The short version: AI notetakers like Otter, Fireflies, Fathom and Read.ai are now sitting in on client calls for millions of small businesses, and most owners have no idea what happens to that recording after the call ends. Some tools use your transcripts to train their models unless you opt out. Some auto-share full transcripts, including the awkward bits, with everyone on the calendar invite. You need a process, not just a plug-in.
The call that made me switch my notetaker off for good
Two years ago I had a discovery call with a prospective client, a founder of a small e-commerce brand who was lovely, sharp, and clearly stressed. Fireflies was sitting in, quietly transcribing, because that’s what I did on every call by then. Halfway through, off the back of a question about her marketing budget, she said something like “honestly I don’t think we’re going to survive the next quarter unless this works.” Not for the record. Just a human being being honest with another human being.
The call ended. Fireflies did its thing, generated a tidy summary, and because the settings were on default, it emailed that summary automatically to every attendee on the calendar invite, including her business partner, who hadn’t been on the call and had no idea she’d said that out loud to a stranger. She rang me, upset, asking why her private comment had landed in her co-founder’s inbox with a timestamp.
Nothing malicious happened. It was a default setting I hadn’t checked. But it taught me something that most posts about AI meeting tools skip over entirely: the danger isn’t the AI making things up, it’s the AI faithfully recording things people didn’t mean to say for the record, and then distributing them further and faster than a human notetaker ever would.
What these tools are doing behind the scenes
Most AI notetakers work the same basic way. You connect your calendar, the bot joins the call as a visible or invisible participant, it records audio, transcribes it, then runs the transcript through a language model to produce a summary, action items, and sometimes a “sentiment” reading. Here’s what varies wildly between tools, and what most small business owners never check:
- Who gets the summary automatically. Fireflies, Otter and Fathom all default to sharing with meeting attendees in some plan tiers. That means a note meant for your team can end up in a client’s inbox unless you change the sharing settings.
- Whether your data trains the model. Read.ai’s and several tools’ terms of service allow use of anonymised or aggregated data to improve their product unless you actively opt out in account settings. Enterprise tiers usually block this by default; free and personal tiers often don’t.
- How long it’s kept. Otter’s free plan keeps transcripts indefinitely unless you delete them manually. Fireflies gives you retention controls but only on paid plans.
- Whether it announces itself. Some bots join with a visible name and icon. Others can be set to join silently or with a generic name, which is where the real trouble starts.
The consent problem nobody flags in the sales page
If you’re recording calls with UK clients, you’re processing personal data under the same rules that apply to any other recording, and that includes the fact that the person on the other end usually assumes they’re talking to a human notetaker or nobody at all, not that their words are being fed through a third-party AI company’s servers. The general rule under GDPR is that people need to know their data is being collected and roughly what for. A quick line at the start of the call, “just so you know, I’ve got an AI notetaker running, is that alright with you,” takes four seconds and solves most of the legal and trust problem in one go. Most small business owners skip it because it feels awkward. It’s less awkward than the conversation you’ll have if a client finds out afterwards.
If you want the fuller detail on what counts as informed consent for recorded data, the underlying rules sit inside the General Data Protection Regulation, and it applies whether the recording is a cassette tape from 1995 or a cloud-based AI transcript in 2026.
The uncomfortable bit: it’s making you worse at your job, not better
Here’s the part most write-ups about AI notetakers avoid saying outright. Since I started relying on a bot to catch everything, I listen less carefully. I used to take my own scrappy notes, which forced me to decide in real time what mattered. Now I half-listen because I know the transcript will catch it, and I’ve noticed my follow-up questions on calls have gotten lazier. Clients notice this too, even if they can’t name it. There’s a difference between someone who’s present and someone who’s outsourced their attention to a bot and is quietly checking Slack.
The fix isn’t to ditch the tool. It’s to use it as a backup, not a substitute for listening, and to still write three or four notes of your own during the call on what mattered to the person in front of you. The AI catches the facts. You need to catch the feeling, and no summary tool does that reliably yet.
Step by step: setting these tools up
- Say it out loud at the start of every external call. “I’ve got an AI notetaker on, that alright?” Ninety-nine times out of a hundred, people say yes and forget about it in ten seconds.
- Turn off auto-sharing to all attendees. On Fireflies this is under Settings then Integrations then Notifications. On Otter it’s in Workspace Settings. Do this once, and check it again after any plan upgrade, because settings sometimes reset.
- Opt out of model training if the option exists. Look for a line in account settings, usually under Privacy or Data, that says something like “use my data to improve the product.” Switch it off unless you actively want your client conversations helping train someone else’s AI.
