The moment I could tell which ghostwriter wrote a post — without being told

I knew something was wrong before I could name it.

I’d taken on a new client, a founder whose previous ghostwriter had moved on. I sat down to read six months of their LinkedIn posts, the way I always do before writing a single word for someone. Reading for voice. Reading for tells. Reading for the shape of how they actually think.

Except the shape wasn’t theirs.

I scrolled through their feed, and I could feel another writer in the posts. Same cadence. Same opening moves. Same structural tricks. Same three-word punchlines dropped at the same beats. I wasn’t reading a founder’s public voice. I was reading a ghostwriter’s recipe, executed one hundred and forty-seven times with a different logo at the top.

Then I opened another client’s inherited work. Different industry. Different personality. Same writer’s fingerprints.

Then a third.

That’s when I named it. Voice Decay. The slow, systemic erosion of a founder’s public voice when a ghostwriter approximates instead of extracts. The client becomes a manufactured item rather than a stand-alone thinker. And most of them have no idea it’s happening.

This isn’t a craft problem. It’s a business model problem.

Here’s what I’ve come to understand, and it’s not a gentle take:

Voice decay is laziness from ghostwriters.

Not all ghostwriters. Not the good ones. But the ones operating at volume: agencies with forty, a hundred, two hundred clients running through the same machine every week, they cannot extract a real voice from every founder on their roster. It’s not possible. There isn’t enough time. There isn’t enough attention. There aren’t enough of them to go around.

So they built a shortcut. A template. A prompt in ChatGPT. A library of hook formulas and framework skeletons that can be dressed up with surface-level details about the client, their industry, a couple of client stories, the word they use for their customers, and shipped as “personalised” content.

And it looks personalised. If you only ever see your own posts.

The clients think they’re getting bespoke work. They’re not seeing the other thirty-nine founders receiving near-identical structural content the same week. They have no frame of reference. The delivery is polished, the posts go out on time, engagement is fine, so they assume the system is working.

The system is working. Just not for them. It’s working for the agency that’s charging £400 a month per seat and running a 2,000-post-a-year content factory with a tiny team and one master prompt.

The tell

I can open a LinkedIn feed and, within three posts, tell you which template seller wrote them.

One agency in particular.. I won’t name them, but anyone who has looked carefully at the LinkedIn ghostwriting space for longer than a year will know exactly who I mean, runs the same weekly template across every single client on their books. Small surface adjustments. Different proper nouns. A nod to the client’s industry in paragraph three.

But the architecture is identical. Every week. Across a personal trainer, a SaaS founder, a trauma counsellor, a recruitment consultant, a mindset coach. The same structural bones. The same rhythmic tics. The same “I used to think X. Then Y happened. Here’s what I learned.” spine, dressed in whatever industry costume fits.

The client thought they were unique. They weren’t. They were one of many running through a factory that had stopped extracting and started manufacturing a long time ago.

And I felt quite disappointed reading it. Because these are real founders. They have real conviction, real stories, real opinions that would, if excavated properly, build them an actual authority position in their market. Instead they’re being processed.

The real cost nobody wants to name

Here’s what agency ghostwriting at volume doesn’t put in the pitch deck:

A founder with a decayed voice cannot sell the way a founder with a real one can.

You can have beautiful engagement. You can grow followers. You can clear 50,000 impressions on a Tuesday post. And still not generate a single inbound DM from the kind of person who would actually buy from you, because the content has no diagnostic edge. It doesn’t name anything. It doesn’t challenge anything. It doesn’t take a position a competitor wouldn’t also take.

It’s competent. It’s consistent. It’s dead.

This is the part that makes me angry. Not the ghostwriters chasing volume, they’re just running the business model they chose. It’s that the founders paying for it are being sold an authority position and receiving a performance of one. They don’t know the voice is decaying until six months in, when the DMs dry up, or the clients they do close feel confused because the person on the Zoom doesn’t sound anything like the person on the feed.

The ghostwriters at volume are not thinking about what comes out the other end. What it means for the business. What the implications are. They just want to put something in the AI, fire it out, and take the retainer.

That’s the laziness. Not the writing. The thinking.

Why I made extraction non-negotiable

Every founder I work with goes through the same process before I write a word of content for them.

We sit on a call. Sometimes two. I ask them questions that have nothing to do with content strategy. I ask them what they believe that the rest of their industry is wrong about. I ask them the story they keep telling privately that they’ve never told publicly. I ask them what they would say to their market if they didn’t care about being likable.

I record all of it. I transcribe it. I read it back looking for the phrases nobody else would use. The positions that have bite. The angles that sound like them and only them.

Then I write from that. Not from a template. Not from ChatGPT’s output layered with a “client tone” modifier. From their own words, structured into an argument.

This is slow. It’s why I can’t have forty clients. I can have a few, and I can do this properly for them, but it costs more because the work costs more. That’s the whole model.

The output looks different. When you read a founder I work with, you’re reading them. Their convictions, sharpened. Their voice, structured. Their authority, positioned. Not a recipe with their name on it.

