The short version: AI notetakers like Otter, Fireflies, Fathom and Zoom AI Companion are now sitting in on client calls whether or not anyone asked, and that’s quietly costing small business owners deals, candour and trust. Announce the bot, offer a real opt-out, and know what happens to the transcript after the call ends, or don’t use one at all.
Worth reading next: What AI Notetakers Are Doing To Your Sales Calls (And How To Fix It).
Worth reading next: Why Your AI Meeting Notetaker Might Be Breaking the Law (And Killing Y.
The bot that joined without asking
Three months ago I was on a discovery call with an operations director I’ll call Mark, who runs a 40-person logistics firm in the Midlands. Two minutes in, a little icon popped into the corner of his screen: “Fireflies.ai Notetaker has joined the meeting.” Mark stopped talking. Literally mid-sentence.
“Sorry, who’s that,” he said. It wasn’t a person. It was a transcription bot his own marketing manager had set up to auto-join every calendar event with an external guest. Mark didn’t know it was there until it announced itself. He spent the rest of that call choosing his words like he was giving evidence.
That call taught me more about AI adoption in small business than any report I’ve read. The tool wasn’t the problem. The silence around it was.
What these tools are quietly doing behind the scenes
Every AI notetaker does roughly the same three things: joins the call as a visible or semi-visible participant, records audio and often video, then produces a transcript and summary you can search, share or paste into a CRM. The differences are in price and what happens to your data next.
- Otter.ai: free tier gives 300 minutes a month, Pro is around 16.99 US dollars a month per user, and it stores transcripts on Otter’s servers unless you actively delete them.
- Fireflies.ai: free tier is limited to 800 minutes storage, Pro sits around 10 US dollars a month, and by default it can use call content to improve its own models unless you switch that off in settings.
- Fathom: free for individuals, paid team plan around 19 US dollars a month, popular because it auto-syncs summaries straight into HubSpot and Salesforce.
- Zoom AI Companion: bundled into paid Zoom plans, no separate visible bot icon, which is exactly why it worries me more than the others.
- Read AI: around 19.75 US dollars a month, and it goes further than transcription, scoring participants on “engagement” and sentiment during the call, which is a strange thing to have quietly happening in the background of a sales conversation.
None of that is expensive. A solo consultant can run one of these for less than the price of two coffees a week. The cost isn’t financial. It’s what it does to how people talk to you.
The bit nobody wants to say out loud
Here’s the uncomfortable truth: recording a call, even with a friendly little icon and a summary at the end, makes people more careful and less honest. Not less honest as in lying. Less honest as in holding back the thing that matters.
Sales trainers push AI notetakers as a way to catch every objection word for word. What they don’t say is that the moment a prospect knows every word is being logged, the objections get softer and vaguer. “I’m not sure this is the right time” instead of “I think your price is too high and I don’t trust the ROI numbers you showed me.” The real objection, the one you needed to hear, doesn’t make it onto the transcript because the prospect never says it out loud in the first place.
I picked up a version of this from the sales thinking in Business Lessons from Alex Hormozi, where the whole point of a good discovery call is getting to the real, uncomfortable objection fast. A visible recording bot works against that. People perform for the transcript. And a transcript full of polite hedging is worse than no transcript at all, because it looks like data. It feels rigorous. It isn’t.
There’s a second, quieter issue. Most of these tools’ free and mid tier plans reserve the right to use call content to train or improve their models unless you dig into settings and switch it off. If you’re a consultant, coach, lawyer or accountant, that means client conversations, sometimes covered by an NDA you signed personally, are sitting on a third party’s servers with terms you almost certainly didn’t read line by line. I hadn’t, until a client’s compliance officer asked me to send our data processing agreement with Fireflies. I didn’t have one. I had a free account and a login.
When a notetaker helps and when it quietly costs you the deal
I’m not against these tools. I use one most weeks. But I’ve learned where they help and where they hurt.
They help on internal team calls, project reviews, and any meeting where the goal is a record of decisions and action points. Nobody’s holding back on a Tuesday standup because a bot is listening. They help enormously for anyone with a hearing difficulty or anyone working across a language barrier, where an accurate transcript is a genuine accessibility win, not a nice extra.
They hurt on first discovery calls, pricing negotiations, and anything HR-adjacent: performance conversations, exit interviews, redundancy discussions. They hurt on coaching calls where the whole value is the client saying the thing they haven’t said to anyone else yet. And they hurt more than people admit on client calls generally, because the client rarely feels they can say no once the bot is already in the room.
The right way to bring one into a call
After the Mark call, I built a small policy and I’ve stuck to it for three months. It’s not complicated.
