The short version: AI meeting notetakers like Otter, Fireflies and the built-in Zoom and Teams transcribers are useful for internal calls, but they’re less reliable than most people assume with regional accents, crosstalk, and numbers, and they can quietly damage client trust if you drop one into a call without asking first. I still take my own notes on every call that matters, and the recording is a backup, not the record.
Why every small business suddenly has a bot on the call
Eighteen months ago almost none of my clients had a notetaker on their calls. Now I’d say seven or eight out of ten do, whether that’s Zoom’s built in AI Companion, Fireflies, Otter, or Teams Premium’s recap feature. It happened fast, and I understand why. A one-hour discovery call generates roughly 6,000 to 9,000 words of transcript. Typing that yourself, or even trying to keep proper notes while you’re also meant to be listening and asking good follow-up questions, is hard. I get it. I’ve been the person scribbling half a sentence while the client is three points further along, and I’ve missed things because of it.
So the appeal is obvious. What’s less obvious, because almost nobody who sells you these tools wants to say it, is how often they get things wrong in exactly the moments you need them to be right.
The call where it mattered
Last year I ran a strategy call with a client based in Manchester, a home care franchise owner with a strong regional accent, on a call that also had her operations manager joining from a car, on speaker, with road noise in the background. The AI transcript came back afterwards looking perfectly professional. Clean paragraphs, timestamps, a tidy summary at the top. Except in the summary it had written that she wanted to “reduce headcount by 20 percent,” when what she’d said was she wanted to reduce her ad spend by 20 percent and reinvest it in training. One word, completely different meeting. I caught it because I’d been taking my own rough notes alongside the recording, out of habit rather than distrust. If I’d forwarded that AI-generated recap to her team without checking it, as I’ve seen other consultants do, it would have caused a awkward internal conversation about redundancies that were never on the table.
That’s the story I tell people now when they ask if these tools are worth using. Yes. And check every single recap before it goes anywhere near a client, every time, no exceptions.
The accuracy problem nobody puts in the marketing
Every AI notetaker vendor quotes transcription accuracy in the high 90s. What they mean is accuracy on clear audio, one speaker, no accent variation, no background noise, no overlapping speech. That’s not most business calls. Add a second speaker talking over the first, a regional British accent, an Israeli or Indian English speaker, or someone dialling in from a car or a warehouse floor, and accuracy drops fast. I’ve seen figures cited around 70 to 80 percent word accuracy in those messier real world conditions, which sounds fine until you realise that the words most likely to get mangled are numbers, names, and negatives, exactly the words that change the meaning of a sentence. “We’re not going ahead with the £15,000 package” and “We’re going ahead with the £15,000 package” differ by one small word that transcription software drops surprisingly often.
This matters more the more diverse your client base is. I work across the UK, the US and Israel, and the calls where the AI notes are weakest are almost always the ones with a mix of accents on the line, or where two people are talking at once because someone’s excited about an idea. Those are often the most important five minutes of the whole call.
The trust problem clients don’t always say out loud
Here’s the part most people writing about this topic skip over. Some clients, particularly the ones over 50, or in more traditional industries like law, accountancy or construction, do not love being recorded and transcribed by an AI, even when they agree to it on the call. They agree because saying no feels awkward. I’ve had two separate clients mention, weeks later, that they’d felt slightly on edge during a call because they knew it was “being listened to by a robot.” Neither said anything at the time. Both told me eventually, once the working relationship was established enough for it to feel safe to say.
That’s an uncomfortable thing to admit as someone who runs a business built partly on being an early and enthusiastic adopter of AI tools, but it’s true, and I’d rather say it than pretend every client is delighted their sales call is being transcribed and summarised by software. Some are. Some tolerate it. A few actively dislike it and never tell you.
How I use them now, step by step
- I ask before I record, out loud, in the first minute of the call, not buried in a calendar invite nobody reads: “Is it alright if I let the AI notetaker run so I can focus on the conversation instead of typing?” That single sentence has never once been refused, but asking it changes the dynamic completely compared with the bot just silently joining.
- I take my own short notes anyway, on paper or in a plain document, three to five bullet points per call maximum. Not because I don’t trust the tool, but because the act of writing something down by hand is what makes me remember it later, and it gives me something to check the AI summary against.
