AI Meeting Notetakers Compared: Fireflies and Google Meet in Hindi and Gujarati

If your calls switch between English and Hindi or Gujarati, check the language list before you trust an AI notetaker. Mine did not handle it.

I tried two AI notetakers on my own calls: Fireflies and the Take notes feature in Google Meet. I used them on 2 meetings, each about 30 minutes long. Across those calls I hit two problems. The notes mixed up names and who was speaking. And they missed the parts of the conversation that were in Hindi or Gujarati rather than English.

What each tool says it supports

	Google Meet, Take notes	Fireflies
How it joins	Built into Meet	Joins your call as a notetaker
Languages listed	8: English, French, German, Italian, Japanese, Korean, Portuguese, Spanish	English first; Spanish, French, Portuguese, Italian and 100+ others "in beta"
Hindi / Gujarati	Not listed	Not named; would fall under beta
Who can use it	An eligible Workspace edition or Google AI plan, held by the meeting organiser	Free plan, with more transcription credits on Pro and Business

Both lists were checked on 30 September 2026.

Why the Hindi and Gujarati parts went missing

Google's list matches what I saw from Meet: Hindi and Gujarati are not on it. Fireflies does not name them either, and it describes everything outside English as beta.

The two were not equally bad, though. Fireflies did a little better on those stretches. It picked up something from the Hindi and Gujarati parts. Meet gave me nothing from them at all, as if those minutes of the call had not happened. Neither produced notes I would send to anyone, so this is a difference of degree. I am not recommending either one. If your calls are bilingual and you have to pick one of these two today, Fireflies is the one that at least tries.

It also explains the shape of what came back. A notetaker that drops the sections it cannot follow leaves a transcript that looks complete. There is no gap, no marker, no note saying it stopped understanding here. The English sentence before the Gujarati stretch sits next to the English sentence after it. Read the notes without having been on the call and you would never know a decision had been discussed in between. That is worse than an obvious blank, because a blank tells you to go and check.

Wrong speaker names are a separate problem, and in my notes it was the more dangerous one. The words were roughly there; they were attached to the wrong person. A line saying the wrong colleague committed to a date is a claim about who agreed to what, which makes it more than a transcription slip. If the notes say the wrong person agreed to something, fix that before you send them on.

Before you rely on one

On Meet, the organiser's Workspace edition decides whether Take notes is there at all, so on someone else's call the choice may not be yours.

Run it on one low-stakes call in the languages you actually speak, then read the notes against your memory of the call. Two calls was enough for me to see both faults, so you do not need a long trial. You need one call you were paying attention to, because the test is not whether the notes read well. It is whether they match what you remember happening.

Three things I would check on that call. Did it get the names right, and did it put the right words against the right name. Did anything said in a language other than English make it into the notes at all. And are there any decisions in the notes that you do not remember being made, which is the failure that comes from a tool stitching two halves of a conversation together across a gap it did not report.

If the answer to any of those is no, the notes are a draft that still needs checking before anyone treats them as the record. I still had to read them against my own memory of a 30 minute call, which is most of the work I wanted to avoid. For a call where it matters who said what, write your own three lines at the end instead.

If you already connect your Google Calendar to an AI assistant, as in my Claude Pro connector setup, the notes are one more thing it can read, so they need to be right. A wrong name in a transcript stops being a typo once something else starts quoting it back to you.