If you manage any sort of expert-backed content for your organization, you’ve probably juggled this tradeoff many times: expert content is what actually earns trust and sets your organization apart from everyone producing the same generic, keyword-stuffed slop, but it’s also the slowest to produce. Subject matter experts (SMEs) — be they clinicians, engineers, lawyers, or anyone explaining their field — have limited time to write or review content, yet understandably high standards for what gets published under their name.
So most teams settle, either for less content or more generic content that doesn't need as much of the expert's attention. In practice, that often means defaulting to the broad topics everyone in the industry already covers — the “what is X” explainers that establish baseline credibility but don’t require much expert nuance to write.
“There’s so much friction in that old model that, a lot of times, the marketer just cuts the expert out and starts writing these very safe, basic posts,” explains Chadwick Savage, founder of The Rabble Group, an agency that focuses on producing content for high-trust, high-expertise industries.
Those basic posts are also the least effective. For one, they’re the most competitive ground in any industry, with your biggest competitors already writing the same explainers targeting the same head terms. But they also matter less in a world where AI search adds context about your customer to each search, even if they don’t type it, so the same query now requires a far more specific answer.
“Say ChatGPT goes out and finds six ‘what is IOP therapy’ pages, but yours is the only one that's speaking specifically to that parent,” explains Chadwick. “Being able to write to those long-tail keywords allows you to show up in that AI search better.”
Reducing friction to improve specificity
Chadwick thinks he’s found the answer in a robust content engine he built to help him reduce friction in creating expert-backed content. After creating an SEO brief and content roadmap for a client, Chadwick feeds that information into a Claude workflow that lightens the load of weekly planning and drafting. Every Sunday, a scheduled task scans industry news, checks it against the client's content roadmap, and decides what to write that week, bumping up any ideas that seem especially timely and then sending initial drafts for review.
But perhaps the most important piece is that the engine has a constantly-updating, highly detailed voice profile for each SME to help it create content that needs fewer and fewer edits over time.
Chadwick remembers one clinician he worked with who initially tore apart an article he was reviewing. “He marked the first draft up for four hours, told us it was awful,” says Chadwick.
Chadwick took the feedback from that initial draft, asked Claude to create a voice profile for the clinician based on his edits, and then rewrote the other ten drafts they had in progress using that profile.
“He was done reviewing all ten in three hours, just marking small changes,” says Chadwick.
Yes, the time saved is nice, especially when engaging with experts in your organization whose primary job isn’t marketing. But, perhaps more importantly, the reduced friction allows Chadwick’s clients to publish more expert-backed content, therefore opening space to target more specific topics and keywords.
“The reduction in friction allows us to speak to things we actually care about, instead of just defining the terms of our industry,” he explains. The results? Between April and August, one of his behavioral health clients jumped from 32,000 search impressions and 500 clicks to 84,000 impressions and 1,200 clicks.
Now, Chadwick has built that improvement loop into his content engine so that it keeps getting better without him — or the experts he’s writing for — ever having to articulate that feedback to AI.
How to create a voice feedback loop for SME content
1. Start with a detailed voice profile
The foundation of this whole process is a detailed voice profile for the AI to reference when drafting articles, both for your organization and separate profiles for any individual SMEs you may be writing for.
Detailed is the key word here. Most voice profiles Chadwick sees clients write on their own are too similar to everyone else’s. “I'm professional, empathetic, authoritative. I'm a thought leader, knowledgeable yet approachable. All of that is vague garbage,” he explains. “A voice profile earns its keep only when it is specific enough that it could not belong to anyone else.”
That’s why his voice profile template covers 16 different aspects of how you write, from sentence rhythm and vocab dos and don’ts to signature rhetorical moves to how you like to build an argument and present data.
Voice-profile template, given away by The Rabble Group (rabblegroup.com) as part of "How to build a content engine."
This is the scaffold, not our filled-in voice. The fastest way to fill it in is with your AI assistant, not by hand: give it a handful of samples of how you actually write, have it draft each section from them, then read what it produced and correct anything that does not sound like you. You are the judge, not the typist. This one file does more to keep an AI-assisted engine sounding human than anything else in the system. -->
<!-- How to fill this in: paste 3-5 samples of your real writing (emails, posts, pages you are proud of), ask your assistant to draft each section from them, then push back on anything that is off until it sounds like you. -->
---
version: 1
last_updated: YYYY-MM-DD
---
# Voice profile: [YOUR BRAND]
A structured description of how your brand sounds and how it must never sound. Every draft the engine writes passes through this file, so it is worth getting right. You do not have to write it from a blank page: paste in samples of your real writing and let your assistant draft each section from them, then correct what is off. You bring the writing and the judgment; it brings the first draft. Vague answers produce vague content; specific answers are what make the writing sound like a person who knows the subject.
## 1. One-line identity
Who is speaking, in a sentence. Not a tagline - a description a stranger could use to recognize your voice. Example shape: "A senior operator talking plainly to a busy owner making a high-stakes decision."
## 2. Who you are talking to
The actual reader. Their job, their stress, what they already know, what they are skeptical of, the moment they are in when they read you. Write for one person, not a segment.
