The short version: AI notetakers like Otter, Fireflies, Fathom, and Zoom AI Companion make people on your calls talk less openly, and most of them use your transcripts to train their models unless you dig into settings and switch it off. That combination is quietly costing small businesses trust, information, and deals. Fix it with a five minute settings check and one honest sentence at the start of every call.
The moment I noticed it
It was a discovery call in March this year with a heating and plumbing business owner in Ashford who wanted help with his marketing. Ten minutes in, I asked what was slowing his sales down. He paused, looked at his screen, and said "can we turn that thing off first." He meant the little bar at the top of Zoom that said Fireflies Notetaker was recording and transcribing. Once it was gone he told me the real problem: three of his engineers were about to leave for a competitor and he hadn't told his business partner yet. That's not information he was ever going to say in front of a bot with a name and a logo, even one he'd invited himself.
I've run hundreds of these calls over the years. That was the moment I clocked that the tool built to capture more information was, in that call and plenty since, the reason I got less of it.
What's happening when that bot joins your call
Every AI notetaker on the market right now, Otter.ai, Fireflies.ai, Fathom, Gong, Zoom's own AI Companion, does three things at once: it records audio and video, it transcribes in near real time, and it generates a summary afterwards that gets emailed round or dropped into your CRM. Useful. I use these tools myself most weeks.
But research on video call behaviour keeps landing on the same finding: people self-censor more when they know a transcript exists, and they self-censor even more when the thing making the transcript is a piece of software rather than a human note-taker they can read the mood of. A human taking notes can be told "off the record" with a look. A bot can't. So the conversation gets safer, blander, and further from the truth than it would be without it. If you're trying to sell, coach, or understand a prospect's problem, that's a real cost, not a small one.
The setting almost nobody checks
Here's the part that gets skipped in most write-ups of these tools. Read the terms of service for the free and mid-tier plans of most AI notetakers and you'll find a clause that lets the company use your call data, transcripts included, to train or improve its AI models unless you explicitly opt out. Otter's consumer and small team plans have carried language like this for years. Fireflies has a similar default on its lower tiers. Even Zoom faced enough backlash in 2023 over training-data language in its terms that it had to publicly clarify and adjust its policy. The default, across the industry, leans toward "we can use this unless you say no," not the other way round.
That matters more than it sounds. If a client tells you their turnover, their staffing problems, a pricing dispute with a supplier, or something personal that came up because the call ran long and got human, that recording may sit on a server being used to make someone else's product smarter. Under UK GDPR, if there's personal data in that call, you need a lawful basis to record and process it, and "the software vendor defaults to yes" isn't one. In the US, eleven states including California require every party's consent before a call can legally be recorded, not just the host's. Most people running these bots have never opened the settings menu to check which side of that line their plan sits on.
Where this costs you money
Three specific ways I've watched this play out with small business owners I work with:
- Slower deals. Prospects give shorter, safer answers, so you need more calls to get to the real objection, which stretches your sales cycle and burns your time.
- Lost referral trust. A client who feels recorded rather than listened to is far less likely to introduce you to their network. Referrals are still, five years into rebuilding my own business in public, the highest converting channel I have, and they run entirely on the client feeling like a person and not a data point.
- Compliance exposure you didn't budget for. A single complaint to the ICO or a state attorney general about an unconsented recording is a far more expensive afternoon than the £8 to £16 a month these tools cost.
None of that means bin the tools. It means treat them like you'd treat any other bit of business infrastructure that touches other people's information, with a policy, not just a login.
My rule now, step by step
This is what I do, on every call, not just the ones where it's obviously sensitive:
- Say it out loud in the first thirty seconds. "I've got an AI notetaker running so I don't have to scribble while we talk, happy to turn it off if you'd rather." Nine times out of ten people say leave it on. The one time in ten matters more than the nine.
