The short version: Record yourself talking instead of sitting down to type, run the transcript through AI, and one 20-minute voice note can become a week of LinkedIn posts, a newsletter, and a blog outline before lunch. The part nobody tells you is that AI makes the words easy and the thinking still hard, and if you skip the human edit pass you end up with seven pieces of content that all say the same soft nothing.
Why I stopped typing my content
For most of the last five years, rebuilding my business in public, I typed everything. Blog posts, LinkedIn updates, emails to my list. It worked, but it was slow, and on the days I was tired (which after 53 years and a business that fell apart and had to be put back together, is most days) I just didn't write anything at all.
Then I started recording voice notes on my morning dog walk, a 40 minute loop near my house where I talk to nobody but the dog and my phone. I'd talk through whatever was on my mind: a client problem, something I'd noticed in the news, a mistake I'd made that week. I wasn't trying to write content. I was just thinking out loud.
It turns out thinking out loud, transcribed and cleaned up, is better content than most of what I was typing on purpose. It has rhythm. It has the actual words I'd use in conversation. It doesn't read like it was written by someone trying to sound clever.
The 20-minute voice note that became a week of content
Here's a real one, from a Tuesday a few weeks back. I'd just come off a call with a client who runs a small accountancy firm in Leeds, six staff, and she was furious because a competitor was using ChatGPT to answer client questions faster than her team could, and losing her nothing but making her feel behind. I recorded 22 minutes on my walk about that call: what she said, what I told her, why "AI will replace you" is the wrong fear and "AI will make your slow competitors look fast" is the right one.
That one recording, transcribed through Otter.ai, became:
- A 900-word LinkedIn post about the client call (published as-is, with her permission, name changed)
- Three shorter LinkedIn posts pulling out single ideas from the transcript
- One newsletter section, roughly 400 words
- A blog outline with five headings, which became the post you're reading a version of that pipeline in right now
- Two tweet-length quotes I could reuse for weeks
Twenty-two minutes of talking to a dog produced a week of content. That's the appeal, and it's real. But it only works because of what happened after the transcript, not because of the transcript itself.
The step-by-step pipeline
This is the exact process, not a rough sketch of one:
- Record while you're doing something else. Walking, driving, making dinner. Sitting still to record makes people perform. Moving makes people talk normally.
- Transcribe it. I use Otter.ai, which gives you 300 free minutes a month and around £16 a month if you need more; Fireflies.ai is similar, around £8 to £10 a month depending on the plan. A 20-minute recording gives you roughly 2,500 to 3,500 words of raw transcript, filler words and all.
- Feed the transcript to ChatGPT or Claude with a specific instruction, not a vague one. "Summarise this" gives you mush. "Pull out the three strongest single ideas from this transcript and write each as a standalone LinkedIn post in my voice, using my exact phrases where possible" gives you something usable.
- Ask for the boring formats too. A newsletter section. A one-paragraph version for an email signature link. A five-heading blog outline. Same transcript, different container.
- Rewrite the first two sentences of everything by hand. This is the step almost nobody does and it's the one that matters most. More on why below.
- Publish on a schedule you can keep, not the schedule the AI-generated content calendar suggests. One voice note a week, turned into five pieces, beats four voice notes a month that never get processed.
The tools and what they cost
People ask me constantly what this "" costs to run. Here's the real number, not a vague one: for a solo business owner doing this weekly, you're looking at somewhere between £0 and £30 a month total.
- Transcription: Otter.ai free tier covers 300 minutes a month, which is roughly 15 voice notes at 20 minutes each. Most people don't need to pay for this at all.
- Writing: ChatGPT Plus or Claude Pro, both around £16 to £18 a month. You only need one.
- Light editing for audio or video versions: Descript, around £10 a month if you're turning any of this into short video clips too.
Total real cost: £16 to £34 a month for a pipeline that used to take me four or five hours a week of typing and now takes about 90 minutes, split between recording, prompting, and the editing pass I'll come to.
Where this quietly falls apart
Here's the bit most people writing about "AI content pipelines" skip, because it makes the tools look worse than the sales pitch. If you take the AI's output and publish it with only light tidying, five pieces of content from one transcript will sound like the same paragraph copied five times with different opening lines. I've seen this happen to smart people whose LinkedIn feed used to have personality and now reads like a corporate newsletter that's been run through a translation app twice.
