Straight answer: AI meeting note-takers like Fireflies, Otter and Fathom are useful, but if you let them replace the human job of listening and following up, your team gets slower and less accountable, not faster. The fix isn't dropping the tool, it's putting rules around it that most businesses never bother to write down.
What happened when I handed my meetings over to a bot
In early 2025 I ran a six month experiment with my own small content team. Every client call, every internal planning meeting, every discovery call, got recorded and transcribed by Fireflies. I did it for the obvious reason: I was tired of scribbling half a page of notes while trying to hold eye contact on a video call, then losing the page by Thursday.
For the first two months it felt like a small miracle. Full transcripts, automatic summaries, action items pulled out and dropped into a shared doc. I stopped taking notes almost entirely because why would I, the bot had it covered.
By month four I noticed something I didn't like. In a client review, I got asked why we hadn't actioned something they'd raised three weeks earlier. I checked the transcript. It was there, word for word, timestamped, summarised correctly by the AI. Nobody had read it. The summary existed. The doing didn't.
That's the bit nobody selling you a £24-a-month AI notetaker puts in the demo video: a perfect record of a conversation is not the same thing as someone being responsible for what happens next. If anything, a good transcript can make a team feel like the job is done the moment the meeting ends, because "it's all written down somewhere."
The uncomfortable part
Here's the bit I had to admit to myself. Before AI notetakers, when I took my own scrappy notes by hand, I remembered more of the meeting, not less. Writing something down in your own words forces your brain to process it. Reading a polished AI summary afterwards, weeks later, does not. There's decent cognitive science behind this (it's sometimes called the generation effect), and it's the same reason students who type lecture notes verbatim tend to retain less than students who summarise in their own words.
So the tool that was supposed to make my team more on top of things was quietly making us worse listeners in the room, because we knew we didn't have to be. That's an uncomfortable thing to say out loud when half the marketing world is selling AI meeting tools as a pure win. They're not a pure win. They're a trade. You get a searchable archive. You risk losing the muscle of active listening and personal accountability that used to come from the fact that if you didn't catch it, nobody did.
The numbers that made me change how we use it
I went back through three months of transcripts and did something most people never bother doing: I counted. Out of 40 recorded client and team meetings, 31 had action items flagged by the AI summary. Of those 31, only 19 had been completed by the date someone had committed to in the call. That's a 61% follow-through rate on things a machine had already written down for us in plain English, with names attached.
Compare that to the six months before we started recording anything, when I was still taking rough handwritten notes and typing up three bullet points after every call myself. Follow-through on those self-typed action items ran closer to 80%, by my own rough tracking in a simple spreadsheet. The AI gave us more complete records and worse execution. That's not what any vendor puts in a case study.
The rules I use now
I didn't ditch the tool. I still use Fireflies for most client calls, because for a business with clients across the UK, US and Israel, having an accurate record of what was agreed matters more than it used to when everything happened in one room. But I changed how we treat it. Five rules, all boringly simple, all things almost nobody sets up by default:
- One human owner per meeting. Every call has a named person, not the AI, responsible for reading the summary within 24 hours and turning it into two or three actual tasks in our project tool. The AI summary is a draft, not a to-do list.
- No recording without a stated reason. We stopped auto-recording everything. If there's no clear reason we'll need the transcript later, we don't record, because most short internal check-ins don't need a permanent archive and the habit of recording everything just creates noise nobody reviews.
- Someone still writes three lines by hand. Whoever owns the meeting jots three lines in their own words before they look at the AI summary. It's a two-minute habit that keeps the retention benefit of note-taking without giving up the archive.
- A weekly 15-minute review, not a daily scroll. Instead of everyone dipping into transcripts constantly, we review outstanding action items from the week's meetings in one short Friday slot. It's the single change that pushed our follow-through rate back up, from that 61% to somewhere around 85% over the following quarter.
- Clients get told they're being recorded, every time, in plain words. Not buried in a calendar invite footer. Said out loud, at the start of the call, with a chance to say no.
