The short version: AI notetakers like Otter, Fireflies and Fathom save you real time on admin, but they quietly change how clients behave on calls and create a pile of stored transcripts that becomes a liability the moment something goes wrong. I used one on every call for over a year, lost a client because of it, and now only switch it on when I have a specific reason to.
The 14 months I let a bot sit in on every call
From late 2024 through most of 2026 I had an AI notetaker join every single client call. Discovery calls, catch-ups, awkward "can we push the deadline" conversations, all of it. I started with Otter, moved to Fireflies for about eight months because the summaries were sharper, then tried Fathom for the last stretch because it was free for solo users and did a decent job pulling out action points.
The pitch is obvious and it's true: you stop scribbling notes mid-conversation, you get a searchable transcript, you get a tidy summary emailed to you twenty minutes later, and you can quote a client back to themselves word for word six weeks later when they claim they never agreed to the scope you're now delivering. For someone rebuilding a business in public, that last bit matters more than people admit. I've had a client insist in writing that we never discussed a second phase of work, and I pulled the transcript line where they said "yes, let's do phase two once the first bit lands." That single moment probably paid for a year of subscription fees.
So the tools work. That's not in question. What I'm less convinced about, after fourteen months of running one on autopilot, is whether they should be on by default.
The call where it cost me a client
Here's the story that changed my mind. I was on a discovery call with the owner of a small bathroom fitting business in Kent, maybe eight employees, decent turnover, wanted help getting his Google Business Profile and local reviews sorted along with some basic AI-assisted content. Twelve minutes into the call he stopped mid-sentence and said, "who's Fireflies?"
The little bot had joined as a visible participant with its own name and camera-off tile, same as it did on every call. I explained it was a notetaker, that it transcribes so I can focus on the conversation instead of typing. He went quiet for a second, then said something I've thought about a lot since: "I don't love that I don't know where that's going."
He wasn't wrong to ask. He had no idea whether the recording sat on a server in the US, whether it trained a model somewhere, whether anyone else at my end could pull it up later. I gave him a straight answer, he seemed fine with it in the moment, and then he never booked the follow-up call. I'll never know for certain it was the bot, but it's the only thing that changed between a warm inbound enquiry and a lead that went cold within 48 hours.
What these tools store, and for how long
This is the bit most write-ups skip. On Fireflies' standard business plan, recordings and transcripts sit on their servers indefinitely unless you manually delete them. Otter's business tier keeps data for the life of your account. Fathom keeps recordings for a rolling period tied to your plan, but the transcript and summary text often lives on in your linked CRM or Slack workspace long after the recording itself expires. None of that is hidden in bad faith, it's in the terms most of us never read before clicking accept, but it means every sales call, every client complaint, every off-the-cuff comment about a competitor is sitting somewhere as searchable text.
For a UK-based consultant working with clients across the US and Israel, that's not a small detail. GDPR requires a lawful basis for recording someone and a way for them to ask what you're holding and have it deleted. Most small businesses running these tools have never once had that policy written down, me included until this year.
The uncomfortable part nobody selling these tools wants you to sit with
The pitch is efficiency. The reality, fourteen months in, is that the transcript archive becomes a second inbox you never asked for. I had over 300 recorded calls sitting in Fireflies by the time I did a proper clear-out. I had gone back and reread maybe six of them. The rest were just there, unread, unactioned, a growing pile of other people's words that I was legally responsible for and functionally never used.
That's the uncomfortable truth: the value of the notetaker isn't in the archive, it's in the single moment you need to pull one specific quote from one specific call. You don't need 300 recordings for that. You need a habit of flagging the calls that matter and being disciplined about deleting the rest. Nobody selling you the subscription tells you that, because "record everything, forever" is the model that keeps you paying £20 to £30 a month per seat indefinitely.
Where they still earn their place
I'm not anti-AI on calls, I use AI every working day for far more sensitive things than transcription, and I've written before about exactly how I use AI without any technical background. Notetakers are useful in three specific situations:
- Contract and scope calls, where a word-for-word record protects both sides if there's a dispute later.
