Skip to content
AI and content

An AI content agency sells volume. Volume is free.

Producing content stopped being expensive about two years ago. Deciding what is worth producing did not get cheaper by a single minute, which makes it the entire cost now.

So AI runs backstage here, on the research and the drafting and the repurposing. It is never what you are buying.

The production pipeline4 of 8 automated
  • AIResearch and source gathering
  • AIOutline against the brief
  • AIFirst draft
  • AIRepurposing into other formats
  • HumanDeciding it is worth publishing
  • HumanThe claim, and whether it is true
  • HumanThe opinion it was never given
  • HumanPressing publish

The dashed line is the whole argument. Everything below it is judgement, and judgement is what you are actually buying.

The inversion

Three things collapsed in price. Three did not move at all.

Every argument on this page follows from this table. Most ai content services are priced against the top three rows, which are now close to worthless as a thing to sell.

The task2022Now
Drafting two thousand words6 hours4 minutes
Researching a topic properly3 hours20 minutes
Repurposing into six formats4 hours10 minutes
Knowing which topic is worth itunchangedunchanged
Having an opinion worth readingunchangedunchanged
Being accountable for a claimunchangedunchanged

Illustrative timings. The direction is not in dispute, and neither is the fact that the bottom three rows are where all the remaining difficulty lives.

The line

Four jobs a model does not do here.

Not because of a policy, but because it cannot. Each of these needs somebody who was present, is accountable, or has something to lose.

  • 01

    The customer conversation

    A model can summarise a call it is given. It cannot book one, notice the thing the customer almost said, or ask the follow-up question that turns a complaint into a piece nobody else could write. Every genuinely differentiated article on this site started as a recording.

  • 02

    The claim, and whether it is true

    Language models are fluent about things that are not the case, and they are most fluent precisely where the detail is obscure enough that nobody checks. Somebody has to be professionally accountable for every factual sentence, and that person has to be a person.

  • 03

    The opinion it was never given

    Ask for a view and you get the average of everything written on the subject, which is by definition what your competitors already published. A position that costs you something to hold cannot be generated, because it was never in the training data.

  • 04

    The decision not to publish

    The single most valuable act in this pipeline is throwing a finished draft away. No tool will ever suggest it, because no tool bears the cost of a mediocre page sitting on your domain for three years.

Read this before you sign anything

What an AI content agency cannot promise you.

This is the newest category on the site and the one where the pitch and the mechanism have drifted furthest apart. An AI content agency is usually selling throughput, and throughput is the one thing that stopped being scarce. Four consequences follow.

  1. 01

    Volume is not the constraint, so buying more of it changes nothing.

    If you could already publish twelve pieces a month and it was not working, forty will not work either. It will work less well, because the same amount of attention is now spread across more pages and the average quality of what represents you goes down.

  2. 02

    Nobody can promise this passes as human, and it is the wrong goal.

    Detection tools are unreliable in both directions and I would not build a service on beating them. The goal is not undetectable output, it is output somebody with standing was willing to attach their name to. Those are different targets and only one of them is stable.

  3. 03

    It cannot learn a voice you have not established.

    Brand voice work needs source material: things you have already written, recorded or argued. Given fifty pieces of yours, a model will hold your register convincingly. Given a two-line brand adjective list, it produces the same house style as everybody else who did that.

  4. 04

    Speed makes bad strategy fail faster and more visibly.

    This is the uncomfortable one. A weak plan executed at four times the pace produces four times the wrong pages, on your domain, at once. Every efficiency gain here is a multiplier on the quality of the decision in front of it.

None of this argues against using the tooling. It argues against buying it as the service, which is what the category is currently structured to sell you.

The work

Six workstreams, and the machine touches three.

Notice which three. The drafting is the cheapest step in this list and the only one the category advertises.

  1. 01

    Weeks 1–2

    Decide what deserves to exist

    The expensive half, and the half nobody is selling. Which questions your buyers actually ask, which of them you can answer better than anyone, and which topics you are declining. An ai content strategy that starts at the generation step has skipped the only part that was ever scarce.