- Set a retention limit. Thirty to ninety days is plenty for most client work. Delete anything older automatically rather than letting years of transcripts pile up as a liability sitting in someone else’s cloud.
- Use a separate account, or no bot at all, for sensitive calls. Legal conversations, HR issues, anything involving a client’s financial trouble or a difficult personal situation. Some things should stay off the record entirely.
- Check the export options before you need them. If a client ever asks you to delete their data, you need to know exactly where the transcript, the summary, and any CRM sync copy of it are sitting, and how fast you can remove all three.
What this costs
Otter’s Pro plan is around $16.99 a month per user with a decent monthly transcription cap. Fireflies Pro runs about $10 a month per user and includes better sharing controls than the free tier. Fathom is free for most individual use and makes its money on team plans. Grain sits closer to $19 a month and leans more into sales team analytics than general note-taking. None of these numbers are the point. The point is that the free and cheap tiers are almost always the ones with the loosest data controls, because tight retention and training opt-outs are usually gated behind the paid plans. If you’re on a free plan purely to save ten dollars a month, you’re often the one paying with your clients’ data instead.
Trust is the actual product here
Brands that get remembered for the right reasons tend to be the ones that treat customer trust as the whole business, not a compliance checkbox. Mailchimp built an entire brand identity around making email permission and consent feel simple rather than sinister, which is worth studying if you want to see how a marketing strategy can turn a legal requirement into a point of difference instead of a chore. GoPro did something similar with user content, always making clear what they were using and why, which is a big part of how they built a brand people trusted enough to hand over their own footage. The lesson carries over to a five-second line at the start of a Zoom call. Tell people what’s happening with their data and most of them will happily go along with it. Hide it and you’ve made a small trust problem into a much bigger one the day it surfaces.
Joe Pulizzi spent years arguing that content businesses live or die on trust built slowly over time, and that one bad move can undo years of it faster than any good move can rebuild it. His business lessons apply just as well to an AI transcript ending up somewhere it shouldn’t as they do to a badly timed newsletter.
When to just turn the bot off
Not every call needs an AI listening in. If a client is going through something personal, if the conversation touches on redundancies, health, family, money trouble, or anything they might regret having on record, turn the notetaker off and take your own scrappy notes instead. You lose the perfect transcript. You keep the trust, and in a small business built on referrals and repeat clients, trust is worth considerably more than a tidy summary email.
If you’re not sure whether your current AI setup across notetakers, CRM, and email tools is creating this kind of exposure without you realising it, that’s exactly the kind of audit an AI consultant for small business should be doing for you in the first session, not the tenth.
Tools like Figma built collaborative software on the premise that people could see exactly who changed what and when, which is worth remembering when you’re choosing AI tools for your own business, because the brands that win long term tend to be transparent about data by design, not as an afterthought bolted on after a complaint.
Free resource: The Sales Call Discovery Question Cheat Sheet.
Free resource: The Discovery Call Prompt Pack.
Frequently asked questions
Do I legally have to tell clients an AI notetaker is on the call?
In the UK, GDPR requires that people know their personal data is being collected and roughly why, and a recorded, transcribed call counts. A short verbal heads-up at the start of the call satisfies this in practice for most small business situations, and it also protects the relationship if the client later finds out the recording existed.
Does Otter or Fireflies use my call transcripts to train their AI?
It depends on the plan and the settings. Free and lower tiers on several tools allow use of transcript data to improve the product unless you opt out manually in account settings, usually under Privacy or Data. Paid business and enterprise tiers typically block this by default. Always check the current settings yourself rather than assuming, because policies and defaults change.
What’s the biggest risk with AI meeting notetakers for small businesses?
The biggest practical risk isn’t the AI inventing things, it’s accurate transcripts and summaries being auto-shared with people who weren’t meant to see them, or off-the-record comments being captured word for word and distributed further than a human notetaker would ever have gone. Checking your sharing settings solves most of this in under five minutes.
Should I stop using AI notetakers altogether?
No, they’re useful for accurate follow-up and action items. The fix is to use them deliberately: tell people they’re on, tighten sharing and retention settings, opt out of model training where possible, and switch the bot off entirely for sensitive or personal conversations rather than letting it run by default on every call.
If you want the full breakdown, here is everything I know about AI marketing.
Related reading: Why Your AI Meeting Notetaker Might Be Breaking the Law (And Killing Your Sales Calls) and Why I Turned Off AI Meeting Notes On My Client Calls (And What I Do Now).