- I never forward the raw AI summary to a client. I read it, correct it, and rewrite the action points in my own words before anything goes out. Takes me about five minutes per call. That five minutes is where the Manchester mistake would have been caught by anyone using the process.
- For anything involving a number, a price, a date, or a decision, I read that specific line back to the client on the call itself: “So just to check I’ve got this right, you’re happy to go ahead at £2,400 a month starting the first of March.” That one habit fixes almost every transcription error before it ever reaches a document.
- I delete recordings after 30 days unless there’s a specific reason to keep one, mostly because GDPR compliance is my responsibility as the person who called the meeting, not the software vendor’s, and clients rarely think to ask whether the recording is still sitting on a server somewhere six months later. I do.
When I switch it off completely
Any call involving a redundancy conversation, a sensitive personal disclosure, a legal matter, or a first meeting with a new client where I want them to feel they can speak freely without wondering who else will read a transcript. In those calls I take notes the old way, by hand, and I tell the other person plainly that I’m not recording. That single sentence has, more than once, changed how open a new client was willing to be in that first conversation. People say different things when they know the only record is the one person in front of them writing a few lines by hand.
The productivity trap underneath all of this
The bit I’d flag to anyone starting out with these tools is what happens to your own listening once you know the AI is catching everything. You relax. You stop holding the thread of the conversation in your head, because you know you can look it up later. Over a few months I noticed my own follow-up questions on calls got worse, not because I’m less capable, but because I’d quietly outsourced the job of remembering to the software, and remembering is half of what makes a good follow-up question possible in the moment, not two days later reading a transcript. I’ve written before about the gap between what AI content does for your business and what it looks like it’s doing, in the piece about the honest math behind 569 AI-written blog posts, and this is the same pattern in a different room. The tool does the visible task well enough that you stop noticing the invisible skill it’s quietly replacing.
Since I noticed that, I’ve gone back to keeping my own three bullet points every call, AI notetaker running or not. It slowed me down for about a week. It’s made every call since noticeably better, because I’m still doing the thinking, and the transcript is just there to catch what I missed rather than to do the remembering for me.
Where they earn their keep
Internal team calls, weekly check-ins, and any meeting where the point is a shared record rather than a relationship, that’s where these tools are worth every penny. When I was rebuilding parts of my own business after a rough few years, documented in the month my website started paying me again, having an accurate log of what was agreed on internal calls with contractors and freelancers saved me from at least three arguments about who’d committed to what by when. Nobody’s trust gets dented by an AI transcript of a Tuesday morning team catch-up. It’s the client-facing, relationship-building calls where you need to think harder about whether the bot belongs in the room at all.
Frequently asked questions
Do I need to tell clients I’m using an AI notetaker on the call?
Yes, and in the UK and EU you’re legally required to under GDPR if you’re recording and processing someone’s voice data, not just polite to. A single sentence at the start of the call covers it: ask permission out loud rather than relying on a line in a calendar invite nobody reads.
How accurate are AI meeting notetakers really?
On clean audio with one clear speaker they’re strong, often 90 percent plus. On real business calls with accents, crosstalk, or background noise, accuracy commonly drops into the 70 to 80 percent range, and the words most likely to get mangled are numbers, names and negatives, the exact words that change what a client agreed to.
Should small businesses stop using AI notetakers because of these accuracy problems?
No, they’re still worth using, particularly for internal team calls. The fix isn’t to abandon them, it’s to never send a raw AI summary straight to a client without reading it first, and to read numbers and dates back out loud during the call itself so any error gets caught in the moment rather than three weeks later.
Which calls should never have an AI notetaker running?
Redundancy or disciplinary conversations, legal matters, and first meetings with new clients where you want them to speak freely. Telling someone plainly that you’re not recording, and taking notes by hand instead, tends to make people noticeably more open in exactly those conversations.
Related reading: I Let an AI Notetaker Sit in on 90 Days of Sales Calls. Here’s What It Caught, and What It Wrecked and Best Free AI Tools for Small Business in 2026.
If you want the full breakdown, here is everything I know about AI marketing.