    Topic decisions18 of 140 kept
    We can answer this betterkeep
    Buyer actually asks itkeep
    Everyone has published itdecline
    We have no evidencedecline

    The generator is not switched on until this list exists.

  2. 02

    Ongoing

    Get the source material a model cannot invent

    Recorded customer calls, founder interviews, support tickets, your own operational data. This is the input that makes AI content creation produce something specific rather than a competent average, and collecting it is the part that actually takes effort.

    Brief inputsper piece
    Customer or founder recordingrequired
    A number we can stand behindrequired
    A named examplerequired
    Competitor blog as a sourcebanned

    A brief missing the first row does not get drafted. It gets rescheduled.

  3. 03

    Ongoing

    Draft fast, on rails

    This is where the ai writing tools earn their place: a first draft against a real brief, in minutes rather than days. It is deliberately the cheapest and least discussed step in the whole pipeline, because it is now the least difficult one.

    Brief to drafthours → minutes

    Before

    Now

    The saving is real and it is spent on the research and the edit, not banked.

  4. 04

    Ongoing

    Edit like it is going out under your name

    Because it is. Human editing here means checking every claim against its source, cutting the fluent paragraphs that say nothing, restoring the specifics a model smooths away, and being willing to send it back. Roughly a third of drafts do not survive this step.

    Edit pass~1 in 3 killed
    Every factual claim sourcedrequired
    Fluent but empty paragraphscut
    Specifics restoredrequired
    Draft sent back or droppedoften

    The kill rate is a feature. A pipeline that publishes everything has no edit step.

  5. 05

    Ongoing

    Make one piece into eight

    The genuinely transformative use, and the one most programmes underuse. An article becomes founder posts, a newsletter section, a community answer and a landing-page section, each properly adapted rather than truncated. This is where content at scale stops being a euphemism.

    One interview8 surfaces
    Long-form articleshipped
    Founder posts4 drafts
    Newsletter sectionqueued
    Answer-shaped passagequeued

    Same recording, eight places. The recording was the expensive part.

  6. 06

    Month 2+

    Watch quality, not throughput

    The reporting deliberately does not lead with how much was published. It leads with whether the pieces earn attention, get cited, and hold up. Content quality is measurable if you choose measures that can go down, and throughput cannot.

    Monthly readleading · business

    Leading

    Pieces earning attention, citations, edit kill rate

    Business

    Content-influenced signups, demo requests

    Shape of a report, not a forecast. Published count is deliberately absent.

If the question is your whole marketing stack rather than the publishing pipeline, that is a different page.

The alternatives

Four ways to buy this. Three are priced on volume.

That pricing is the tell. If the invoice scales with pieces produced, nobody in the arrangement is paid to suggest publishing less, and publishing less is frequently the correct advice.

busyless
AI content agency
A writing tool
Freelance writers
What you are sold
Judgement about what to make, with drafting as cheap infrastructure behind it
Output volume, usually priced per piece
Seats and a credit balance
Hours, and a slower calendar
Where the raw material comes from
Recorded customer and founder conversations, plus your own data
A keyword tool and the existing top ten results
Whatever you paste in
A brief you write
Who is accountable for a false claim
Me, by name, and every claim is checked against a source
Contractually, you
You
The writer, usually
Willingness to publish less
About a third of drafts are killed at the edit
Against the pricing model
Not a concept
Yes, but it costs you the fee anyway
What is reported
Whether pieces earn attention and hold up, never throughput
Pieces delivered
Words generated
Articles delivered
Cost shape
$2,500 diagnosis, then from $5,000/mo for the judgement
$50 to $500 per piece, scaling with volume
$20 to $500/mo, plus all of your time
$300 to $1,500 per article

If you already know exactly which fifty pages you need and why, a per-piece supplier or ai seo content vendor will be cheaper and perfectly adequate. The judgement layer is only worth paying for when the list itself is the uncertain part.

