AI Meeting Notetakers: The Client Call That Changed How I Use Mine

The short version: AI meeting notetakers like Otter, Fireflies and Fathom save real hours, but the transcript often becomes the client's record of what was agreed, not your memory of it, and that's where small businesses get burned. Tell people you're recording, check the summary before it goes anywhere, and switch the bot off for anything involving money, redundancy or a first-time discount request.

The call that made me stop trusting my notetaker

Last spring I was on a Zoom call with a client I'd worked with for about a year, a retainer worth £3,500 a month. Fireflies was running in the background, as it does on most of my client calls. We were talking about renewing the contract and I said something like "if you sign before the end of the month I can look at doing something on price, maybe a discount." That was it. Vague, deliberately so, because I hadn't decided what I could offer.

The AI summary that landed in both our inboxes twenty minutes later said: "Lilach agreed to a 20% discount on the retainer if signed by Friday." Clean sentence. Confident. Completely wrong.

The client's finance person had it forwarded to her within the hour. By the next morning I was on a call explaining that no, I hadn't agreed to 20% off, that would have been £700 a month, £8,400 over the year, and I'd said "maybe" not "yes." I got there in the end, we settled on a smaller reduction tied to a longer contract, but I lost half a day of goodwill sorting out a mess an algorithm made in one sentence.

That's the bit almost nobody selling you these tools mentions. In my experience the summary gets the shape of a conversation right about nine times out of ten. It's that tenth call where it turns a maybe into a yes, or a rough figure into a committed number, that costs you.

What these tools do well

Before I sound like I'm talking you out of them, I'm not. I still use one on nearly every client call. Here's what they're good for:

• Freeing you up to look at the client instead of typing while they talk, which matters more on video calls than people admit
• Catching action items you'd have forgotten by 4pm, especially on back-to-back days
• Giving you a searchable record when a client says "we never discussed that" six months later and you did
• Producing a rough draft of a follow-up email in seconds instead of fifteen minutes

The current pricing, roughly: Otter's free plan gives you 300 transcription minutes a month, then it's around £17 to £20 a month for the Pro tier per user. Fireflies has a free tier with limited storage and a Pro plan around $10 a seat a month billed annually. Fathom is free for individual use with unlimited recordings, and charges for its Team plan if you want CRM syncing. Zoom's own AI Companion now comes bundled free with most paid Zoom accounts, which is probably why fewer small businesses bother paying for a separate tool than did two years ago.

The part vendors leave out of the demo

Every one of these tools is marketed as "never take notes again." What they don't put in the demo video is what happens to the relationship in the room while the bot is running.

People behave differently when they know an AI is transcribing them, and it's not the same as knowing a human assistant is taking notes. I've watched clients, especially older ones or those in regulated industries, go noticeably more guarded in the first ten minutes of a call once I've mentioned the bot is on. The easy chat that usually builds trust at the start of a meeting shrinks. They start speaking in tidier sentences, like they're being deposed. You lose something in that trade, even as you gain a transcript.

The second thing nobody says out loud is that the transcript becomes the record. Not your notebook, not your memory, not even the recording itself, which almost nobody re-watches. The AI's written summary is what gets forwarded, filed, and quoted back at you weeks later. If it's wrong, it's wrong in a document with your name on it that now outranks your own recollection in the client's mind.

How I use mine now

After the discount mix-up I changed how I run these tools. Here's what that looks like in practice:

1. Say it out loud, every time. "Just so you know, I've got an AI notetaker running so I can be fully present, is that alright with you?" Takes four seconds, and under UK GDPR you have a duty to be transparent about processing someone's personal data anyway, so it's not optional in spirit even where the law is fuzzy on strict consent.
2. Read the summary within an hour, not the next day. That's the window before it gets forwarded internally on the client's side.
3. Correct anything involving a number, a date or a commitment before you send it on. I now manually rewrite any line with a percentage or a price in it rather than trust the auto-generated version.
4. Send my own two-line summary alongside it for anything contractual, so there are two versions on record and mine is the one I wrote in full sentences on purpose.
5. Turn it off for specific conversations before they start, not after I remember I should have.

