How I Chose My AI Notetaker (and why I'm sticking with it for now)

AI notetakers are no longer a foreign technology. Google now has Gemini built into Meet. People are loving Otter. Fathom has been popping up more often (I don't like how this one cuts off access while still looking like you're recording. Look out if you try it!)

And Fellow and Granola deserve honorable mentions*

With emergent technology that scales so quickly, my main concern is how adoption naturally moves faster than understanding.

Most conversations I hear about AI notetakers tend to fall into one of two categories. People are either excited about the functionality or concerned about privacy. Both are reasonable. What surprises me when I look more closely is how often those conversations blend together concerns that are actually quite different from one another.

I suspect part of this is a side effect of how quickly technology is being commercialized and adopted.

As I evaluate the various options on the market, I find myself less interested in identifying the "best" tool and more interested in understanding the landscape itself. I want to think through things like:

• What exactly are these tools doing?

• What are the practical tradeoffs?

• Which concerns are legitimate?

• Which ones are misunderstandings?

• And perhaps most importantly, what matters enough to influence my decision?

This article is simply a walkthrough of what I've learned so far, what ends up mattering most to me, and why I choose to use Fireflies.

The first challenge is defining the question

One of the reasons I think conversations about AI notetakers become confusing is that people are often talking about entirely different things while using the same language.

Privacy is a good example.

Someone might say they're concerned about privacy and mean that they don't want a bot joining meetings. Someone else might mean they don't want transcripts stored. Someone else might mean they don't want their meeting content used to train AI models. Another person might be concerned about compliance requirements within their organization.

All of those concerns fall under the broad umbrella of privacy. They are also distinct questions with different answers.

To my surprise, the more I separate the questions, the easier it becomes to evaluate the products.

Instead of asking, "Is this private?" I find myself asking:

• Is meeting content stored (and where, and for how long)?

• Is meeting content used to train AI models?

• Who has access to the transcripts?

• What controls are available to users?

• How does the tool enter meetings?

• What happens if I want information deleted?

• What security protocols do they use (and what does the answer even mean)?

Those questions are related, but they aren't interchangeable. Understanding the distinctions matters because the answers often differ.

There is no zero-risk option

I know, that's not a particularly satisfying thing to accept.

Many of us approach technology decisions hoping there will be a clear winner. A product that delivers all the benefits with none of the tradeoffs. I don't find that here, and I don't think expecting that from any AI technology is realistic.

So I look at the factors.

Every AI notetaker processes information somewhere. Every platform stores information somewhere. Every company has policies, infrastructure, vendors, and compliance frameworks that shape how information moves through the system.

As I test these tools, I stop looking for a tool that eliminates risk and start looking for a tool whose tradeoffs I understand. Not needing to find certainty allows me to focus on defining what I need from the tool and then understanding what I'm actually agreeing to in order to get those benefits.

Here's what matters to me:

I need an AI notetaker to capture my meetings and working sessions so that I can engage fully during the conversation. I want to be able to easily pull from transcripts afterward to create recaps, action items, and follow-up materials. I want to download transcripts and use them across other tools in my workflow. And I want to know I can refer back to a meeting months later if I need to.

I also don't want to spend time managing another system in order to get those benefits.

The tools I spend the most time evaluating**

There are quite a few AI notetakers available today, but I focus primarily on four whose functionality meets my needs:

• Fireflies

• Otter

• Fathom

• Google's Gemini (within Google Workspace)

See below for a quick security and feature comparison.**

What I appreciate most is that each seems to be solving a slightly different problem.

Otter stands out for real-time transcription and collaborative note-taking. Fathom has developed a strong reputation among users who prioritize generous free access and privacy-conscious functionality, though my experience has been somewhat different. Gemini benefits from living directly inside the Google ecosystem, which eliminates some of the friction associated with third-party meeting bots. And Fireflies stands out for search, retrieval, integrations, and its overall approach to meeting intelligence.

I don't come away with the impression that one tool has clearly won. What I do come away with is the sense that different people will likely arrive at different conclusions depending on what they value most.

The distinction that matters most to me

Storage and training are not the same thing.

These concepts are treated as though they are interchangeable surprisingly often. They're related, but they each have different criteria to consider.

A company may store transcripts so users can access them later. That does not automatically mean those transcripts are being used to train future AI models. They may train their own systems on customer data, prohibit training entirely, or take a mixed approach.

Things become even more complex when external AI providers are involved. A company may store customer data within its own platform while also sending information to third-party models to generate summaries, action items, or other outputs.

At that point, an entirely separate set of questions emerges:

• Is the third-party provider allowed to retain the data?

• Can they use it for training?

• What happens to the information after processing?

• Are there enterprise agreements restricting that use or determining timelines and disclosure?

This is where I spend most of my research time. The answers vary significantly between products, and they're much harder to understand than the marketing pages suggest.

For me, understanding this distinction clarifies a huge portion of the conversation.

What I am able to verify with Fireflies is that they state they maintain zero-data-retention agreements with the external foundation model providers used to generate summaries and insights. According to those agreements, meeting content can be processed by external AI providers without being retained for future model training.

Interestingly, Fireflies doesn't rely on a single AI company. Instead, they have formal partnerships with several major model providers, including OpenAI, Anthropic, Perplexity, and Google. When users interact with AskFred, they're leveraging frontier models through Fireflies' platform rather than interacting directly with the model providers themselves.

That distinction matters to me because it brings me back to the questions I was already asking: Who stores the data? Who processes it? Who can retain it?

