Bottom line: AI meeting notetakers like Otter, Fireflies and Fathom are brilliant for saving admin time, but they change how people talk in a room, and most small businesses are using them in a way that quietly damages client trust and makes their own team worse at listening. Fix it with three rules, not by ditching the tool.
The moment I noticed something was off
About eight months ago I started tracking, informally, how many new client calls opened with some version of “is this being recorded?” Out of roughly 40 discovery calls since March, I counted 14 where the client asked before we’d even said hello. Not annoyed, just checking. But the tone of the call changed after that question every single time. People chose their words more carefully. They stopped saying “honestly, our last agency was rubbish” and started saying “we’ve had mixed experiences previously.” Same meaning, completely different level of candour.
That’s the bit nobody selling you an AI notetaker tells you. The tool doesn’t just capture the conversation, it changes the conversation. The second someone knows a transcript is being generated and possibly fed into a CRM or shared with a manager, they perform slightly. And you lose the messy, honest, useful part of the meeting where people tell you what’s wrong.
Why small businesses reach for these tools in the first place
I get the appeal. If you’re running a five person consultancy or a one person coaching business, you cannot afford to have someone typing notes badly while trying to listen. Otter.ai’s free tier gives you 300 minutes a month of transcription, Fireflies.ai bundles in automatic CRM logging, and Fathom has become the default for a lot of sales teams because it plugs straight into Zoom and gives you a clip reel of the “important bits” afterwards. Used well, these save a genuine chunk of admin time. I’d estimate they cut post-meeting write-up time by 20 to 30 minutes per call for most consultants I work with.
The problem isn’t the tool. It’s that most people switch it on, forget it’s running, and then treat the AI summary as a substitute for being present. I’ve watched it happen in my own team. Someone joins a call, the notetaker is on, and halfway through you can see them mentally checking out because “it’s all being recorded anyway.” Then they read the summary afterwards instead of remembering the conversation, and summaries flatten nuance. A client saying “we might have budget in Q2, depends on a few things” becomes a bullet point that reads “Budget confirmed for Q2.” That single flattening has cost me a follow-up email sent too confidently, once, and it was awkward to walk back.
The uncomfortable bit about client trust
Here’s the part that sits uneasily with a lot of the “AI productivity” advice floating around. Recording a call, even with consent, changes the power dynamic. Your client is agreeing to something they didn’t ask for and often can’t fully evaluate in the moment, because they’re mid-conversation and don’t want to make a fuss. In the UK, under GDPR, you technically need a lawful basis and clear notice before recording someone, and “everyone does it now” is not a legal basis. Most small businesses I’ve spoken to haven’t checked this. They’ve just turned the bot on because it came free with their Zoom plan.
I’m not saying stop recording. I’m saying stop pretending it’s a neutral, invisible act. It isn’t. It’s the same trust calculation as Airbnb had to solve when they convinced strangers to sleep in each other’s houses, worth reading their approach to building trust at scale if you want to see how a much bigger company handled the same problem: transparency first, control second, and never assume consent just because nobody objected out loud.
What I do now, step by step
After that run of 40 calls, I changed my process. This is what runs on every client call now:
- Ask before you join, not after you’re in the room. I email a line ahead of first calls: “I use an AI notetaker to capture action points, happy to switch it off if you’d rather I take notes manually.” Giving people the choice before the call starts, when there’s no social pressure, gets an honest answer.
- Say it out loud anyway. Even after the email, I say at the start of the call “just flagging Fathom is recording this for my notes.” Redundant, yes. It also resets the tone and people relax faster once it’s named rather than assumed.
- Turn it off for the sensitive ten minutes. If a client starts talking about a partnership falling apart or staff problems, I pause the bot. Manually. It takes three seconds and it tells the client, without saying it, that this part of the conversation is between us.
- Read the transcript, don’t just skim the summary. The AI summary misses tone every time. I read the actual transcript for anything I’m about to act on, particularly pricing conversations, because “might” and “will” get treated the same way by most summarisation models.
- Delete on a schedule. I keep transcripts for 90 days then bin them. Clients ask about this more than you’d think, and having an answer ready, rather than a shrug, changes how the whole tool is perceived.
None of that took long to set up. The whole change was maybe two hours of thinking and a template email. The return has been fewer awkward “wait, who has access to that transcript” conversations and, I’d guess, more honest early calls, though that one’s harder to put a number on.
