The content writing system produces articles, guides and pages that are researched, structured to be quoted, and written in the right voice for each brand. It runs research, planning, a writer pass and a separate editor pass, then voice and pre-publish checks. Every piece is read and signed off by me before it’s used. Built in-house.
What it does
The content writing system turns a brief into a finished, publish-ready piece of content. It analyses what already ranks and what those pages miss, researches primary sources, plans an outline with word budgets and source assignments, drafts the piece with one pass focused on depth and warmth and a second focused on accuracy and concision, then removes AI writing patterns and produces the meta data, schema and HTML.
How it works
| Stage | What happens |
|---|---|
| 1. Brief and gap analysis | The search, the reader, the target page, and what the pages already ranking cover and miss |
| 2. Research | Primary sources first, such as platform documentation, with every claim recorded in a source ledger with its date and location |
| 3. Outline | Sections, word budgets, answer passages and the sources each section uses, approved before drafting |
| 4. Writer and editor | A writer pass aiming for depth and usefulness, then an editor pass checking sources, cutting vague lines and enforcing the rules |
| 5. Polish | A voice pass to remove AI writing patterns, then meta titles, descriptions, schema, HTML and a consistency report |
Set up per brand
Each brand gets its own version of the system, locked to its voice, audience, spelling, facts and rules. That includes things like US or UK spelling, claims a brand must never make, products it doesn’t stock, and competitors it never names. Versions are set up for client brands, for River Bend and for this site.
A shared, one-size setup gets these mixed up. Separate versions keep a brand consistent across every piece.
What the system never does
- Invent experience. Examples, results and learnings come from real projects. Where none exists, the draft marks a gap for a human to fill.
- Cite without a source. Every factual claim traces to an entry in the source ledger, and undated statistics are cut.
- Publish on its own. Finished pieces are delivered for review, or into a client’s review folder, never pushed live without approval.
Example output
Every piece arrives as a folder containing:
- the final article in Markdown and ready-to-paste HTML
- meta title and description, checked for length
- schema, including Article or BlogPosting and FAQPage
- the source ledger
- a consistency report listing every check and anything still needing input
What I review
I read every piece in full. I check the sources behind key claims, add or confirm first-hand examples, make sure the advice is right for the business and its customers, and edit anything that doesn’t sound like a person talking. Nothing is marked ready until every gap for my input is filled.
Limitations
The system is only as good as its brief and sources. It can’t supply experience it hasn’t been given, which is why first-hand examples come from real project files or from me. It also doesn’t replace judgement about what’s worth writing, which comes from the content strategy.
Where it’s been used
It produced the guides and collection page rewrites in the FalseEyelashes.co.uk case study, the education rewrites and content pipeline for Kwiat, and the guides behind River Bend.
Services it powers
Blog and article writing, product and category page copy, content pipelines and generative AI for marketing. See the rest of the tools on the tech stack page, and how AI is used across the work in the AI use statement.
FAQ
Does the content writing system publish content automatically?
No. The system researches, outlines, drafts and checks content, but nothing is published without me reading, editing and approving it. Where a client has a publishing pipeline, finished pieces are delivered into their review folder for their own team to sign off, not pushed live.
Why have a separate writing system for each brand?
Every brand has its own voice, audience, rules and facts, and a single shared setup gets those mixed up. A separate version for each brand keeps its tone, spelling, banned claims and product rules locked in, so a New York jeweller’s guides use US spelling and a UK retailer’s never do.
How do you stop AI-written content sounding generic?
By giving the system real material and checking the output hard. Briefs include first-hand examples and sources, a writer pass aims for depth, an editor pass cuts anything vague, and a voice check removes the phrasing that makes AI writing easy to spot. Anything without a real example gets a gap marked for me.