
- Helping the bots help us
Once we realized the bots could listen for cues, we began interacting with them directly inside the meeting:
- We would specifically say “AND NOW, THE ACTION POINTS” and mention the person’s name responsible for this item.
- Before moving on, someone would say: “NEXT ITEM ON AGENDA.” to help improving the summary structure.
Once we started speaking more clearly to the bots, things like, “OK, action items now” or “next agenda point”, something interesting happened. The notes got sharper. Less fluff, more structure. Turns out, these tools are like quiet teammates: they do their best work when we give them just a little help.
- Speaker tracking
- In hybrid meetings (room mics + remote attendees), none of the tools hit 100 % accuracy.
- In a fully online format, MeetGeek nailed speaker attribution more accurately, compared to other Fireflies apps, for example.
- User experience (UX) & UI
| UX factor | MeetGeek | Fellow | Fireflies |
| Dashboard clarity | 4/5 | 4/5 | 2/5 |
| Learning curve | Low | Low | Moderate |
Fireflies felt “feature‑rich but menu‑heavy” while the others guided us straight to the recap.
- What do you get for free, and how much does an upgrade cost?
One of the first questions we asked ourselves was simple: How far can we get without paying? Since most teams aren’t ready to commit budget until they’ve seen clear value, we decided to run all tests using only the free plans.
Here’s what each app offered out of the box:
| Feature | MeetGeek | Fellow | Fireflies |
| Transcription minutes/month | 3 h | 5 meetings | Unlimited transcriptions |
| AI summary + action points | Yes | Yes | Yes |
| Storage | 3 months | 5 meetings | 800 min |
We were impressed that all three tools offered essential functionality, like speech-to-text, summarization, and basic exports, even at the free tier. But there were subtle trade-offs.
- MeetGeek offers structured, well-organized summaries and integrations, but capped transcription at 3 hours limit.
- Fellow provided a clean interface that’s easy to onboard with, but had storage limited to five meetings on the free plan.
- Fireflies stood out for unlimited transcriptions but felt less intuitive and sometimes struggled with accuracy, especially in Romanian.
Now, when we looked beyond the free tiers, here’s how the pricing stacked up (per user/month):
| Tier | MeetGeek | Fellow | Fireflies |
| Entry/Pro | $15 | $7 | $10 |
| Business | $29 | $15 | $19 |
| Enterprise | $59 | $25 | $39 |
From this, it’s clear:
- Fellow offers the lowest entry point, making it appealing for teams with tight budgets.
- MeetGeek, while pricier, appeared to give the best value for multilingual support and meeting structure.
- Fireflies sat in the middle but had the most generous transcription limits for those doing a lot of recordings.
In our case, we found the free plans sufficient for running a thorough pilot test, but we also saw where upgrading could add value, especially when dealing with a higher number of meetings or if storage and integrations become essential.
- Best‑fit recommendations
My subjective pick: If you have a multilingual team and you want a more reliable tool between the 3 of them, MeetGeek would be your choice. In case you want a more structured agenda with smaller available budgets, you should probably consider using the Fellow.app.
MeetGeek.ai earned the highest marks across language accuracy, speaker tracking, and ease of use. In an all‑online meeting, each participant on their own mic and speaking English, its summary reflected the main topics of the discussion. Fellow app was almost close to the 1st place as well. If I was to choose on upgrading the plan for one of these 3 apps, I would probably choose Fellow app, to check on further, due to its prices.
General remark: Romanian performance still needs polish for both applications, and I expect that to improve. Speech‑to‑text models keep evolving. English still wins today, but regional‑language accuracy improves with every model update. We fully expect Romanian transcription parity within the next product cycles.
What’s next at ISD
- Paid‑tier trials. To see if premium models improve Romanian recognition.
- Microsoft Teams built-in AI agent tool (only available via Premium subscription). Compare achieved results.
- Larger sample. Promote the usage of AI meeting note taker apps and have a common template of meeting minutes generated at company level.
Stay tuned, more follow-ups will come over, and if you haven’t tried an AI meeting agent, this article will help you pick the tool that suits you best!