The privacy problem almost nobody flags
That last rule matters more than it sounds. Under UK GDPR, recording someone on a call and running it through a third party AI tool for transcription is processing their personal data, and in most cases you need a lawful basis and, practically speaking, their knowledge. A lot of small businesses have quietly turned on "always record" in their AI notetaker settings and never once asked a client if that's fine with them. If your notetaker also auto-joins external calls with vendors, prospects or partners, you're potentially recording people who never agreed to it and who may not know their words are sitting in a US-based server being summarised by a language model. I've had exactly one client, a solicitor, refuse a recorded call outright, and she was right to ask.
This is the sort of detail that gets skipped in the "10 best AI meeting assistants" roundups, because it's not a fun feature to write about. But if you're running a small business and building client trust the way brands like Away built theirs, through consistency and clear communication rather than clever tricks, this is exactly the kind of unglamorous detail that protects that trust.
Where AI notes earn their place
None of this means don't use them. There are three situations where I think an AI notetaker is worth every penny of the roughly £15 to £35 a month per user that tools like Otter Business, Fireflies or Fathom charge:
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.
- Client discovery calls, where you need an exact record of scope, price and expectations agreed verbally, because memory disputes over what was promised cost real money later.
- Multi-timezone teams, where half the people who need the information couldn't attend live and a searchable transcript beats a garbled voicemail summary.
- Compliance-heavy conversations, where you need proof of what was said and when, not just what someone remembers.
Where they earn their keep least: quick internal brainstorms, one-to-ones, and anything where the value is in the thinking that happens in the room, not the record of it afterwards. Howard Schultz built a culture at Starbucks partly by insisting on being physically present and listening to baristas directly rather than working from reports about them, and there's a version of that lesson in how Schultz's business decisions held up over decades. A transcript can't replace someone paying attention in the room.
Building this into how your business runs
If you're a solo founder or a team of five, the mistake is treating an AI notetaker as a plug-and-play fix rather than a process change that needs owners and rules, in the same way Napoleon Hill argued that a decision without a definite person responsible for carrying it out rarely gets carried out at all. His writing on decisiveness, covered well in this piece on Napoleon Hill's business lessons, applies almost exactly to what goes wrong with meeting AI: the decision gets made, it gets recorded beautifully, and then it evaporates because nobody was named as the person who'd carry it through.
The businesses that get real value from AI notetakers tend to be the ones that treat the rollout like they'd treat any other operational change, with someone testing it for a set period, measuring the actual follow-through rate rather than assuming the tool is working because the transcripts look tidy, and adjusting before it becomes a permanent habit nobody questions. That's also, frankly, the exact conversation I have most often when I sit down with a small business through an AI implementation coach engagement: it's rarely the tool that's the problem, it's the absence of a simple owner-and-review system around it.
Documentation habits matter too. Ahrefs built a huge part of its reputation by writing down and publishing exactly how it does things, and there's a lesson in that worth borrowing here: the value isn't in having a record, it's in someone reading the record and acting on it. The detail in how Ahrefs approaches its content strategy is that nothing gets published without a clear next step attached, which is precisely what most AI meeting summaries lack by default.
The short version to put into practice
Keep the AI notetaker if it's saving you time on records you need. Stop letting it be the reason nobody in the meeting is really listening. Name one owner per call. Make them write three lines by hand before they read the AI summary. Review action items on a fixed weekly slot rather than trusting they'll get done because they're written down somewhere. And tell every client, out loud, that the call is being recorded, before you press start.
Frequently asked questions
Do I need client permission to use an AI notetaker on calls?
Under UK GDPR and similar rules elsewhere, yes, in practice you need to tell the other person and give them a genuine chance to object, not bury it in a calendar invite. Say it out loud at the start of the call.
Which AI meeting notetaker is best for a small business?
Fireflies, Otter and Fathom all do a similar job for £15 to £35 a month per user. The tool matters far less than whether someone is assigned to read the summary and turn it into tasks within 24 hours, which most teams skip.
Does using AI notes make meetings less effective?
It can, if people stop paying attention because they assume the transcript covers it. In my own six-month tracking, action item follow-through dropped from around 80% with handwritten notes to 61% once we relied fully on AI summaries, until we added a weekly review habit that brought it back up.
Should small teams stop recording meetings altogether?
No, but record with a reason. Client discovery calls and compliance conversations benefit from a transcript. Quick internal check-ins and brainstorms usually don't need one, and recording everything by default just creates an archive nobody reviews.
Related reading: AI Automation for Professional Services Firms and AI News This Week for Business, 12 July 2026.
