Most AI content sounds like every other AI content because every other team is using the same prompt window. We build the layer that comes before the prompt: the research that gives each piece a real angle, the voice training that keeps the writing yours, and the editorial gate that catches what the model can't see. One executive. Twenty posts a month. Her audience still thinks she's writing them.
01 / What's at stake
The output ships fast. It reads like everyone else's. Three weeks in, your LinkedIn looks like a competitor's, the case study reads like the one you saw on a vendor's site last Tuesday, and the GBP posts across your client portfolio have started to use the same five adjectives. None of it is wrong. None of it is yours.
The reason isn't the model. The reason is that the model is doing all of the work: the research, the angle, the voice, the decision about what to leave out. A senior writer makes a hundred small choices per piece before a single word goes on the page: which angle to take from the three the topic invites, which two recent client conversations to mention, which sentence to cut because it sounds like everyone else. ChatGPT makes none of those choices unless you give it the inputs to make them. Most teams don't, because building those inputs is the actual work.
The cost isn't a bad LinkedIn post. The cost is that the time your team spent on production was supposed to be time spent on strategy, and now it's gone. There's a better version of this. We've built it three times.
The output ships fast. It reads like everyone else's.
02 / What we build
"AI marketing" without these five layers is a prompt window. With them, it's a working department.
01 · Research
Every piece starts with research that's specific to the topic, the audience, and the week. Not a generic prompt rerun with the topic word swapped. We pull from your CRM notes, the client conversations your team logged on Friday, the industry sources your audience actually reads, and the data your competitors are quoting. The angle is decided before the writing starts. The model never has to invent context.
02 · Voice
Most AI tools train on a "brand voice" doc that says things like "warm but professional." That isn't voice. Voice is sentence rhythm, specific word choices, the joke a particular executive makes when she's tired, the phrases she'd never use. We build the voice layer from twenty to fifty real pieces by the actual author, calibrated until her team can't tell which drafts she wrote and which the system drafted for her review.
03 · Editorial
The system drafts. The author reviews. The dashboard shows the week's drafts side by side with the research notes that produced them, the voice-calibration flags (an example: "this sentence reads 14% off your usual sentence length"), and a one-click approve. What used to be six hours of writing becomes five minutes of editing. The author still owns every word that ships.
04 · Publishing
Approved pieces route to the right channel automatically: LinkedIn, blog, GBP post, programmatic SEO page, email sequence. Each channel has its own formatting rules and its own timing window. The author approves once. The system handles the rest, including the failed-post retries and the "did this actually go live" notifications.
05 · Performance
Once content is live, the system tracks which pieces drew which kind of response, surfaces patterns across two months, and feeds those patterns back into the research step for the next round. The system gets sharper as it runs. Not stale.
03 / How we work
01 · Discovery
The first week is shaped around your existing content function: what your team writes now, who writes it, where it stalls, what gets cut for time. No deck. No prescribed workflow. We need to see how the work actually moves through your shop before we redesign any of it.
Week 1 · No fee · You keep the notes02 · Audit and scoping
Out of discovery comes a scope: which content type the system handles first (LinkedIn, blog, GBP, programmatic SEO, email, usually one to start), which author the voice layer is built around, which research sources feed the pipeline, where the human review gate lives. You sign off before any building starts.
Week 2 · Fixed-fee scope · Proposal in 48 hours03 · Build
We build the research pipeline, the voice layer, the editorial dashboard, the publishing automation, and the reporting loop in parallel. Every Friday is a demo. You see what's working, what's not, what the next week looks like. Voice calibration takes the longest. On the LinkedIn Engine, it took two of the three build weeks. We don't ship until the author can't tell her drafts from the system's.
Weeks 3-5 · Weekly demos · No fixed-scope surprises
04 · Handoff
Week six is handoff. One-hour training for the author on the review dashboard. One-hour training for the operator on the research and publishing layers. A runbook covering where every part lives, how to pause the system, how to update the voice layer when the author's style drifts, how to debug a failed post.
Week 6 · Training and runbook · Your team owns it from here
05 · Support
If anything breaks in the first thirty days, we fix it on us. The system is ours until your team is comfortable owning it. After that, most clients move into a monthly review retainer: an hour every month to recalibrate voice, refresh research sources, and tune the reporting loop. A few take the playbook in-house and run it themselves. Both are fine.
Days 1-30 · Backstop included · Monthly review optional04 / Proof
Six hours of weekly content work replaced by a five-minute review. Twenty posts a month, in voice, on schedule. Her audience still thinks she's writing them. Because she still is. We built the system that does the reading and drafting for her.
Six hundred and eleven location pages shipped in one month, schema-tagged, indexed inside forty-eight hours of approval. Not a template-swap. Each page written from local context. Twelve times faster than the manual estimate, and Google didn't penalize a single page.
05 / Questions
Voice isn't a brand doc that says "warm but professional." Voice is the sentence rhythms, the specific word choices, and the things a particular author would never write. We build the voice layer from twenty to fifty real pieces by the actual author and calibrate until her team can't tell her drafts from the system's. The LinkedIn Content Engine preserved one executive's voice across twenty posts a month. Her audience still thinks she's writing them.
Only if you point ChatGPT at a topic and copy-paste the output. The Cashmax site shipped six hundred and eleven location pages in one month, none of them template-swapped, and Google indexed every page inside forty-eight hours of approval. The difference is research depth. Google penalizes duplicate patterns, not AI-assisted writing. Our system produces unique angles per piece by design.
Those are tools. We build systems. A tool is a prompt window. You bring the research, the angle, the voice. A system includes the research pipeline, the voice training, the editorial gate, and the publishing automation. The work product is what your team would have produced if they had eight more hours a day, not what ChatGPT produces in two minutes.
If your monthly content load is one or two pieces a week from one channel, hire the freelancer. We will tell you that on the scoping call. We come in when the volume breaks freelance math: twenty LinkedIn posts plus four blog pieces plus monthly GBP posts across forty locations plus programmatic SEO across six hundred pages. No single freelancer covers that surface area, and managing five freelancers becomes the bottleneck you were trying to solve. The other asymmetry is durability. A freelancer leaves with her notebook. The system stays. The voice-calibration work, the research pipeline, the editorial gate, the publishing automation. Year two of working with us costs less than year one. Year two with a freelancer costs the same, plus the cost of the next freelancer when this one moves on.
About six weeks from kickoff to a system your team can run without us. Week one is discovery. Week two is scoping. Weeks three to five are the build, with a demo every Friday. Week six is handoff and training. Voice calibration is usually the longest part of the build. We don't ship until the author can't tell her drafts from the system's.
Then we route you to a one-week voice doc sprint first. The voice layer is trained on real writing by a real author, and if the author hasn't been writing yet, we have to build the source material before we can build the system. The voice sprint is scoped and priced separately. Honest cost: a week and a fixed fee. Honest reason: a voice layer trained on guesswork ships content that sounds like guesswork.
No. The author reviews every piece before it ships, in the dashboard, in about five minutes a week. Nothing publishes without a human approval. The system handles the reading, the drafting, the formatting, and the publishing. The author still owns every word that goes live. If you want zero-touch autonomous publishing, this isn't the right service.
Next step
The free assessment asks ten questions about your current content function and where the time goes. It tells you whether a marketing system would actually move the number you care about, or whether the bottleneck is somewhere else entirely. No call. No pitch. The report is yours either way.
