Straight answer: AI meeting notetakers like Fireflies, Otter and Fathom are brilliant at capturing what was said and useless at telling you what to do about it, and most small business owners are now sending follow-up emails that read like a transcript dump rather than a human who was listening. Fix the follow-up, not the recording, and you'll close more of the deals you're currently losing to silence.
The tool did its job. The deal still died.
I ran 43 discovery calls between January and March 2026, every single one recorded and summarised by an AI notetaker. Full transcripts, action items pulled out automatically, key quotes highlighted, the lot. On paper, brilliant. In practice, I closed fewer deals in that quarter than I had in the same period the year before when I was scribbling notes on a legal pad like it was 2009.
I went back and read twelve of the follow-up emails my team had sent using the AI-generated summaries as a base. Every one of them had the same structure: "Great speaking with you today, here's a recap of what we discussed" followed by three bullet points lifted almost word for word from the transcript. Correct. Accurate. Completely dead on the page. There was nothing in there that told the prospect we'd understood why they were struggling, only that we'd heard them say it.
That's the bit nobody selling you these tools wants to say out loud. Capturing a conversation is not the same as understanding it, and a prospect can tell the difference within one sentence.
Why "everyone has the notes" quietly makes follow-up worse
Before AI notetakers, if you wanted to write a good follow-up email you had to remember the call, which meant you'd processed it. Now the transcript exists whether you paid attention or not, and that changes behaviour in a way I didn't expect.
Three things happen once a team gets used to AI capturing everything:
- People stop taking their own notes during the call, so they stop forming their own opinion about what mattered
- Follow-up gets delegated to whoever is free that afternoon, not the person who was on the call, because "the transcript has everything anyway"
- Everyone assumes someone else actioned the summary, so action items sit in a shared doc for a week untouched
I've seen this exact pattern in three client businesses now, a marketing agency in Manchester, a small accountancy practice, and a SaaS reseller. All three had adopted an AI notetaker within the past year. All three had a graveyard of "action items" in a shared tool that nobody had looked at since the day they were generated. The AI did the recording perfectly. The humans stopped doing the following up, because the record felt like the job was already done.
The number that made me stop and fix it
Here's the bit that made me change our process. I pulled the data on our 43 calls and found that only 14 had a follow-up email sent within 24 hours. Nineteen went out somewhere between two and five days later. Ten never got a follow-up at all, they just sat in the CRM marked "call complete" with a transcript attached and nothing else.
Ten out of 43. That's 23% of qualified sales conversations where we had a full, accurate, AI-generated record of everything the prospect said, including their exact pain points in their own words, and we did nothing with it. Before we had notetakers, when following up meant sitting down and remembering the call, our no-follow-up rate was under 5%. The tool made us worse at the one thing that moves a deal forward.
What I do differently now (the five-step version)
I haven't ditched the notetaker. I use Fireflies most weeks. But I changed how we use the output, and it's brought our follow-up rate back down to almost zero missed calls.
- Step 1: Someone owns the call before it starts. Not "whoever picks it up in the summary channel". The person who was on the call writes the follow-up, full stop. This is a rule now, not a preference.
- Step 2: The AI summary is a checklist, not a draft. We read it to confirm we didn't miss a commitment or a number, we don't paste it into the email.
- Step 3: The follow-up email opens with one sentence only the human on the call could have written. Something about tone, hesitation, a joke they made, the thing they said off-script. If you can't write that sentence, you weren't listening on the call, and the transcript won't save you.
- Step 4: Follow-up goes out within four hours, not 24. We set a Slack reminder that fires the moment the notetaker posts the summary. Four hours is the window where the prospect still remembers wanting to hear from you.
- Step 5: Action items get a name and a date attached in the same message, not a separate task. "I'll send the pricing breakdown by Thursday" beats "action item: send pricing" sitting unowned in a doc.
None of this is complicated. It's the same discipline good salespeople had before any of this software existed, we've just had to rebuild it on purpose because the tool made it too easy to skip.
The uncomfortable bit about "productivity" tools generally
This is where I'll say the thing most posts about AI productivity tools tend to skip past. Every AI tool that captures, summarises or drafts something for you removes a small piece of friction, and friction is often where the thinking happened. When you had to type your own follow-up email from memory, the act of remembering forced you to decide what mattered. Remove that friction and you get speed, but you can lose the judgement that used to live inside the slow bit.
This isn't an argument against AI tools, I use several every day and I'm not going back to legal pads. It's an argument for noticing exactly which piece of judgement the tool removed, and deciding whether you're going to rebuild it somewhere else or just let it disappear. Most small businesses adopt a tool, get the speed, and never ask that second question.
