The short version: AI note-takers like Otter, Fireflies and Fathom are useful for small business owners running client calls, but the accuracy is patchier than the adverts suggest, the legal disclosure bit is more fiddly than a five-second popup, and the real value isn’t the notes at all. It’s what the transcript shows you about how you talk to clients, which is usually less flattering than you’d hope.
Why I started doing this at all
Eighteen months ago I was losing details. Not big things, small ones. A client would mention their daughter’s wedding in March and I’d forget it by April. Someone would say “we can’t go above 4k a month” in a discovery call and I’d write “budget flexible” in my notes because I was too busy nodding and thinking about my next question to catch it written down.
So I started running an AI note-taker on every client call. Not sales calls only, every call. Onboarding, quarterly reviews, the awkward ones where someone wants to cancel. Forty calls in, I have opinions.
What the transcripts get right
The tools are good at the obvious job: turning speech into text and pulling out a summary with action points. On a clean one-to-one call with two clear British accents and no background noise, accuracy sits around 90 to 95 percent. That’s real, and it’s a genuine time saver.
Where it falls over is anything messier than that. Three people on a call, someone with a strong regional or non-native accent, a bad Wi-Fi connection, someone eating a sandwich (yes, this happened), and accuracy drops hard, sometimes into the 60s. I had one transcript that turned “we need to look at your Q3 numbers” into “we need to look at your cutie numbers.” My client found that funnier than I did.
The summaries the AI generates are also confidently wrong in a specific way: they smooth over disagreement. If a client pushed back on price and I held firm, the AI summary often reads as “agreed on pricing” because that’s the tidiest sentence it can construct from the conversation. You cannot skip listening back to anything that mattered financially. The summary is a starting point, not a record.
The disclosure problem nobody warns you about
Here’s the bit that gets glossed over in every “top 5 AI note-taking tools” listicle. In the UK, recording a call where you’re a participant is generally lawful without the other person’s consent under the general law of confidence and data protection rules, but using an AI tool that processes, stores, and potentially trains on that recording is a different question under UK GDPR. You need a lawful basis for processing that data, and if you’re recording someone else’s voice and personal details, telling them matters both legally and, frankly, decently.
So I say it now, every single call, in the first thirty seconds: “I’m running an AI note-taker for accuracy, is that alright with you?” Most people say yes without blinking. About one in twenty pauses, and a couple of those have said no outright. One client, a solicitor, said “no, and please don’t ask me again,” and I never have.
The awkward truth is that most business owners using these tools skip this step entirely. They add it to Zoom, it whirs away in the corner, and nobody mentions it to the client. That’s not a small oversight. If someone found out afterwards that a full transcript of their confidential business call was sitting in a third party server they never agreed to, that conversation goes badly, and it should.
The bit that changed how I work
Notes were never really the point, though I didn’t know that when I started. The point turned out to be pattern-spotting across dozens of calls at once, which is something you cannot do from memory or scattered handwritten notes.
Around call 22 I searched my transcripts for the word “confused” out of curiosity. It came up eleven times, always from clients, always in the same part of the conversation: right after I explained pricing tiers. Eleven times I had thought I was being clear and eleven times I wasn’t. I rewrote that one explanation, tested it on the next six calls, and “confused” dropped to zero. That’s not a thing I would ever have noticed without a searchable record of my own words.
I’ve written before about how I use AI day to day without being remotely technical, and this is the same principle: the tool isn’t clever, the searching is. Any small business owner running 20 or more client calls a month has this same blind spot sitting in their own conversations, they just can’t see it because nobody writes down what they said, only what they meant to say.
The uncomfortable truth about what it does to trust
Here’s the part I don’t see written anywhere else. Once you know a call is being transcribed, you talk differently, and so does the client. People hedge more. They say “hypothetically” before opinions they’d have stated flatly a year ago. I noticed I was doing it too, softening things I’d normally say straight, because some part of my brain knew a permanent, searchable, exportable record now existed of me saying it.
That’s not necessarily bad. It made me more careful with promises I make on calls, which is no bad discipline for anyone in a service business. But it is a genuine change in how the conversation happens, and the “just switch it on, nobody will mind” advice you see everywhere skips straight past that.
A step-by-step for doing this
- Pick one tool and stick with it for a full quarter before judging it. Otter, Fireflies and Fathom all do roughly the same core job, switching between them constantly just loses you the searchable history.
- Say the disclosure line every time, out loud, in the first minute. Not a buried line in your terms of service, an actual spoken sentence.
- Check your data processing agreement with the tool covers UK GDPR if you have UK or EU clients, and read where the servers physically sit.
- Never let the AI summary be your only record of anything with a number, a date, or a promise attached to it. Listen back or read the full transcript for those bits.
- Once a month, search your own transcripts for words like “confused,” “expensive,” “worried,” or “not sure.” That fifteen minutes is worth more than most CRM reports.
- Delete transcripts you don’t need after 12 months. Storing every call forever is a liability, not an asset, if you ever get a subject access request.
When I’d tell you not to bother
If you do fewer than five client calls a month, don’t bother. The setup faff and the awkward disclosure conversation aren’t worth it for that volume, just take normal notes. And if your clients are in regulated industries, legal, medical, financial advice, check with them directly first rather than assuming your standard disclosure line covers it. Some sectors have stricter recording rules than general small business ones, and it’s cheaper to ask than to unpick afterwards.
If you’re weighing up whether this kind of thing is worth building into how your business runs day to day, or you want someone to set the workflow up rather than bolting on another app, that’s exactly the sort of practical groundwork an AI consultant for small business should be doing with you, not just recommending tools off a list.
I’ve been rebuilding this business in public for a while now, and the quiet, unglamorous wins like this one, catching my own bad habits in a transcript, have mattered more than anything flashy. It’s the same slow, unshowy work behind the month my website started paying me again. None of it looks exciting from the outside. It just works if you do it.
And to be clear, this is a different problem from the one I wrote about when I published 569 AI blog posts and Google indexed 5 percent of them. That was AI generating content nobody asked for. This is AI recording content that already existed, your own client conversations, and just making it visible to you for the first time.
Frequently asked questions
Is it legal to record client calls with AI note-takers in the UK?
Recording your own call is generally lawful, but processing someone else’s personal data through a third party AI tool needs a lawful basis under UK GDPR, and telling the other person you’re recording is the safer, more decent option even where it isn’t strictly mandatory. Say it out loud on every call rather than relying on small print.
Which AI note-taker is most accurate for client calls?
In my own use, Otter, Fireflies and Fathom perform similarly on clean, quiet, one-to-one calls, around 90 to 95 percent accuracy. All of them drop noticeably with multiple speakers, accents, or background noise, so don’t pick a tool on accuracy claims alone, pick one and test it for a quarter.
Do AI meeting summaries capture disagreements accurately?
Not reliably. AI summaries tend to smooth conflict into agreement because that produces a tidier sentence. Never rely on the summary alone for anything involving price, dates, or a promise, always check the full transcript for those.
Is it worth using AI note-takers if I only have a handful of client calls a month?
Probably not. The setup, the disclosure conversation, and the data storage housekeeping only pay off once you’re running enough calls that pattern-spotting across them becomes useful, typically five or more a month. Below that, normal notes do the job fine.
Related reading: AI Meeting Notes: What They Get Right, What They Miss, and When I Switch Them Off and AI Meeting Notes Are Quietly Costing You Client Trust.
I go much deeper on this in the AI marketing guide.
