The short version: I ran an AI notetaker (Fireflies, then Fathom) on every client call for six months and it saved me roughly 6 to 8 hours a week in follow-up emails and typed-up actions. It also created a written transcript of every promise I ever made on a call, which is brilliant until a client uses it against you in a dispute. Consent, not the note-taking, is the part that costs you clients if you get it wrong.
Why I started recording every call in the first place
I was losing detail. Not big things, small ones. A client would say "can you send that proposal by the 14th, not the 20th" halfway through a rambling 45-minute call, and by the time I sat down to write the follow-up email I'd remember the deadline but not the exact date. Multiply that across 15 to 20 calls a week and you get a business that runs on slightly wrong memory.
So in mid-2025 I started running Fireflies.ai on Zoom and Google Meet calls, then switched to Fathom around month three because the summaries were shorter and less padded with filler. Both sit in the call, transcribe it, and spit out a summary with action points within a couple of minutes of the call ending. Neither is expensive: Fathom's paid tier was around £15 to £19 a month per seat when I was using it, Fireflies similar. This isn't a tool review though, it's what six months of doing it did to how I run client work.
The first awkward moment
Three weeks in, on a call with a client I'd worked with for two years, her ops manager stopped mid-sentence and said "wait, is that thing recording us?" I said yes, it takes notes so I don't miss anything. She went quiet, then said "I'd rather you didn't, if that's alright." I turned it off. That call taught me more than the other 117 combined: some people do not want a permanent, searchable, quotable transcript of what they say on a call, even when nothing sensitive is being discussed. It is not about privacy in the abstract, it is about not wanting a throwaway comment to exist forever in text.
After that I built a one-line habit into every first call with a new client: "I use an AI notetaker for accuracy, happy to turn it off if you'd rather I didn't." Out of roughly 30 new clients across the six months, four asked me to turn it off, always. Two asked me to turn it off only for the first fifteen minutes, when they were still working out how much to trust me. Nobody ever asked me to turn it back on once it was off.
What it changed, in real numbers
Across 118 recorded calls between June and November 2025:
- Average follow-up email time dropped from around 12 minutes (typing from memory or scribbled notes) to about 3 minutes (editing an AI-generated summary and pasting it in)
- I stopped missing action items entirely on long calls. Before this, on any call over 40 minutes I'd typically lose one small commitment. After, zero, because the transcript catches everything even the throwaway "oh and can you also..."
- I recovered roughly 6 to 8 hours a week that used to go on writing up notes, based on tracking my own time for four weeks in September
- Three disputes over what was agreed on a call got resolved in minutes rather than becoming a "he said, she said" back and forth, because I could pull the exact line from the transcript
That last point cuts both ways, and it's the part most posts about AI notetakers skip past.
Where this gets uncomfortable
A transcript of everything you say on every call is a record that can be used against you as easily as for you. I had one client, a small logistics firm in Leeds, query an invoice six weeks after a project wrapped, claiming we'd agreed a lower scope on the call. The transcript showed exactly what was agreed, word for word, and it favoured me that time. But I've also gone back through my own transcripts and found myself promising things off the cuff, on a Friday afternoon call, that I wouldn't have put in writing if I'd thought about it for thirty seconds. An AI notetaker doesn't just record your client's commitments, it records yours, and it doesn't forgive a badly worded sentence the way a fuzzy human memory does.
There's also a legal layer that most "how to use AI notetakers" articles don't mention at all: in the UK, recording a call without telling the other party can breach UK GDPR because a transcript containing someone's spoken words and opinions is personal data, and processing it without a lawful basis (which usually means telling them and getting at least implied consent) is a problem. Under UK GDPR, individuals also have a right to know what data you hold on them and to request it, which means a client can ask for the transcripts of your calls with them. I've had one do exactly that, eighteen months into a contract, for reasons that were never fully explained. I'd rather have a habit of clean consent from call one than find out my recording practice is the reason a client's solicitor is emailing me.
