AI Notetakers Are Quietly Damaging Client Trust (What I Do Instead)

Straight answer: AI notetakers like Otter.ai, Fireflies and Zoom's AI Companion save you time, but they change how honest people are on a call, and most small business owners never ask permission before hitting record. The fix isn't ditching the tool, it's a three-line policy you say out loud before every call, plus knowing which conversations should never be transcribed at all.

The call that made me change my policy

Eighteen months ago I was on a discovery call with a financial services client, call her Rachel, based in Leeds. Fireflies joined automatically, the way it does when it's synced to your calendar. Twelve minutes in she stopped mid-sentence and said, "Sorry, who's Fireflies?" I explained it was an AI notetaker, purely for my own reference. She went quiet, then chose her words for the rest of the call the way you choose words when you know a lawyer might read them back to you one day.

We got the contract eventually. But the call was worse. Shorter answers, no tangents, none of the loose talk where clients usually tell you the real problem instead of the polished version. I lost the good bit of the conversation because a bot was in the room and she knew it.

That's the part nobody selling these tools tells you. The moment someone knows they're being transcribed, they stop talking like a person and start talking like a witness.

What's really happening when the AI joins the call

Otter.ai, Fireflies, Fathom, Grain, Zoom AI Companion, Microsoft Copilot in Teams, they all do roughly the same thing: join as a visible or semi-visible participant, transcribe in real time, then generate a summary with action items. Pricing is not nothing either. Otter's paid tier is around 16.99 dollars a month per user, Fireflies runs about 10 to 19 dollars depending on the plan, and Zoom's AI Companion is bundled but only on certain paid tiers. If you've got three people on your team using this on client calls, you're looking at 400 to 700 dollars a year just in subscriptions, before you count the time spent editing summaries that got someone's name wrong.

In the calls I've run over the past year, roughly one in five clients visibly reacts when they notice the notetaker on screen, a pause, a comment, a change in tone. That's not a scientific study, it's just what I've watched happen on my own calendar, but it's consistent enough that I stopped ignoring it.

And the summaries themselves aren't as reliable as the marketing suggests. I've had Fireflies attribute a client's own concern about pricing to me, because the transcript engine mixed up overlapping speech. If that summary gets forwarded to a colleague or a lawyer without anyone checking it against the recording, you've got a real problem, not a small one.

The bit that makes people uncomfortable

Here's the uncomfortable truth most posts on this topic skip past: a transcript is a document, and documents get discovered. If a client dispute ever ends up in front of a solicitor, every AI-generated meeting note you've ever taken about that client is potentially something that can be requested and read. Not just the polished summary, the full transcript, including the bit where you complained about their invoice being late, or joked about a competitor.

I know one consultant, a coach working with SMEs in the Midlands, who had a client relationship go sour over a scope disagreement. Her AI notetaker had transcribed a throwaway comment she'd made about "just getting this one over the line for now." Out of context, in a formal complaint, it looked like she'd admitted the work was rushed. It wasn't, but she spent three weeks and a fair bit of money proving that. She's told me since she now only records calls where the content itself needs a paper trail, and turns it off for everything else.

That's the real lesson. AI transcription isn't a neutral productivity habit. It's a decision to create a permanent, searchable, forwardable record of everything said in that room, including the things you'd normally forget by Friday.

The three-line policy I use now

I still use AI notetakers. I'm not precious about the tech, I use ChatGPT and Claude every day for this business. But I changed how I introduce them, and it's fixed most of the awkwardness:

• Before the call starts, I say: "I use an AI notetaker so I don't miss anything, it's just for my records, are you happy with that?" Out loud, every time, not buried in a calendar invite nobody reads.
• If there's any hesitation at all, I turn it off and take notes myself. I've done this maybe a dozen times this year. Nobody has ever been upset that I asked.
• I never transcribe pricing negotiations, anything involving a complaint, or a first call with a nervous client. Those are the calls where people need to think out loud without it becoming evidence.