Engagement

Buy the decision. The drafting comes free with it.

Priced on judgement rather than throughput, deliberately, because a fee that rises with volume creates an incentive I do not want to have when the right answer is to publish four things instead of twenty.

Start here

Discovery

$2,500one-off · two weeks

What your buyers actually ask, which of it you can answer better than anyone, what to decline, and an honest read of whether more content is your problem at all. Frequently it is not.

  • The topic list, with the declines stated
  • An audit of what you have already published, including what to delete
  • Where your evidence and recordings will come from
  • A verdict on whether volume is actually your constraint
Book the diagnosis

Most common

Ownership

From $5,000per month

The whole pipeline owned: the decisions, the source material, drafting on rails, an edit with the authority to reject, repurposing across every surface, and reporting that never leads with how much was published.

  • Topic decisions made from customer evidence, not a keyword export
  • Every claim checked against a source before it ships
  • About a third of drafts killed at the edit, on purpose
  • One recording repurposed across eight surfaces
Talk it through

The strategy half can also be bought on its own as a fixed-fee piece of work, if what you need is the list rather than the pipeline.

Let's build

One call. Real plan, not a pitch.

30 minutes. We talk about what's already working, who owns content today, and whether a fractional Head of Content is actually the right move. If it isn't, I'll say so.

Direct calendar

Book a 30-min intro call

Loading scheduler…

Calendar busy?

Send a note instead.

One sentence on the bottleneck. I'll reply within 24h with a sharper next step.

Or send a note

FAQ

Questions, ahead of time.

  • Do you actually use AI to write our content?+
    Yes, for research, first drafts and repurposing, and I would rather say so plainly than pretend otherwise. It never decides what gets made, never supplies an opinion, and never publishes. Every piece runs from a brief built on a real recording, through a human edit that checks each claim, to somebody deciding it is good enough. The tooling is infrastructure, not the product.
  • Will Google penalise AI generated content?+
    Google's stated position is that it rewards helpful content regardless of how it was produced, and penalises content made primarily to game rankings. That is a distinction about intent and quality rather than authorship, which is why the pipeline here is built around the decision and the edit rather than the generation. The March 2024 spam policies target scaled production without regard for whether anyone is helped.
  • How do you keep it sounding like us and not like everything else?+
    By feeding it a large amount of your existing material, and by keeping the specifics that a model naturally smooths away. Fluency is the default failure mode: the draft reads well and says nothing anybody could disagree with. The edit exists mainly to put back the numbers, names and admissions that make a piece recognisably yours.
  • Can we just publish more, faster?+
    You can, and it is usually the wrong instinct. If twelve pieces a month were not producing results, forty will not either, and the extra thirty-two dilute the attention your good work would have received. The speed is better spent on more research per piece and on repurposing what already works.
  • What happens to our existing writers?+
    In most teams they stop drafting from scratch and start doing the two things that got more valuable: running the customer conversations and editing with the authority to reject. That is a more senior job than the one they had, and the teams that handle this well tend to keep everybody and publish rather less.
  • How is this different from your AI marketing page?+
    This page is about the content pipeline specifically: what a model may draft and where the line is. The other one is about the whole marketing function, where the recurring problem is not capability but adoption, and most of what was bought is sitting unused. Different question, and worth reading if your issue is a stack rather than a publishing process.
  • Do you disclose AI involvement to readers?+
    I do not add a label to every article, because the meaningful claim is that a named human takes responsibility for the content, and that is true here. What I will not do is market the work as human-crafted while a model does the drafting, which is the actual dishonesty in this category. Anything I say about the process on this site is accurate.
  • What does this cost compared with buying content per piece?+
    Per piece it is more expensive, and that comparison is the trap. A two-week Discovery is $2,500 and ongoing ownership starts at $5,000 a month for the judgement layer, with production capacity priced separately. If what you need is fifty cheap pages and you already know they are the right fifty, a per-piece supplier is genuinely the better buy.