The bit that costs you money: your team stops listening
This is the part I think gets skipped in almost every “AI notetaker productivity hack” post. The bigger cost isn’t client trust, it’s your own team’s listening muscle. I noticed one of my associates, sharp, good with clients normally, started asking questions in follow-up calls that had already been answered two calls earlier, because he’d stopped retaining anything live and was relying entirely on the searchable transcript archive. Technically he could find the answer in 30 seconds. Practically, the client noticed he’d asked the same question twice and it read as not paying attention, because it wasn’t.
We fixed it with a small rule: no laptop open during discovery calls unless you’re the one sharing a screen. Notetaker still runs, but nobody’s typing, nobody’s half-reading the live transcript while the client talks. It’s a bit like the discipline Joe Pulizzi talks about in his own content work, doing fewer things with full attention beats doing everything with half attention, and there are some useful business lessons from Joe Pulizzi on exactly that trade-off if you want the fuller argument.
Where this fits with your wider marketing
If you’re using AI transcripts to feed content, and a lot of small businesses now mine sales calls for blog topics and objection-handling copy, be careful what you lift directly. A client’s specific complaint about a competitor, said honestly on a recorded call, is not the same as a testimonial you can publish. I’ve seen a version of this go wrong where a punchy quote from a discovery call ended up almost word for word in a case study, and the client noticed. If you’re building your content engine off real conversations, the same discipline that goes into good Mailchimp style email marketing, permission-based, opted in, clearly labelled, should apply to anything sourced from a recorded call. Ask before you quote, every time, even internally.
There’s also a UX lesson buried in here that most people miss. Companies like Figma built their whole reputation on making collaboration feel visible and consensual rather than hidden, worth a look at their approach to building a brand people trust with shared work because the same principle applies to a recorded call: show people what’s happening to their input, don’t just capture it silently and process it later.
What good practice looks like on paper
If you want a simple policy for your business rather than ad hoc decisions on every call, write down four things and stick them somewhere your team can find them: who gets notified before recording starts, how long transcripts are stored, who has access to them inside your business, and what happens if a client asks for theirs deleted. Takes about 20 minutes to draft. I did mine on a Sunday afternoon with a cup of tea and it’s saved several “hang on, can you send me that transcript” scrambles since.
If your business runs a lot of client calls and you’re not sure how to set any of this up without it becoming a whole IT project, this is exactly the kind of practical fix a decent AI consultant for small business can sort in an afternoon rather than a project you keep putting off. It’s not complicated, it just needs someone to decide the rules once.
None of this means stop using AI notetakers. I still use one on every call. It’s just no longer running silently in the background while I pretend nothing’s changed. It’s a visible part of the meeting now, and that one shift, saying it out loud, pausing it when it matters, reading the transcript instead of trusting the summary, has done more for client trust than any onboarding email I’ve ever written.
Frequently asked questions
Do I legally have to tell clients I’m recording a call with an AI notetaker?
In the UK, yes, under GDPR you need a clear lawful basis and you should give people notice before recording, not after. Most one way notification banners inside Zoom don’t count as meaningful consent if the client hasn’t seen or acknowledged them, so say it out loud at the start of the call as well.
Which AI notetaker is best for a small consultancy or agency?
Otter.ai works well for solo consultants who mainly need transcription and a free tier. Fireflies.ai is stronger if you want automatic CRM logging across a small team. Fathom is the one most sales-heavy teams end up on because it integrates cleanly with Zoom and pulls out action items without much manual tidying.
Does recording client calls hurt sales conversion?
I don’t have a clean study to point to, but in my own tracking, clients who asked whether a call was being recorded gave noticeably shorter, more guarded answers to open questions for the rest of that call. That’s not proof it hurts conversion, but it does mean you’re getting less honest information to work with, which shows up later as missed objections you didn’t hear the first time.
Should I turn the AI notetaker off for sensitive conversations?
Yes. If a client starts discussing something personal, a staffing problem, a partnership dispute, a budget they’re not meant to be sharing, pause the recording manually and say you’re doing it. It costs three seconds and it signals respect that a blanket “always recording” policy never will.
Related reading: The AI Notetaker Sitting In On Your Client Calls Might Be Breaking Your NDA and AI Meeting Notes Are Quietly Making Your Team Forget How To Think.
Want the complete version? Read where I break down AI marketing.
This overlaps with my main post on the topic: ai notetakers client trust problem.
