Content engineering: an AI content system that still sounds like you
A content engineer builds the system that makes the content, instead of writing each piece by hand. I build that system for B2B teams who want content to feed their pipeline, and I design it to keep the voice of the person whose name is on the posts.
If your team says "We need a system, not just random posts", the short answer is this: five files before the first post, one machine that looks for ideas, one that writes from an interview with the person, and checks on every draft, with AI in the middle of the process and never at the start. The rest of this page shows each part: what the job is, the five files, the voice profile, where AI sits, and how one system holds several voices.
What does a content engineer do?
A content engineer designs the system that produces content, then runs it and keeps improving it. The phrase has two meanings, and both are in use today.
The older meaning is about structure. Wikipedia describes content engineering as "the practice of organizing the shape and structure of content by deploying content and metadata models". In plain words: models, tags and schemas, so a piece of content is stored once and reused in many places, and machines can read it.
The newer meaning is about AI. Ahrefs puts it like this: "A content engineer designs the systems that produce content, rather than producing content piece by piece." That is the person who builds the line where research, drafting, editing and publishing run with AI in the middle.
My work is the newer meaning, with one condition: the system has to keep a human voice. At Frontal, where I'm Content Lead, this is the practice I run, and we call it Content Engineering.
From my post of 2 Jun 2026:
"Copywriter is becoming copy engineer. Content marketer is becoming growth operator. The job titles just haven't caught up to what people are already hiring for."
Is content engineering a good career?
I think yes, for someone who can write and think in systems at the same time. A founder sent me this line, which I shared in my post of 2 Jun 2026: "So hard to find someone who's good at content AND can think in systems." My answer there, to the people who can do both: "There are very few of you. That's the whole opportunity." I have no salary or hiring numbers to show you, so this is my view from the work, not a market study.
"We need a system, not just random posts"
A content system is the set of files, steps and checks that turns what a founder knows into posts in their own voice, week after week, without a blank page on Monday. Random posts are what happens without it: every week starts from zero, and the quality depends on how the week went.
In my post of 9 Apr 2026 I describe three levels:
- Content as a task. "No plan. Every week starts with a blank page and the question "what should I post?""
- Content as a system. Pillars, a posting calendar, maybe some templates. It runs on discipline, and when the founder gets busy it goes generic, because "a content calendar without a voice is organized mediocrity."
- Content as a machine. A voice profile, an AI layer that finds ideas in your real conversations and your real data, and a human review step where you edit instead of creating from scratch.
"The jump from 1 to 2 is discipline. The jump from 2 to 3 is infrastructure."
What I build is the third level. Its anatomy is public, and I'm happy to show it. The rules and prompts inside it stay private.
- Five files before the first post. The voice profile first, then the ICP, the content pillars, a grading profile and brand guidelines (below).
- Two machines. One looks for ideas. The other writes, and it starts with an interview, where the person adds the story only they know (below).
- Three checks read every draft before it reaches the person.
- A learning loop. From my post of 22 Jun 2026: "the system feeds on its own numbers. It remembers what worked, drops what faded, and gets a little sharper every day."
Check your own setup first
If you run content with AI today and something feels off, look at four things before you change anything else. From my post of 13 May 2026: "Voice, ICP, feedback loop, ideation system." When one of them is missing, the output is fast and clean for a few weeks, then every post starts to open the same way, and the audience notices before you do.
What documents does a content team need before writing for a founder?
Five files: a voice profile, an ICP, content pillars, a grading profile and brand guidelines. The writing starts after them.
| File | What it holds |
|---|---|
| Voice profile | How the person talks and thinks: the words they use, the words they would never use, the opinions they would defend. Built from a conversation (next section). |
| ICP | Who the posts are for, in tiers, with each buyer's problem written the way that buyer would say it. |
| Content pillars | 5 or 6 themes the person already holds a position on, in their own language. Every post traces back to one of them. |
| Grading profile | What a good post means for this one person, written down, so every draft is judged against the same standard. |
| Brand guidelines | For the visuals, not the writing. |
The ICP comes before any topic idea. From my post of 2 Jun 2026: "The topic is downstream of the ICP, not the other way around."
Most of the work on an account happens under the post, in these files. From my post of 8 May 2026, the part nobody sees is "building + maintaining the voice profile that keeps it from sounding like everyone else (and AI)".
How do I create a voice profile for a founder's content?
In a real conversation, never with a form.
"A Google Form asking "what is your tone?" has never once been answered truthfully."
A form fails because nobody watches their own voice. From my post of 9 Jul 2026: "So when a form asks, you guess. And the guess is the person you wish you were, not the person you are."
