The Coaching Practice That Replaced Its Content Team With Three Prompts

The coach had been paying $4,500 a month to an offshore content team for 18 months when we met.

She wasn't unhappy with the team. She was unhappy with the output. The LinkedIn posts sounded like a competent stranger wrote them. The newsletter felt like every other coach's newsletter. Her engagement was flat. She wasn't getting inbound leads from content anymore even though she was posting 4x a week.

The problem wasn't the team. It was that her content had stopped sounding like her.

Why outsourced content goes flat

Outsourced content writers default to the average. They have to. They write for 20 different clients, all in adjacent niches, and they cannot hold each client's voice in their head with high fidelity.

So they write to a competent center. Inoffensive. Linkedin-able. Generic.

If you're a coach whose differentiation is your voice — your perspective, your turn of phrase, the things you say that make your clients say "I needed to hear that" — outsourced content actively erodes your brand. It makes you sound like every other coach.

The fix is not to fire the outsourced team. The fix is to start with raw material that is unmistakably YOUR voice and then have AI (or the team) transform it. Garbage in, generic out. Voice in, voice out.

The three prompts

The system was three prompts. That's it. Not a tool. Not a platform. Three prompts.

Prompt 1: Transcribe and clean a voice memo. Coach records a 7-minute voice memo on her phone whenever she has a thought. Whisper transcribes it. Claude cleans the transcript into readable prose while preserving every distinctive phrase, every fragment, every voice tell. Output is a "raw essay" of about 800-1200 words in the coach's actual voice.

Prompt 2: Turn the raw essay into a LinkedIn post. Take the raw essay. Pull out the strongest single argument. Compress to 1200 characters with line breaks, no hashtags, ending on a question or a punch. The prompt has a 12-example library of the coach's prior best-performing posts and explicitly tells Claude to match that pattern, not LinkedIn's average.

Prompt 3: Turn the raw essay into a newsletter. Same input. Different output. 600-900 words. Subject line that doesn't read like a robot wrote it. PS line that always references something specific (client, anecdote, current event).

All three prompts run from the same source material. The voice memo is the single source. The outputs are derivatives.

What the system replaced

The outsourced team produced: - 4 LinkedIn posts a week - 1 newsletter a week - 1 blog post a week - Various carousels and reels scripts

The new system produces: - 4 LinkedIn posts a week - 1 newsletter a week - (blog and carousels are now done by the coach's part-time VA using the raw essays as starting material)

Net savings: $4,500/mo down to about $600/mo (VA hours + Claude API spend + Whisper API spend).

The catch

The catch is the coach has to record voice memos.

The system runs on her voice. No voice, no output. If she stops recording, the content stops.

In month one she nearly killed it because she didn't record for 8 days and the VA had nothing to work with. We solved this with a single-question Tuesday morning text. "What's on your mind today, give me 5 minutes." She'd record on her commute. The text became a forcing function.

If you build something like this you need a forcing function. The AI is a multiplier. The input is the bottleneck.

What about quality

Engagement was up 60% by month three. Inbound DMs were up 4x. Newsletter open rate climbed from 27% to 38%.

I'm careful with these numbers because LinkedIn algorithms move and inbound is noisy. But the direction was unambiguous and held for the next six months.

The thing that moved the engagement was distinctiveness. The posts didn't read like content marketing anymore. They read like a person had something to say. Algorithmic boost follows distinctiveness because comments and shares follow distinctiveness.

What didn't work

The first version of Prompt 2 generated really good LinkedIn posts that did not sound like the coach. Claude has a default LinkedIn voice and it is competent but generic.

We fixed it by stuffing the prompt with 12 of her best prior posts (we picked them based on the posts that had generated DMs, not the ones with the most likes) and explicitly telling Claude: "your job is to match THIS voice, not to write a good LinkedIn post."

You can't outsource voice without giving the AI the voice to match. That's the whole game.

The other thing that didn't work: trying to schedule batches. We tried generating a week's posts on Mondays. The result was a week of posts that felt out of sync with anything happening that week. We went back to same-day generation. Voice memo Monday morning, posts Monday afternoon. Voice memo Wednesday morning, posts Wednesday afternoon.

The freshness is part of the voice. Batched content loses the freshness.

What I'd tell another coach starting

If you're paying for content and it doesn't sound like you, the problem is the input.

You can fix it without firing anyone. Tell your writer that you're going to record voice memos and they're the source material. Watch what happens to the output over the next four weeks.

If you don't have a writer and you're trying to do this solo, build the three prompts yourself. They're not complicated. Each one is a paragraph of instruction plus 8-12 examples of your prior best work. Total setup time is 90 minutes.

The AI part is the easy part. The voice part is the work.