When to switch it off completely

There are conversations where I close the app before I even open the call:

• Any negotiation where I'm about to name a price or discount for the first time, until I've said the actual number myself, in writing
• Anything touching a client's staffing, redundancy or a personal issue they've brought to me
• Calls where a client is venting about a competitor or a supplier, since that gets stored as text somewhere I don't fully control
• The first call with a brand-new client, before there's enough trust for "I'm recording this" to land well

If you want to build a wider reflection habit around your own calls without a bot doing the interpreting for you, it's worth trying the same kind of structured self-review that works with ChatGPT prompts for self-discovery and reflection, asking yourself what you committed to versus what you meant to say, rather than outsourcing that judgement to a transcript.

The bit that's useful and underused: turning transcripts into content

Here's where I think most small businesses waste the tool. You've now got a written record of every question a client has ever asked you on a discovery call, in their own words. That's raw material.

Once a month I go through my call transcripts and pull out the questions that came up more than twice. Those become FAQ entries, short posts, or sections I add to an existing guide rather than a brand new one, which is really the same instinct behind why content curation is useful for your business, you're not always creating from scratch, you're organising what's already sitting in front of you. Brands that do this well tend to build entire content libraries out of real customer language rather than guesswork, which is part of what made a strategy like Wix's marketing approach work, real stories and real language rather than invented case studies.

It's also worth checking whether any of that content shows up when someone asks an AI tool a question about your industry. Running something like the 15-minute AEO audit will tell you fast whether the FAQs you're building from real client questions are even visible to the AI search tools people are increasingly using instead of Google.

If you're managing this across a whole team

One person misreading their own transcript is an awkward phone call. Five account managers each running notetakers with no shared policy on what gets recorded, corrected or shared is a genuine risk exposure, especially once client data starts sitting in a third-party AI company's servers with no consistent review process. If you're at the point where more than one or two people in the business are using these tools on client calls, it's worth getting an outside look at how the workflow is set up rather than each person improvising their own rules; that's the kind of gap an AI consultant for small business gets asked to sort out more often than people expect.

The uncomfortable truth about accuracy

I'll say the thing the sales pages won't. These tools are good enough to replace your notebook. They are not yet good enough to replace your judgement, and treating the summary as gospel is how small businesses end up honouring commitments they never made. The technology is improving fast, but "fast" still means it gets a client's specific number wrong often enough that you cannot skip reading it before it leaves your outbox. The time you save typing notes, you need to spend re-reading what the AI decided you said.

That's not a reason to stop using them. It's a reason to stop treating them as finished work.

Related reading: It's Not About the Money.

If you want the full set, start at the AI for Small Business: 25 Plain-English Guides to Automation, Marketing, Sales and Admin.

Related: my ai consultant contract what should be in it page.

Frequently asked questions

Do I have to tell clients I'm using an AI notetaker on our call?

You should, both as good practice and because UK GDPR expects transparency when you're processing someone's personal data, which a voice transcript is. A quick "I've got an AI notetaker running so I'm not scribbling while we talk, that alright?" at the start of the call covers it and takes seconds.

Which AI meeting notetaker is best for a small business?

Fathom is the strongest free option if you want unlimited recording without paying anything upfront. Fireflies suits teams that want cheaper per-seat pricing as they grow. Otter has the longest track record and integrates well if you're already inside Google Workspace. There isn't one clear winner, it depends on whether you need CRM syncing, which most of the free tiers don't include.

What do I do if a client already has a wrong AI summary from a call?

Correct it in writing as soon as you spot it, ideally within a few hours, and send a short two-line email restating what was agreed. Don't rely on a phone call alone to fix it, because the written summary is what gets filed and referred back to later, so your correction needs to exist in the same format.

Should I use an AI notetaker on sales or discovery calls?

Yes, they're useful there since you'll want an accurate record of what a prospect said they need. Just review the summary before you build a proposal around it, and don't let it record the specific figures until you've named them yourself in a follow-up email you wrote.

Further reading

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