## 3. The core tension
The thing your voice is FOR and the thing it is AGAINST. This is the spine. Example: "For honest, compounding work. Against the hype an entire industry runs on."
## 4. Register and tone
Where you sit on the dial: formal or casual, warm or dry, punchy or measured. Name it, then name where NOT to go ("never brash, never salesy"). When in doubt, which way do you lean?
## 5. Sentence rhythm and mechanics
Short or long sentences? Verbs or nouns? Paragraph length? Read a paragraph of your best writing aloud and describe what you hear.
## 6. Person
First person singular ("I"), plural ("we"), or a mix, and when. How do you address the reader ("you")?
## 7. Signature moves
The two or three rhetorical moves that are distinctly yours. Example: "Name the hype, then offer the honest alternative." "Show, do not claim." "Translate every piece of jargon into plain language."
## 8. How you open
Your rule for first lines. What you lead with (a specific claim, the reader's frustration) and what you never open with (mission statements, "in today's landscape," company history).
## 9. How you build an argument
The shape of a section. Example: name the problem -> the honest contrast -> the proof -> the next step. What has to be present for a section to earn its place?
## 10. How you use data and proof
Named sources or vague authority? How many numbers, and how do you introduce them? The hard rule most brands need: never claim a figure you do not have.
## 11. How you deliver hard truths
When you have to tell the reader something they will not like, how do you do it? Bluntly, gently, with a caveat? Your voice is defined as much by this as by anything.
## 12. Vocabulary you use
Words and phrases that are characteristically yours. The house terms. The way you refer to your own work.
## 13. Banned words and phrases
The hard list. Every buzzword, hype word, and cliche the engine must never produce. Be specific and long - this is the most-used part of the file. (Start from the anti-patterns document and add your own.)
## 14. The irreverence dial
What you are unimpressed by and plainspoken about (usually your industry, jargon, hype) versus what you are always serious and respectful about (usually the people your customers serve). Getting this line wrong is the fastest way to sound off-brand.
## 15. Formatting rules
The mechanical house style. Example: no em dashes (use spaced hyphens), no curly quotes, no emoji, sentence case headings. The things a reviewer should be able to grep for.
## 16. Channel overlays
How the voice shifts by channel. The blog, the newsletter, and social are the same brand in different registers (a newsletter is shorter and more direct than a pillar page; social is more personal). Note the deltas here, or keep a separate short overlay file per channel that inherits this one.
Given SMEs may struggle to define these things about themselves, Chadwick suggests dropping 3-5 samples of their writing into a chatbot, asking it to draft an initial voice profile based on that writing, and then working together to fix anything that feels off.
2. Have AI draft, but keep the human in the loop
With a voice profile in hand, Chadwick has AI create initial drafts of articles for the SMEs he works with, including a few additional steps to lighten the load for the SME as much as possible.
First, before the draft even hits his desk, Claude runs them through an “anti-patterns” checklist that looks out for AI tells, banned jargon, unverified citations, or anything else that is against brand guidelines.
Then, Claude sends the draft to Chadwick, and he reviews and makes adjustments or asks for changes.
“We're working together at this point,” he says. “It's not running autonomously.”
From there, the SME gets final say. Chadwick sends the draft along, and the expert marks it up directly, whether that's tracked changes in a Word doc or edits in a CMS draft. The goal is never to remove them from the process, but to make sure they’re spending time doing what only they can do — judge if the piece is accurate and useful — rather than having to tweak the same vocabulary or sentence structure time and time again.
"What we want to do is reduce the friction for a clinician writing it, not cut them out,” says Chadwick.
3. Bake feedback into the AI workflow
Finally, Chadwick has Claude compare the final published piece with the draft it initially wrote, and make suggestions for how to change the voice profile to allow it to produce more aligned drafts in the future.
This, says Chadwick, is what prevents this from being an AI content mill that churns out slop.
“A mill produces the same quality forever. An engine improves, because it watches what happened to the last thing it made and adjusts,” he explains.
He has built this as two skill files that run automatically as part of his content workflow: one that reads what content has been published on a client’s website and then one that compares published articles to the AI’s original draft, notices what’s different, and suggests updates to the voice profile based on patterns it sees.
But, if you aren’t using Chadwick’s full workflow, you could also manually drop in the original draft, final draft, and current voice profile into your AI of choice, and ask it to find the changes and suggest voice profile improvements.
Sample suggestions for voice profile improvements
This is another place where Chadwick emphasizes that human judgement is necessary. When he was building this workflow, Claude offered to update the voice profile automatically based on the SME’s edits. “I specifically told it no,” he says. He reads through the suggestions, and tells Claude which to implement.
While you can download Chadwick’s full content engine here, you don’t need the full stack to start, and can even integrate this into your own workflow. Build one detailed voice profile, let your expert mark up a draft, and feed those edits back in before writing the next one.
And remember: The goal is never removing your SMEs from the picture — it’s just making their jobs easier.
“The human judgment is not replaced by the loop,” emphasizes Chad. “It is recorded by it, and reused.”