- Check the training data setting before the call, not during it. In Otter it's under Settings then Privacy. In Fireflies it's under the workspace admin settings. Fathom is opt-out by default on paid plans, which is one of the reasons I switched a chunk of my calls to it this year. Five minutes, once, per tool.
- Use the pause function for the parts that matter. Every one of these tools has a manual pause button. When a client starts talking about redundancies, a legal dispute, or anything that isn't strictly about the work, I pause it. Takes one click.
- Send the summary before you act on it. AI summaries hallucinate details, misattribute numbers, and occasionally invent commitments nobody made. I send mine to the client with "does this match what we agreed" before it goes anywhere near a proposal or an invoice.
- Delete on a schedule. I clear transcripts older than ninety days unless there's a specific reason to keep them. Most plans let you set this to automatic. Most people never do.
If you want someone to sit down with you and build an actual written AI tools policy for your team in an afternoon rather than working it out call by call the way I did, that's exactly the kind of practical work an AI consultant for small business should be doing with you, not just picking tools.
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.
When I still switch it on without a second thought
Internal team meetings. Webinar recordings where everyone already knows they're on camera. Client onboarding calls where the whole point is a clear written record both sides can refer back to. In those settings the transparency cost is close to zero because everyone already expects to be recorded, and the summary saves an hour of typing. The line I draw is discovery calls, first sales conversations, and anything where I'm asking someone to be honest about a problem before they've decided to trust me. That's where the bot changes the conversation, and where I now default to off unless the prospect says otherwise.
This is the same lesson I keep relearning across every AI tool I've adopted since I started rebuilding this business five years ago: the tool that promises to capture more almost always changes the thing it's capturing. I made a version of this mistake at scale when I tested AI content production and published 569 posts only to watch Google index five percent of them. Same pattern, different tool: adopt fast, check the fine print later, pay for it in results you didn't see coming. The fix both times was the same too, slow down enough to read the settings before you scale the habit.
If you're weighing up which AI tools are worth your time at all before you get anywhere near a notetaker, I've written honestly about how I use AI day to day without being technical, and there's a shorter, odder list of 50 useful websites most business owners have never heard of that includes a couple of privacy checking tools worth five minutes of your time.
What rebuilt my business after five hard years wasn't a tool, it was trust showing up in the numbers, which I wrote about honestly in the month my website started paying me again. An AI notetaker that quietly erodes trust on the exact calls where you need it most works directly against that, no matter how good the summary looks in your inbox afterwards.
Frequently asked questions
Do I legally need consent to use an AI notetaker on a business call?
In the UK, if the call involves anyone's personal data, you need a lawful basis under GDPR, and relying on a vendor's default settings isn't one, so tell people it's running and let them object. In the US, eleven states including California, Illinois, and Florida are two-party consent states, meaning every person on the call must agree before it's recorded, not just the person who set the meeting up.
Which AI notetaker doesn't use my data to train its model?
Fathom's paid Team plan is opt-out of training data by default as of 2026, which is one reason I moved a chunk of my calls there. Otter and Fireflies both let you turn training data use off, but on their free and lower-cost individual plans you have to find the setting yourself, it isn't off by default. Always check the current terms before you assume, vendors change these policies more often than most people notice.
Will telling clients I'm recording the call make them more guarded?
Slightly, for the first minute. But the alternative, them finding out later or noticing the bot mid-call and wondering what else you haven't mentioned, costs you far more trust than a straight sentence up front ever will. Say it, offer to turn it off, then get on with the conversation.
Is it worth using an AI notetaker at all if it makes people more guarded?
Yes for internal meetings, onboarding calls, and anything where a written record helps both sides. No, or at least pause it, for early sales conversations and any moment where you're asking someone to be honest about a problem before they trust you. The tool isn't the mistake, using it the same way on every single call is.
Related reading: Why I Turned Off My AI Meeting Notetaker After 14 Months (And What It Was Really Costing Me) and I Tested an AI Phone Receptionist for Three Months on a Real Small Business.