What to do if you suspect your voice has decayed

Three questions. Answer honestly.

1. Could someone else’s ghostwriter have written your last five posts? If yes, not because they’re bad, but because the structural DNA is generic, your voice has decayed.

2. Do your posts say something your direct competitor couldn’t also say? If not, you don’t have a voice. You have a frequency.

3. Have you had a DM from an ideal buyer this month that referenced something specific you said, not just a vague “love your content”? If not, the content is performing engagement, not authority.

If all three land badly, you don’t have a content problem. You have a ghostwriter problem. Or more accurately, you have a ghostwriting model problem.

The fix is not a new template. It’s extraction.

The ask

If you’re a founder reading this and something just tightened in your chest, hit reply. Tell me what your current content setup is.

I’ll tell you, honestly, whether it’s decaying your voice or keeping it alive.

Not a pitch. A read.

— Sarra

“54% of LinkedIn posts are now AI-generated.”

“AI agents are replacing $3K/month ghostwriting agencies.”

“Your job is over.”

Every week, someone screenshots a new tool and posts it with the subtext: you’re obsolete.

They’re reading the data backwards.

What the numbers actually say is this.

LinkedIn’s 2026 algorithm update didn’t just reduce reach on generic content. It created a 237x performance gap between posts that earn genuine engagement and posts that don’t. Company pages now reach 1.6% of their followers. One well-crafted post that earns depth and discussion is worth more than months of filler.

The algorithm isn’t neutral about this. It’s making judgments, fast, and punishing the obvious.

That 237x gap isn’t a warning to ghostwriters. It’s a hiring brief.

Subscribe

Here’s what I think is actually happening under the surface.

When something floods a space, any space, the people in it learn to filter automatically. They don’t consciously decide “that’s templated AI output.”

They just feel the familiar rhythm of it, the particular shape of a hook that starts with “I used to think…” and ends with a lesson about resilience, and they scroll. Without thinking. Instantly.

The algorithm is doing the same thing, but faster and at scale.

AI content flooded the platform. The platform adapted. Now 54% of posts are functionally deprioritised, not with a label, but with reduced reach, reduced distribution, reduced chance of being seen by anyone who matters.

The people still building audience in this environment aren’t posting more. They’re posting differently.

There’s a distinction I keep coming back to when I talk to founders about this.

There are services that offer you more of the same, faster. More posts. More frequency. Lower cost per word. They compete on efficiency, which means they compete on price, which means they race to the bottom.

And then there are things that offer you something in a completely different category. Not better content production. Different content altogether, the kind that sounds like you, carries the specific weight of how you actually think, and builds the kind of trust that converts.

Most AI tools and a lot of ghostwriters, if I’m being honest, are in the first camp. They’re offering an improvement on what already exists.

What I do is the second thing. And those two things don’t compete with each other, because they’re not the same product.

When a founder understands that their LinkedIn presence is a trust asset, not a content production line, they stop asking “how do I get this cheaper?”

They ask “how do I get this right?” Price stops being the question. It was never really the question.

Share

What’s telling me the most right now isn’t the stats.

It’s the tools.

People are building GitHub repositories, going viral on X, dedicated to making AI writing sound less like AI. Anti-slop writers. Brand voice files. Prompts to strip the template out of text that started as a template.

Sit with that for a second.

The market is building tools to solve a problem that the market created.

Voice is the hardest thing to fake, so everyone is trying to fake it, which means everyone is building tools to detect the faking, which means the faking is getting worse, which means the detection gets sharper.

This is not a problem that a better prompt solves. This is a structural problem.

And the structural answer to “everyone sounds the same” is “someone who makes you sound like yourself.”

If you don’t control your own voice, something else will fill that space. A template. A tool. A generic version of what you might have said if you’d had time to think about it properly.

That version will reach 1.6% of your followers. It will earn you a spot in the 54%. And your audience, the sharp ones, the ones you actually want, will scroll past it without knowing why.

Here’s what I want ghostwriters to stop being afraid of.

The tools aren’t competition. They’re market validation.

Every new AI writing product that “learns your voice” is proof that voice is the hardest part. If it were easy, people would stop building tools to solve it.

The fact that they keep building them, that founders keep buying them, tells you exactly how much the problem costs and how desperate people are to fix it.

The interview, the listening, the reading between the lines of what someone says versus what they mean, that’s not something that gets automated. It barely gets taught.

The 237x gap isn’t closing. It’s widening as more AI content enters the platform and the algorithm gets sharper.

That gap is the entire business case. For all of us who do this properly.

Not the ghostwriter who drops a voice note into a tool and calls it done. The one who sits in a conversation, catches the phrase someone uses that they don’t realise is interesting, notices the offhand comment that’s actually the whole point, and builds something from that.

That work isn’t faster because of AI.

It’s better.

And right now, better is exactly what the algorithm rewards.

She'd been publishing for eighteen months.

Eighteen months of content, and not a single piece sounded like the person who built the thing.

