The short version: AI notetakers like Otter, Fireflies and Fathom are now joining a huge share of business calls without anyone thinking twice about what happens to that recording afterwards. Most small business owners have no consent process, no idea where the transcript is stored, and no clue it might be used to train a model or turn up in a legal dispute later. Fix the process before someone asks you about it, because eventually someone will.
The morning a client's bot outed mine
I was on a discovery call with a prospective client back in early 2026, a founder of a small logistics company in Manchester, and about four minutes in her screen showed a little box pop up: "Fathom Notetaker has joined the meeting." She hadn't told me it would be there. I hadn't told her mine was recording either, because Zoom's own AI Companion had quietly switched itself on after an update.
So there we were, two AI notetakers sat in on a call about trust and consultancy, neither of us having agreed to it out loud. She laughed. I didn't, because I'd just realised how often this must be happening on calls where nobody laughs at all.
Why every other call has one now
This isn't a fringe habit anymore. Zoom built its AI Companion into every paid tier by default. Otter.ai passed 25 million users by the end of 2025. Fireflies, Fathom, Read AI and Gong have all pushed hard into the small business market, some with free tiers that only need a calendar connection to start joining every call automatically, whether you remember inviting them or not.
The pitch is fair: nobody wants to type notes while trying to listen. I use one myself, most weeks, because chasing a client's action points from memory is a rubbish use of anyone's time. The problem isn't the tool. It's that almost nobody using it has thought past "this saves me twenty minutes."
The bit nobody's solved: consent
Here's the uncomfortable part. Under UK GDPR, recording a call and processing that recording (which is exactly what an AI transcript is) needs a lawful basis, and for most small business calls that means clear consent, given before the recording starts, not buried in a calendar invite footer nobody reads. The ICO has been explicit about this since well before AI notetakers existed, and the arrival of AI hasn't changed the law, it's just made breaking it far more common because the bot joins by default now rather than by choice.
Most small businesses I talk to when I'm doing AI consultant for small business work have never once said the words "this call is being recorded and transcribed by an AI tool, is that alright with you" out loud. Not once. The bot just joins, quietly, the same way it joined mine and my Manchester client's call, and everyone gets on with the meeting as if that's normal, because it's become normal, which is exactly the problem.
The fix costs nothing and takes ten seconds: say it at the start of the call. "Just so you know, I've got an AI notetaker running so I can focus on the conversation rather than typing, are you happy with that?" I've never once had anyone say no. I have, twice, had people say ", can we turn it off for this bit" when something sensitive came up, and being asked meant they trusted me more by the end of the call, not less.
What happens to the transcript afterwards
This is the part most articles about AI notetakers skip entirely, and it's the part that should worry you more than the recording itself. That transcript doesn't just sit in your account. Depending on the tool and the plan you're on:
- It may be used to train the company's models unless you've actively opted out in the settings, something buried three menus deep on most free and lower tier plans
- It's searchable by anyone else on your team who has access to the workspace, including old contractors you forgot to remove
- It's stored indefinitely by default on most tools, meaning a throwaway comment a client made eighteen months ago about a competitor, or a pricing figure they asked you to keep quiet, is still sitting there
- It can be pulled into discovery if a client dispute ever ends up anywhere near a solicitor's letter, and transcripts read very differently out of context than conversations do
I had a client last year whose bookkeeper mentioned, off the cuff on a recorded call, a cash flow problem that had nothing to do with our project. Six months later that transcript got pulled up during an unrelated dispute with a supplier, because it existed, was searchable, and nobody had ever deleted it. Nothing came of it in the end, but it was a uncomfortable few weeks watching a throwaway sentence turn into evidence.
Where this quietly connects to your search visibility too
There's a second layer to this that almost nobody in small business marketing has clocked yet. Some AI notetaker platforms are starting to feed anonymised or aggregated call data into the same large language models that now answer questions in ChatGPT and Perplexity, the same models you're trying to get cited by if you've ever run something like the 15 minute AEO audit on your own website. Your call content and your website content are both, increasingly, raw material for the same systems. Worth knowing before you say something on a client call you'd never write in a blog post.