How I build one:
- A conversation, one question at a time. What the person cares about, what annoys them in their field, the opinion they would defend. After a while they forget they are being studied, and the real voice comes out.
- Speech before old posts. Old posts are not a clean sample. From the same post: "Because your old content is already polluted." Templates and a bit of AI are already in them, and feeding them in first teaches the machine "to copy the makeup, not the face."
- The words the person rejects go in too. A list of what they would never say is as useful as what they would.
- Two lengths. In my post of 9 Jul 2026 I gave away two versions. The short one takes about 20 minutes: "Enough for a twin that works." The deep one is about 100 questions, "For when your voice is your whole brand and you want it near perfect. An hour or two."
- It keeps growing. New call transcripts keep updating it, so the profile follows the person as their thinking moves.
What a voice profile gives you, from my post of 30 Mar 2026: "It captures how you think, what you believe, and what makes your perspective different from the 50 other people posting about the same topic."
How do I make AI written LinkedIn posts not sound like AI?
Give the AI something only you could say before it writes, and keep the judgment for yourself after it writes. Part of the AI sound comes from what goes in: empty input, and prompts that ask for nothing in particular.
From my post of 20 Aug 2026:
"AI is not necessarily writing badly, but part of the blame lies in the prompts."
"Make it punchier." "Add a CTA." With prompts like these, as I wrote in the same post, "You have essentially asked it to produce the statistical middle of everything ever written."
The order I work in:
- Experience
- Idea
- Thinking
- AI
- Editing
- Judgment
- Publish
"AI belongs in the middle as a production step. It can never substitute for having something to say."
What to give it first, from my post of 1 Jun 2026: "Your real numbers. The opinion you'd defend out loud, the ideas that make you feel itchy. The story from your ordinary day that AI couldn't invent (yet)."
"AI content is just as good now"
It is as clean, and that is the problem. When every post is clean, clean stops meaning anything.
"Polish used to be the proof you were worth trusting. Now, polish is the tell."
AI did change something real. From my post of 20 Aug 2026: "So I don't think AI has lowered the bar for good content. It has lowered the cost of reaching the bar." The time it saves is meant for the thinking and the editing, not for more volume.
How do I repurpose one founder interview into multiple LinkedIn posts?
Keep everything the interview gives you as raw material, then let each idea become several angles, and each angle a different post. Treat one interview as material for many posts, not as one.
In the system, the interview is where the person adds the story only they know. After it, nothing is thrown away. From my post of 24 Aug 2026, the material I keep is "client questions, sales objections, things from calls, interesting examples, mistakes, half-formed arguments, screenshots and voice notes." And: "I don't decide whether something is "content" when I capture it. I just save it."
Then one idea opens into several posts. From the same post, each idea can become "the argument for it, the argument against it, the mistake it corrects, the example that proves it, or the experience that complicates it."
The interview gives the perspective, and the market gives the subject. As I put it in my post of 24 Aug 2026: "An external signal gives me a subject. My internal archive gives me a perspective."
When one of those posts works, don't post it again. Take the lesson and put it into a new idea for the same people. From my post of 15 Jul 2026: "In ads, you multiply the asset. In content, you multiply the lesson."
How do I scale LinkedIn content for several founders with AI without losing their voice?
One system, one voice profile per person, and AI only in the middle. That is an AI-assisted system, and it is a different thing from AI-generated posts, where a model writes and nobody's thinking is in the post.
The opposite choice exists too. In its own account of its marketing, published on 15 Jul 2026, the B2B software company Clay writes: "None of this comes from a PR team. The executives write their own posts, sometimes with a teammate sharpening a weak hook, and none of it is AI-generated."
The system I build is AI-assisted, and this is where the line sits:
| Step | Who does it |
|---|---|
| The idea, the story, the opinion | The person: in the interview, and from their own week |
| The voice | A voice profile for each person, built from a conversation with them |
| The draft | AI, in the middle, working from that person's own files |
| The checks | Three checks read every draft before it reaches the person |
"A tool doesn't make content soulless. Outsourcing your thinking does."
Several voices in one system is a skill before it is a tool. From my post of 2 Jun 2026, the job is "Holding a real opinion in their name, across 100 posts, without sliding back into the AI default or leaking your own voice in". The system helps, and the design matters more than the model. From my post of 17 Apr 2026: "The way you design the workflow matters more than the model you use."
Where to go next
- Content to drive B2B pipeline: how content becomes a go-to-market channel, and how to tell whether it brings in pipeline.
- LinkedIn programs for teams: the founder, executives and employees posting together.
- To have your team trained to run its own system, or to have it run for you, Frontal's two services have their own pages: Done With You and Done For You.