How I set it up without annoying clients
If you want to try this in your own business, here's the version I'd do from day one rather than working it out the hard way over six months:
- Say it out loud on the call, every time, for the first three calls with any new client, even if the tool announces itself. "I'm running an AI notetaker so I don't miss any actions, happy to pause it if you'd rather" takes eight seconds and saves an awkward moment later
- Put a line in your contract or terms saying calls may be transcribed by AI tools for accuracy and internal use. Most clients never mention it, but the one time it matters, you want it written down somewhere they agreed to
- Never let the summary go out unedited. AI summaries get tone wrong more often than they get facts wrong, and a client reading a slightly clipped, robotic version of their own words back to them feels worse than no summary at all
- Delete transcripts on a schedule rather than keeping them forever. I keep mine for 12 months then bulk delete, because a permanent archive of every call is a liability, not an asset, once a project is long finished
- Turn it off for anything involving money disputes, redundancy conversations, or anything personal. Those are exactly the calls where a client wants to speak freely, not be quoted back to themselves
What I'd tell someone starting this next month
If you're not technical and the idea of adding another tool to your stack feels like one more thing to learn, I get it, that's exactly the ground I cover in how I use AI day to day without being a techy person. An AI notetaker is one of the lowest-effort AI tools to add because it needs almost no setup, it just sits in the background of a meeting you're already having. The learning curve is entirely about how you talk to clients about it, not the software itself.
The bigger lesson from six months of this wasn't about the tool at all. It was that adding AI into any client-facing part of your business changes the relationship slightly, in ways that have nothing to do with efficiency. A notetaker makes a call feel more formal to some people and completely unnoticed to others, and you won't know which type of client you're on the phone with until you've asked. I found that out the same way I found out most things while rebuilding this business after a rough few years, which I've written about more openly in the month my website started paying me again: by doing the thing imperfectly first, then fixing the process once it had already gone slightly wrong in public.
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.
If you're weighing up whether to bring in outside help to set tools like this up across a whole team rather than just your own calls, that's a fair use of an AI implementation coach, because the tool itself takes ten minutes to connect and the consent, storage and contract side takes considerably longer to get right.
One last thing worth saying plainly: don't assume the transcript is accurate just because it sounds confident. I've caught Fathom mishearing "fourteenth" as "fortieth" and quietly correcting nobody, the same overconfident-but-wrong pattern I saw when I tested AI content generation at scale and wrote up what happened when I published 569 AI-assisted blog posts. AI tools that sound certain are not the same as AI tools that are correct, whether they're writing your blog or writing down what your client just told you.
Frequently asked questions
Is it legal to use an AI notetaker on client calls in the UK?
Yes, as long as you tell people it's being used and you have a lawful basis for processing the recording, which for most small businesses means clear consent and a line in your terms about it. Recording without any mention of it at all is where the legal risk sits, not the tool itself.
Which AI notetaker should a small business start with?
Fathom and Fireflies are the two most common starting points, both with free tiers, both around £15 to £19 a month for the paid version per seat. Otter.ai is another widely used option. Try one for a month on internal calls before you put it in front of clients.
Do clients mind being recorded by an AI notetaker?
Some do, most don't. In my own six months of use, about 1 in 8 new clients asked for it to be turned off, always on the first call, and none of them changed their mind later. Asking upfront rather than assuming consent is what keeps the relationship intact either way.
What's the biggest risk of running an AI notetaker on every call?
It's not accuracy, it's permanence. A transcript captures throwaway comments and half-formed promises from both sides of the call and keeps them forever unless you actively delete them, which means you're creating a legal record of everything said, not just the useful parts.
Related: looking for an AI consultant for small business? Here is how I can help.
Related reading: AI Notetakers Are Quietly Costing You Client Trust. Here's What I Do Instead and Should You Let an AI Notetaker Into Your Client Meetings?.