Where this shows up in your numbers
If you want to know whether this is happening in your business, don't ask "do we use an AI notetaker." Ask these three things instead:
Pull your last 20 sales calls. How many got a follow-up within 24 hours? How many follow-up emails could have been sent to any prospect, with no detail specific to that conversation? And how many action items from meeting summaries from the last 30 days are still unactioned?
If your numbers look like my 23%, you don't have an AI problem, you have a follow-up ownership problem that AI made easier to hide. The fix isn't a better tool. It's putting a human name and a clock on every summary the second it's generated, the same discipline brands that win on customer experience build in everywhere else in the business. Look at how deliberately GoDaddy structured its customer-first marketing approach around never leaving a customer touchpoint unowned, or how Headspace built its entire brand voice around sounding like one specific calm human rather than a corporate summary generator, something worth studying in their Headspace marketing strategy. The tone consistency isn't an accident, it's a rule someone enforces on every single piece of output.
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.
What good AI-assisted follow-up sounds like
Here's a real (lightly anonymised) example from one of the calls that did convert. The prospect had mentioned, almost as an aside, that she'd been burned by a previous agency that "disappeared after the deposit." The AI transcript flagged this under "concerns raised" as a single generic line. The follow-up email my colleague sent opened with: "I noted what you said about the last agency going quiet after they'd been paid, that's exactly the thing I want to make sure doesn't happen here, so here's what our first 30 days look like week by week." That's not a transcript summary. That's a human who heard a specific fear and answered it directly. She signed two weeks later.
Behavioural economist Richard Thaler's work on choice and friction is relevant here too, people don't decide based on the complete list of facts, they decide based on the one thing that reduced their anxiety, and there's a good breakdown of how that plays out commercially in these business lessons from Richard Thaler. An AI summary gives you all the facts. It doesn't tell you which one to lead with. That judgement call is still entirely yours.
A word on tone, because this is where most AI follow-up falls apart
The bullet-point-recap email isn't just boring, it actively signals "this was templated" even when it wasn't. Prospects who've had a handful of sales calls in the past year have seen enough AI-flavoured follow-ups to spot the pattern instantly: bold header, three bullets, a CTA. Coco Chanel built an entire empire on the opposite principle, stripping things down to what mattered and removing everything ornamental that didn't serve the person wearing it, a discipline worth revisiting in these Coco Chanel business lessons. Your follow-up email doesn't need more structure. It needs less template and one true sentence.
If your team is engaging prospects rather than just informing them, you're often better served pairing that human follow-up with something interactive the prospect can act on immediately, a pricing calculator, a fit assessment, something that gives them a reason to come back to you rather than just read and archive. There's a good breakdown of why interactive tools convert better than static recaps in this piece on engaging and converting visitors with interactive calculators, and the same principle applies to sales follow-up, not just website traffic.
When to bring someone in to fix the whole system
If you're reading this and recognising your own CRM full of untouched action items, this usually isn't a one-person fix. It's a process problem across your whole sales and marketing stack, and it's exactly the kind of thing I walk small businesses through when I work as an AI implementation coach, mapping where a tool removed friction and where that friction needs rebuilding on purpose rather than left as a gap nobody notices until the pipeline dries up.
Frequently asked questions
Should small businesses stop using AI meeting notetakers?
No. Tools like Fireflies, Otter and Fathom are good at capturing accurate records of calls. The problem isn't the tool, it's treating the summary as the finished follow-up rather than a checklist you build a human response around.
How quickly should a sales follow-up go out after a call?
Within four hours if possible, and no later than the same working day. Our data across 43 calls in early 2026 showed follow-ups sent after 24 hours converted at roughly half the rate of those sent within four hours.
What's the biggest mistake small businesses make with AI meeting notes?
Pasting the AI-generated summary directly into a follow-up email. Prospects can spot a templated recap instantly, and it signals nobody on the call was truly listening, even when they were.
How do I know if AI tools are hurting my sales process?
Check your last 20 calls: how many got follow-up within 24 hours, how many action items are still sitting unactioned after 30 days, and how many follow-up emails could have been sent to any prospect with zero changes. If those numbers are high, the tool isn't the problem, but it's making the ownership gap easier to hide.
Related reading: How to Use AI to Save 10 Hours a Week: My Real Weekly Breakdown and AI Meeting Notes: What They Get Right, What They Miss, and When I Switch Them Off.
I go much deeper on this in the AI marketing guide.