The reason is simple. AI is very good at restructuring what you gave it and reasonably good at matching a tone you've shown it. It is not good at inventing the specific detail that made your original voice note worth listening to in the first place. In my example above, the client's name (changed for privacy), the fact she runs six staff, the fact it was a Tuesday and she called me straight after the meeting, furious, before she'd even had lunch, that detail is what makes the post readable. AI won't add that unless it was already in your transcript, and it will happily strip it out if you ask for something "punchier" or "more concise," because concise, to an AI, usually means generic.
So the uncomfortable truth is this: the pipeline doesn't save you from having something specific to say. It saves you from the typing. If your voice note is 20 minutes of vague opinions with no names, numbers, or actual moments in it, you will get five vague posts out the other end, produced faster than before, and they will perform worse than the one honest post you typed slowly six months ago. AI content pipelines make good raw material go further. They don't make thin raw material good.
What separates the brands that make this work
Look at the brands that have built distinct voices and you'll notice the same pattern: specificity, repeated relentlessly, not volume for its own sake. I've broken down how Duolingo built a brand voice around one unhinged mascot rather than trying to sound clever across fifty different formats, and how Calm built theirs around slowness and restraint, the opposite instinct to "post more." Hotjar's marketing strategy leaned hard into showing real customer data rather than generic tips, and GoPro's approach works because every piece of content is somebody's actual footage, not a repackaged idea. None of them scaled by producing more generic content faster. They scaled by having a strong enough original thing that repackaging it kept working.
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.
That's the test I'd apply to your own voice note before you run it through anything. If you read the transcript back and there's nothing in it a competitor couldn't have said, the AI pipeline will just help you publish nothing, faster, five times a week. If there's a real client, a real number, a real mistake you made, the pipeline turns that one good thing into five good things instead of one.
A smaller trick worth stealing
One thing I do that's easy to copy: at the end of the AI prompt, I add "keep any specific numbers, names, or dates exactly as I said them, don't smooth them into generalities." That single instruction fixes about 80 percent of the "sounds like everyone else" problem, because it stops the model from doing the thing models default to, which is rounding your specific point into a general one that reads more like advice and less like something that happened to you.
If you're building anything interactive off the back of this content, like a quiz or a pricing tool to sit at the end of a newsletter, it's worth reading how interactive calculators convert visitors, because the same principle applies: specific inputs, specific outputs, no generic middle.
If you want help building this
I set this pipeline up for a handful of clients now, because most small business owners can follow the steps once but can't tell, on their own, whether their raw material (the voice note itself) is strong enough to survive the AI pass. That judgement call, more than the tools, is where an outside pair of ears helps. If you'd rather have someone build and stress-test this with you than guess at it alone, that's exactly the kind of practical setup I cover through working with an AI implementation coach, rather than another course that tells you which app to download.
Also worth mentioning, because people forget this: automation glue like IFTTT can quietly handle the boring bit, pushing your transcript from your recording app straight into a shared doc or your AI tool of choice without you touching a single file. I've written before about how IFTTT built its whole product around exactly this kind of invisible connection, which is the same idea, just applied to your own content workflow instead of theirs.
Frequently asked questions
How long should a voice note be for this to work?
Fifteen to twenty-five minutes is the sweet spot. Under ten minutes and you usually haven't gone deep enough into one specific story or idea to get more than one usable piece out. Over thirty and the transcript gets rambling enough that the AI struggles to find the strongest three ideas without your help.
Can AI make content sound like me, or will it always sound generic?
It can sound like you if you give it enough of your actual words to copy from, which is exactly what a transcript provides. The generic sound comes from asking for a summary instead of asking it to preserve your specific phrases, names, and numbers. Tell it explicitly not to smooth those out and the output improves sharply.
Is it worth paying for transcription tools or is the free tier enough?
For most solo business owners, Otter.ai's free 300 minutes a month is enough, that's about 15 voice notes at 20 minutes each. You only need to pay if you're recording daily or doing long client calls on top of your own content notes.
What's the biggest mistake people make with this pipeline?
Publishing all five or six pieces from one transcript without rewriting the opening lines by hand. AI tends to open everything with a version of the same sentence structure, so if you skip that one edit, your whole week's content reads like it came from the same template, because it did.
Related reading: Why Your AI Meeting Notetaker Might Be Breaking the Law (And Killing Your Sales Calls) and Why Your Analytics Are Hiding How Many Customers AI Is Sending You.