Under UK GDPR, this isn't optional politeness either, it's the law. If you're recording and processing someone's voice data, you need a lawful basis and, in most cases, their clear consent, especially if the recording leaves the platform and sits in a third party AI tool's servers. The ICO has guidance on recording calls and meetings, and "the AI joined automatically, I didn't think about it" is not a defence that holds up.

What I do with the transcripts afterwards

Where these tools earn their keep is after the call, not during it. I run the transcript through ChatGPT with a simple prompt: pull out every commitment I made, every date mentioned, and any objection the client raised that I didn't fully answer. That single prompt has saved me from at least four missed follow-ups this year that would have quietly cost me a renewal.

I also mine transcripts for content, carefully, and only after checking with the client. A throwaway question from a call about email marketing ROI became the seed for a piece I built around an interactive calculator that let visitors estimate their own return, because I noticed on three separate calls that clients wanted a number, not a paragraph. That's the actual value of these transcripts, spotting the same question coming up again and again across calls you'd otherwise forget individually.

If you're already collecting good material from client conversations, it's worth reading up on why content curation matters for a small business, because half the content you need already exists in your call history, you just haven't organised it.

When AI notetakers help

I don't want this to read as anti-AI, because it isn't. There are calls where a notetaker is the right call, no pun intended:

• Internal team meetings, where everyone already knows and expects it.
• Long onboarding calls with new clients where you're covering ten different process details and nobody, including you, will remember them all.
• Training sessions or workshops you're delivering, where the transcript becomes a resource you can hand out afterwards.

Where it goes wrong is treating every call the same way, flicking the tool on by default because it's synced to your calendar, and never asking whether this specific conversation needs a permanent record at all.

If you're rebuilding how AI fits into your business processes more broadly, not just call notes but the whole client workflow, that's the kind of thing worth getting outside eyes on rather than guessing. It's exactly what I help small businesses work through in AI implementation coaching, because most of the damage isn't from using AI, it's from using it on autopilot without a policy behind it.

A quick reflection before you decide your own policy

Before you set a blanket rule for your business, it's worth being honest with yourself about why you're recording in the first place. Is it because you forget details, or because you don't trust yourself to remember without a safety net, or because you've never questioned the default settings on your calendar? I've used a version of this same self-check with ChatGPT prompts for self-discovery when I was deciding which habits in my business were serving me versus which ones I'd just inherited from a tool's default settings. The answer usually tells you exactly which calls to record and which ones to leave alone.

And if you're building any of this into your marketing content afterwards, worth knowing that AI search tools like ChatGPT and Perplexity are increasingly the place people find advice like this, which is a separate but related reason to check whether AI search is even finding your content in the first place.

For the rest of the series, see the AI for Small Business: 25 Plain-English Guides to Automation, Marketing, Sales and Admin.

Related: ai consultant for hr teams onboarding policy questions.

Related: ai consultant contract what should be in it.

Frequently asked questions

Do I legally have to tell clients an AI notetaker is on the call?

Yes, in the UK you need a lawful basis to record and process someone's voice, and in practice that means clear, verbal consent before the call starts, not a line buried in a calendar invite. The ICO treats this the same as any other recorded call.

Which AI notetaker is best for a small business?

Otter.ai and Fireflies are the two most widely used for solo consultants and small teams, priced roughly 10 to 19 dollars a month per user. Fathom is worth a look if you want something that integrates directly into Zoom without extra logins. None of them are meaningfully more accurate than the others once you're dealing with overlapping speech or strong accents.

Should I turn off AI notetakers for sensitive client calls?

Yes. Pricing negotiations, complaints, and first calls with anxious or unsure clients are exactly the conversations where a transcript can hurt you later if it's ever read out of context. Take manual notes instead and let the conversation stay unrecorded.

Can AI transcripts be used against me in a dispute?

Yes, a transcript is a document like any other, and if a client relationship ends up in a formal complaint or legal dispute, anything you've transcribed and stored can potentially be requested and read back, including offhand comments that meant nothing at the time.

Where to check the details

• Anthropic's own documentation
• the NIST AI Risk Management Framework