The rules I use now
After the Manchester call, and after the bookkeeper story, I changed how I run this. It's not complicated, and it's not expensive, it just needs saying:
- Announce it every time. Ten seconds at the start of the call, every call, no exceptions, even with clients I've worked with for years.
- Check the training data setting. Every tool I've named here (Otter, Fireflies, Fathom, Zoom's AI Companion) has a toggle to stop your transcripts training their model. It's off by default on nearly every plan I've checked. Turn it off.
- Set a deletion schedule. I delete transcripts older than 90 days unless there's a specific reason to keep one. Most tools let you automate this. Most people never touch the setting.
- Audit who has workspace access quarterly. Remove former staff and contractors the same day they leave, not "at some point."
- Never let the bot into calls involving anything sensitive, redundancy conversations, pricing negotiations you don't want written down verbatim, anything a client explicitly asks to keep off the record. Turn it off, take your own scrappy notes instead.
None of this takes long once it's a habit. Working out whether the time saved by the notetaker is worth the small admin of managing it is the kind of thing worth sitting down and doing the maths on, the same way I'd tell anyone to run the numbers before adding any new tool, using something as simple as the interactive calculator approach I use with clients to work out real return on time for any piece of software, not just the flashy AI ones.
What this says about trust, not just tools
The businesses that handle this well aren't the ones with the fanciest privacy policy page, they're the ones who are simply upfront about what they're doing and why, the same instinct that's built genuine trust for brands doing this at scale, the way I've written before about how Miro built a brand people trust largely through transparency about how their product works rather than glossing over the mechanics. Small business owners can borrow that instinct for free. Say what the bot is, say what happens to the recording, say it every single time. It costs you nothing and it's the single biggest difference between a client who feels informed and one who feels surveilled.
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.
Brands that get this instinct right early, the way Red Bull built consistency into everything it does, tend to build the kind of trust that compounds. A ten second sentence at the start of a call is your version of that same consistency.
If you're not sure whether your current setup is compliant, or you want someone to sit down and map out where your client data lives, an hour with an AI implementation coach is far cheaper than finding out the hard way after a client asks a pointed question about where their transcript went.
Frequently asked questions
Do I legally need consent to use an AI notetaker on a client call in the UK?
Yes. Recording and transcribing a call counts as processing personal data under UK GDPR, and the ICO expects clear consent given before recording starts, not buried in a calendar invite. Saying it out loud at the start of the call is enough for most small business situations, and it takes about ten seconds.
Does Otter, Fireflies or Zoom's AI Companion use my call transcripts to train their models?
On most free and lower tier plans, yes, by default, unless you go into settings and switch it off. Paid business tiers on some platforms exclude training data automatically, but this varies by tool and changes often enough that it's worth checking your account settings every few months rather than assuming.
How long do AI meeting platforms keep my transcripts?
Most keep everything indefinitely unless you set a deletion schedule yourself. Very few small business owners ever do this. Setting transcripts to auto delete after 90 days is a sensible default for most client work, longer only if there's a specific reason to keep something.
Is it worth using an AI notetaker at all given these risks?
Yes, the time saved on writing up notes and chasing action points is worth it for most small businesses. The risk isn't the tool, it's using it without a consent habit, a training data check, and a deletion schedule. Ten minutes setting those three things up removes almost all of the downside.
Related reading: The AI Note-Taker Nobody Invited: What It's Doing With Your Client Calls and The Best AI Productivity Tools for Entrepreneurs in 2026 (What I Use, Daily).
For the bigger picture, see my full guide to AI marketing.
Useful references
Related: ai notetaker client calls nda risk.
There is a Liverpool-specific version: an AI consultant covering Liverpool.