That's not a writing problem. That's an infrastructure failure.

Here's what made me think about this.

A stat dropped last month that I haven't been able to stop turning over. 83% of consumers can now detect AI-generated content. Not suspect it. Detect it. They read something and their brain flags it before they can articulate why.

72% say they feel deceived when they find out.

And 75% of marketers are still using AI to generate their content anyway.

(Read those three numbers again. Slowly. That's not a gap. That's a chasm with a gift shop.)

Subscribe

The trust problem nobody's naming

CNN called 2026 the year of anti-AI marketing. Merriam-Webster made "slop" their Word of the Year. The dictionary literally named the problem before most agencies noticed they were causing it.

Here's what's actually happening. Your audience doesn't say "this was written by AI." They say something worse. They say nothing. They scroll past. They feel that something's off, the same way you feel when someone's being polite but doesn't mean it, and they move on to someone who sounds like a person.

The absence of a reaction is the reaction.

64% of consumers, according to Adobe's Trust Report, will actively lose trust in your brand if they discover your content was primarily AI-generated without disclosure. Not "feel mildly annoyed." Lose trust. The thing that takes years to build and seconds to destroy.

And here's the bit that keeps me up. The brands leaning hardest into unscripted, messy, genuinely human content are standing out more than ever. Not because authenticity is new. Because it's now scarce.

That's the market flip nobody warned you about

What AI slop actually looks like (and why your agency won't tell you)

I want to show you something. Because I think most founders don't realise what's happening to their content between the briefing call and the publish button.

What the agency sends you:

> "At [Company], we believe in delivering innovative solutions that empower our clients to navigate today's complex business environment. Our client-centric approach ensures that every engagement is tailored to meet your unique needs, driving measurable results and long-term value."

What the founder actually said in the briefing:

> "Honestly, I just got sick of watching people get ripped off. We'd been in the industry fifteen years and kept seeing the same corners being cut. So we built something that actually works, and now our clients stay for an average of four years. Not because we lock them in. Because they don't want to leave."

One of those is a human being with fifteen years of scar tissue and a reason to exist. The other is a paragraph that could describe literally any company in any industry in any country on Earth.

Both cost the same to produce.

One builds trust with that silent 83% who are scrolling your page at 10pm deciding whether you're worth a conversation. The other makes you invisible.

Share

The three-second test

Here's something you can do right now. Takes three seconds. No framework, no system, no consultation needed.

Open your most recent LinkedIn post. Read the first two sentences out loud.

Would you say those words to someone in a pub?

If the answer is no, your content doesn't sound like you. It sounds like a committee. And committees don't win clients.

(I ran this test on 30 founder profiles last month. 26 of them failed. Not because the founders are bad writers. Because somewhere between their brain and the published post, every rough edge, every opinion, every bit of personality got sanded off.)

The rough edges are the whole point. That's where the voice lives.

Why this is getting worse, not better

94% of B2B buyers used an LLM during their purchase journey last year. ChatGPT, Claude, Gemini. They're asking AI about you in private conversations you'll never see.

And here's the thing about LLMs. They pull from your published content. Your LinkedIn, your website, your Substack. If every piece of your content sounds like every other piece of content in your industry, the machine has nothing distinctive to reference. You become interchangeable in the AI layer too.

You're now invisible in two dimensions. The human scroll and the machine summary.

The founders who sound like themselves, who have a specific point of view, specific language, specific rhythm, those are the ones the machines can actually distinguish. Because there's something there to distinguish.

A voice isn't a nice-to-have anymore. It's infrastructure.

The uncomfortable question

I'm not going to pretend this is easy. Building a real voice system takes time. It takes someone sitting with you for long enough to hear the thing you do with language when you forget you're being professional. It takes pattern recognition. It takes the willingness to sound like yourself, which is harder than it sounds when you've spent twenty years learning to sound like everyone else.

But here's what I know after pulling apart about 50 brands at this point.

Every single person has a voice. No exceptions. The founders who think they're boring are usually the most interesting. The ones who think they have nothing to say are sitting on twenty years of opinions they've been too polite to publish.

The voice is already there. You just don't have a system for it yet.

And in a world where 83% of your audience can smell the AI, where 72% feel deceived by it, where the dictionary literally named the problem and the news is reporting the backlash, having a voice that sounds like an actual human being isn't a luxury.

It's the only defensible position left.

Sarra Richmond is The Ghost.

She finds the voice you can't get on the page.

50 years of reading people. One operating system.

If your content could've been written by anyone, that's fixable.

I watched someone demo their "voice clone" last week.

Custom GPT. Trained on 200 LinkedIn posts. 50 emails. A handful of podcast transcripts. The whole thing took about forty minutes to set up, and ho…

Read more

He had a voice. He just didn’t have the infrastructure to see it.

Read more

Read more

Every founder, CMO, and marketing lead I’ve worked with has said them. Usually while staring at a draft somebody on their team just sent over.

“It doesn’t sound like us.”

Five words. And beh…

Read more