Most content agencies sell deliverables.
They tell you how many blog posts you’ll get per month. How many social posts. How many emails in the nurture sequence. Maybe a quarterly thought leadership piece. The pricing is built around volume — more posts equals higher tier.
It’s the wrong unit.
The actual asset a content agency produces isn’t a stack of blog posts. It’s the brand voice the posts are written in. And brand voice is the load-bearing thing every other content asset rests on.
Why voice is load-bearing
If your voice is undifferentiated — if your blog could be your competitor’s blog with the logo swapped, if your newsletter sounds like every B2B newsletter, if your LinkedIn presence reads like every other LinkedIn presence — none of the deliverables matter. The frequency doesn’t matter. The volume doesn’t matter. The hours invested in content production are subsidizing a thing nobody will remember.
Conversely: if your voice is specific, distinct, and recognizable — if a reader could tell a paragraph of your writing from a paragraph of your competitor’s, even with the bylines removed — then almost everything else becomes leverage. The same hour of writing produces an asset that compounds in attribution. The same newsletter earns the open. The same blog post gets shared and remembered.
Voice is the asset. Deliverables are how voice gets distributed.
Most agencies don’t price it this way because voice work is hard to scope, hard to scale, and hard to invoice for. Templates and frameworks are easier. So agencies sell what’s easier to sell, and the client buys what’s easier to buy, and everyone agrees the deliverables are the thing.
The deliverables aren’t the thing.
Why templates don’t work
I’m going to make a claim that will sound extreme: most content templates produce worse content than no template at all.
Hear me out.
A template is a fill-in-the-blank structure designed to make average content easier to produce. The promise is consistency — you’ll get a passable output every time without having to think too hard about structure.
The reality is that templates strip out the things that make writing distinctive. Templates are designed against the worst-case scenario (the writer who can’t structure a sentence) and they punish the best-case scenario (the writer with a specific point of view that doesn’t fit the template’s shape).
When you template your content, you guarantee a floor. You also guarantee a ceiling. And in a category where attention is the constraint and differentiation is the moat, a ceiling on distinctiveness is a death sentence.
The newsletter that gets opened is the one that sounds unmistakably like a specific person. The LinkedIn post that gets shared is the one with a real argument structured by the argument’s needs, not by a template’s needs. The thought leadership piece that earns inbound is the one nobody else could have written.
You can’t template your way to that. Templates produce content that’s professional and forgettable.
The AI question
This is the part where I’m supposed to argue that AI can’t replicate voice, that the LLMs will produce average prose forever, that human writers are irreplaceable.
I’m not going to argue that. I don’t fully believe it.
Here’s what I actually believe about AI in writing.
I use AI as a collaborative partner and research assistant constantly. It’s useful. It saves hours per piece. It surfaces angles I wouldn’t have considered. It helps me stress-test arguments. It accelerates research that would otherwise be tedious. It drafts paragraphs I then rewrite three or four times until they sound like me.
I’d be lying — and most working writers would be lying — if I claimed otherwise.
But that’s not the same as saying AI can produce your voice on its own. It can’t. Not because the model is bad — the model is increasingly impressive — but because voice isn’t a style. Voice is a series of editorial decisions about what to say, what not to say, what to emphasize, what analogy to reach for, how to open, how to close, what argument is actually worth making to your specific reader in this specific moment.
AI is genuinely good at producing prose. It’s not good at editorial judgment. It cannot decide what’s interesting or true or worth your reader’s attention. It cannot know what your specific business needs to communicate to your specific customer in this specific moment of your specific market.
You can put a transcript of your last sales call into an LLM and ask it to draft a blog post. You’ll get something. The something will probably be passable. It won’t be load-bearing — because the load-bearing part of the post is the editorial work, and the editorial work is what you, the human, have to do.
Where the trade actually is
The actual trade isn’t AI vs human. The actual trade is between two different kinds of human attention.
One kind is the operator-founder who has all the editorial judgment but no time to write. Their voice exists in their head; it never gets onto a page. AI doesn’t solve their problem — they don’t have hours to spend prompting and editing, and they don’t trust raw AI output enough to publish under their name without significant revision. So they publish nothing.
The other kind is the agency that produces in volume by using AI as the primary draft engine and not investing in voice capture. The output is fast, cheap, and indistinguishable from a thousand other agencies’ output. The client gets a stack of “content” that fails to compound because none of it sounds like them.
The actual answer is in the middle: a writer who has captured the operator’s voice deeply, who knows what the business is trying to produce, and who exercises real editorial judgment about what to say and how to say it. That writer treats AI as a tool — useful, frequently — but treats voice as the asset.
The volume of AI usage is mostly noise. What matters is whether the person at the keyboard has internalized the voice they’re writing in, and whether they’re using their editorial judgment to ensure the output sounds like the person it’s supposed to sound like. With or without an LLM in the workflow.
A human is accountable
Here’s the bottom line I come back to, regardless of where the technology goes.
A human is accountable for what gets published under your name.
If you have a content program and one piece misses badly — wrong tone, wrong claim, embarrassing position — a human has to answer for it. A human has to own the call. A human has to look at the published piece, take responsibility, and decide what to do differently next time.
That accountability isn’t abstract. It’s the thing that makes editorial judgment possible. When a human is on the hook, the editorial decisions sharpen. When nobody is on the hook — when content is just what the system produced — the editorial decisions collapse into whatever the template or the model produces.
The voice is the asset because the voice is where the accountability lives. Strip away the voice and you strip away the accountability, and you’re left with content nobody owns. Content nobody owns can’t compound, because there’s no consistent point of view for it to compound into. There’s no asset. Just output.
So when an agency tells you about the deliverables, ask about the voice. Ask how it gets captured. Ask whose voice the content actually sounds like — the founder’s, the company’s, or no one’s in particular. Ask how the agency makes sure each piece sounds like you and not like the agency’s other clients.
If they can’t answer those questions, you’re buying volume, not voice. The deliverables will get produced. The asset will not.
The Signal newsletter is where the more concrete version of this thinking lands every week — what to ask for, what to demand, what to look for in any content being written under your name. Subscribe below.