- Name the tool in the calendar invite. “This call will be recorded and transcribed by Fireflies for note-taking purposes” written plainly, not buried in a meeting link footer.
- Ask out loud in the first thirty seconds. “I’ve got a notetaker running so I can focus on the conversation instead of typing, are you happy with that or would you rather I switch it off.” Nine times out of ten people say yes. The one time out of ten matters more than the note-taking.
- Check the data setting before the call, not after. Turn off any option that lets the vendor use the transcript for model training. It takes ninety seconds in the account settings and most people never do it.
- Delete or redact anything sensitive from the transcript before it goes anywhere, especially before it lands in a CRM that three other people on your team can open.
- Turn the bot off entirely for anything involving numbers the client hasn’t agreed to share widely, or anything HR related. Some conversations deserve nothing more permanent than your own memory and a handwritten note.
None of this is bureaucratic. It’s five minutes of thinking that most small businesses skip because the tool installed itself into the workflow before anyone asked whether it should be there.
Turning the transcript into something worth having
Where these tools earn their keep is after the call, not during it. A good summary saves you the twenty minutes you’d otherwise spend writing up notes, and it gives you a first draft of the follow up email while the conversation is still fresh. I run that draft through a quick tidy up and send it within the hour rather than the next morning, which on its own has improved my close rate on proposals more than any AI feature has.
If you’re building that follow up into a proper nurture sequence rather than a one off email, the workflow discipline I wrote about in the Mailchimp Marketing Strategy piece applies directly here: the AI can draft the first message, but the sequence around it, the timing, the segmentation, the actual point of each email, still needs a human deciding what matters.
Trust is the actual product in most service businesses, long before the invoice is. Airbnb built an entire brand on convincing strangers to trust each other with house keys, which I broke down in the Airbnb Marketing Strategy piece, and the lesson transfers: the moment people feel watched instead of trusted, the relationship changes shape even if nothing bad happens with the data.
Working from home makes this harder to spot
If you work from home and live on video calls the way most consultants and freelancers now do, this creeps up on you faster than in an office. There’s no colleague glancing over to ask “is that thing recording us.” It’s just you, your laptop, and a tool your calendar app quietly connected months ago. I only noticed how many of my calls had a bot in them when I sat down and counted: over one month, 34 external calls, 19 of them had a notetaker join automatically because it was set to auto-join anything with two or more participants. I hadn’t chosen that setting on 12 of them. My calendar tool had.
Before you decide whether the convenience is worth it, do the sum rather than guess. I wrote before about how interactive calculators pull people in and keep them thinking because the numbers are right there instead of hidden. Do the same exercise on your own calendar: how many minutes does the tool save you a week, versus how many client relationships depend on people feeling they can speak freely in front of you. For most solo consultants the honest number comes out lower than the marketing for these tools suggests.
If you want someone to sit down with your actual tool stack, work out which calls should have an AI notetaker and which shouldn’t, and write you a one page policy your team will follow, that’s exactly the kind of practical audit I do through AI implementation coaching. It’s a two hour conversation, not a six month project, and most small businesses need the former far more than the latter.
What I’ve settled on
I still use a notetaker on most calls. I announce it every single time, I’ve turned off the model training setting on every tool I use, and I switch it off completely for anything that smells like a hard conversation. That’s the whole policy. It cost me nothing to write and it’s already stopped one client asking, halfway through a call, who else was listening.
Frequently asked questions
Is it legal to record a client call with an AI notetaker without telling them?
In the UK, under UK GDPR, recording a call without informing participants can breach data protection rules because you’re processing personal data without a lawful basis or transparency. In most US states you need at least one party’s consent, and some states require everyone’s consent. The safe rule everywhere: say it out loud at the start of the call, every time.
Which AI notetaker is best for a small business on a tight budget?
Fathom’s free plan is the strongest free option right now, with unsolicited summaries and CRM sync. Otter’s free tier works fine for occasional users but caps out quickly at 300 minutes a month. If cost is the only factor, start free and only pay once you can point to a specific feature you’re missing.
Should I let an AI notetaker join sales discovery calls?
Be cautious. Discovery calls work best when the prospect says the real, uncomfortable objection out loud, and people soften their language when they know they’re being recorded. If you use one, ask permission clearly in the first thirty seconds and watch whether the tone of the conversation changes once you do.
What happens to my transcripts after the call ends?
It depends on the tool and your account settings, not just the tool’s marketing page. Most free and many paid tiers reserve the right to use call content to improve their models unless you switch that off manually in settings. Check this before your first call, not after a client asks you for a data processing agreement you don’t have.
Related reading: The AI Notetaker Sitting In On Your Client Calls Might Be Breaking Your NDA and AI Meeting Notetakers Are Quietly Costing You Client Trust.