These tools all answer those questions differently.

In many AI products, multiple companies are involved in delivering a single experience. The platform you're using may store information while another company processes it. One company may retain transcripts while another is contractually prohibited from doing so. Enterprise agreements, retention policies, and model-provider relationships all influence what happens behind the scenes.

And of course, if companies update their terms and conditions, these things can change over time.

The concerns are real

I want to be careful not to dismiss concerns simply because I've arrived at a conclusion.

There are legitimate reasons some people and organizations choose not to use these tools.

The most common concern I encounter is the meeting bot itself. Some organizations are entirely comfortable having a participant join for transcription purposes. Others are not.

Some institutions have restricted or blocked third-party meeting bots altogether because of concerns around data handling, unexpected meeting entry, or broader privacy requirements.

If I worked in an environment with strict compliance requirements or significant legal constraints, I'd likely make a different decision. Fireflies does offer SOC 2 Type II compliance and support for frameworks such as GDPR, HIPAA, and FERPA. But I'd still want to understand the specifics before making assumptions about how those protections apply in a given situation.

This is part of why context matters so much. Technology decisions rarely happen in a vacuum. What works well for a solo advisor, consultant, or small business owner may not be appropriate for a university, healthcare organization, or enterprise environment.

Privacy, security, and consent are not the same thing

One additional distinction that becomes important to me is that privacy, security, and consent are often treated as though they are interchangeable.

They are not.

A platform can have strong security controls, clear data retention policies, and robust compliance certifications while still creating a consent concern if participants don't realize they're being recorded or don't understand how the information will be used.

Likewise, a meeting can be fully consensual while still raising legitimate questions about storage, retention, training, or access.

For me, this is less about the technology itself and more about transparency. I want people to know when a meeting is being recorded, what tool is being used, and why it's there. Once that information is available, people can make informed decisions about their participation and organizations can make informed decisions about their policies.

And of course, I'll always remove my notetaker from a call if someone isn't comfortable with it.

Why I choose Fireflies

What ultimately keeps Fireflies at the top of my list isn't a single feature. It's the combination of privacy practices, functionality, and day-to-day usability.

The feature I underestimate most is retrieval.

Going into this process, I assume transcription accuracy will be the most important factor. What I realize instead is that most of these tools are now reasonably good at transcription. The bigger question is whether I can quickly find what I need later, and how much time it takes me to do so.

Fireflies consistently stands out.

I can locate decisions, action items, deadlines, feedback, and specific moments without rereading an entire transcript. The search experience is strong, and the ability to move between transcript and audio makes verification fast when something looks off.

I also use Fred, their built-in AI assistant, far more than I expect. Instead of reading through an hour-long meeting or hours of a working session, I can ask direct questions about what was discussed and retrieve the information within seconds. It has become one of the most practical features in my workflow.

The weekly digest and prep emails ahead of upcoming calls are another pleasant surprise. I originally assume email notifications will be a feature I turn off (who wants more automated emails?!), but they end up being an easy way for me to reconnect with what happened across a busy week and prepare for upcoming sessions without manually piecing together notes from multiple meetings.

It's rare to get platform-generated emails and actually open them consistently.

Why I'm sticking with it for now

Emphasis on "for now."

Technology changes quickly. Products evolve. Policies change. New competitors emerge. What feels like the right choice today may not be the right choice a few years from now.

At the moment, Fireflies provides the best balance of functionality, usability, and privacy considerations for the way I work. That doesn't make it the best option universally.

In fact, if you're already using Otter, Fathom, Gemini, or another platform and it's working well for you, I'm not sure I'd recommend switching.

What I would recommend is understanding what the underlying mechanisms are doing with your data and deciding whether those tradeoffs are worth the value you're receiving from the tool.

Understanding the landscape well enough to make an informed decision is key.

Notes:

—

*When considering Fellow and Granola, my preference for hands-off automation (auto-join bot), deep data interaction ("Ask Fred"), and immediate email distribution (recaps and prepping for each week's calls) makes botha poor fit for me. Fellow is about meeting lifecycle management (better for larger orgs) not a solo practitioner's workflow. And Granola is designed for people who want to actively type custom shorthand notes during the call, which the AI then cleans up afterward. It works great, I just want to sit back, converse, and let the AI do 100% of the capturing work.

**Tool Evaluation

Feature/Metric	Fireflies	Otter	Fathom	Gemini
Best For	Sales teams and deep CRM automation.	Real-time captions and collaborative notes.	Free users and video-heavy recording.	Enterprise-level Google Workspace teams.
Capture Method	Audio/Video bot joins the calendar link.	Audio/Video bot joins the calendar link.	Visual side-panel or manual bot trigger.	Built directly into Google Meet (No bot).
Compliance	SOC 2 Type II, GDPR, HIPAA, FERPA.	SOC 2 Type II, GDPR.	SOC 2 Type II, GDPR.	SOC 1/2/3, ISO 27001, HIPAA, FedRAMP.
AI Training / Data Use	Strict Opt-Out (Zero Data Retention policy).	Varies by account type and agreement.	Opt-Out available for organizations.	No Training on paid Workspace accounts.
Free Plan Value	Highly restricted; gates advanced summaries.	Limited monthly transcription minutes.	Generous; unlimited free recordings. RISK: In my experience, they cut off recordings after token limits were reached without a clear warning.	Paid Workspace add-on required.

Info sourced June 3, 2026, things can change. Always verify current policies directly before making assumptions about these things!