- Team handover calls, where three people need the same accurate summary and nobody wants to write it twice.
- Discovery calls with a new client, provided you tell them upfront and give them the option to say no.
What I stopped doing is running it on internal chats with my own contractors, casual catch-ups with existing clients I've worked with for years, and anything where the person on the other end hasn't explicitly agreed. That covers maybe 60 percent of the calls I used to record without thinking.
How I run calls now
These days the notetaker is off by default. Here's the actual process I follow:
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
- Before any call with a new contact, I say in the calendar invite: "I sometimes use an AI notetaker for accuracy, let me know if you'd rather I take manual notes." That single line has never once caused a problem, and it means the person isn't ambushed mid-call.
- For contract or pricing calls specifically, I turn it on and mention it out loud at the start, not as a footnote in an invite.
- Every quarter I go into Fireflies and delete anything older than 90 days that I haven't flagged. It takes about twenty minutes and it means the archive never becomes the unmanaged pile it was before.
- I don't record any call where a client is upset, giving me sensitive personal information, or discussing a competitor by name. Manual notes only. Some conversations shouldn't live in a searchable transcript on a third party server, however good the AI is.
If you're running a small team and want someone to build this kind of policy rather than bolt it on after a client asks awkward questions, that's exactly the sort of practical, unglamorous thing an AI implementation coach should be doing with you in the first session, not the flashy stuff.
A simple rule before you switch one back on
If you're using or considering an AI notetaker, ask yourself one question before every call: would I be comfortable if the other person read this transcript back to me in six months, out loud, in front of someone else? If the honest answer is no, don't record that call. It's the same discipline I've had to apply to AI content generally. I learned the hard way, after publishing 569 AI-assisted blog posts and watching Google index just 5 percent of them, that "AI can do this" is a completely different question from "should this run unsupervised." The same logic applies to a bot quietly recording someone's business problems back to you.
The month my traffic and revenue turned a corner, documented honestly in the piece about my website paying me again, wasn't the month I added more automation. It was the month I got more deliberate about which parts of the business stayed manual on purpose. AI meeting notes are a small example of the same pattern.
What to check before your next call
If you want to compare tools rather than take my word for it, Otter, Fireflies, Fathom and Read.ai are the four worth trialling, all with free tiers generous enough to test on ten or so real calls before you commit to a paid seat. I've also rounded up a much wider set of underused business tools, several of them relevant here, in my list of 50 websites you probably didn't know existed.
Whichever one you pick, write down your own three-line policy on when it's on, when it's off, and how long you keep the transcripts. It took me fourteen months and one lost client to work out mine. You can save yourself both.
Frequently asked questions
Do I legally have to tell people I'm recording a call with an AI notetaker?
In the UK, under GDPR, you need a lawful basis to record someone and should inform them before the call starts, ideally in the invite and again verbally. In the US it varies by state; some, like California, require consent from all parties on a call, not just one. When in doubt, say it out loud at the start of the call rather than relying on a small on-screen name.
Which AI notetaker is best for a small business?
Fireflies gives the sharpest summaries and integrates well with CRMs, Otter is the most established with a generous free tier, and Fathom is worth trying if you want a free option for a solo consultant. None of them are meaningfully better at the core transcription, the differences are in summary quality and where the data sits afterwards.
Should I use an AI notetaker on every client call?
No. Use it on calls where an accurate written record protects both sides, contract discussions, scope changes, complex handovers. Skip it on casual check ins, calls involving sensitive personal information, or any conversation where the other person hasn't clearly agreed to being recorded.
What happens to the transcripts after the call?
Most tools keep recordings and transcripts on their servers until you manually delete them, often for the life of your account on paid plans. Set a quarterly reminder to review and delete anything you haven't flagged as important, otherwise you end up with hundreds of unmanaged recordings you're still responsible for.
Related reading: AI Phone Receptionists for Small Business: What They Cost and Where They Quietly Lose You Money and AI Phone Agents for Small Business: What Happens When You Let AI Answer Your Calls.
