Claude and ChatGPT configured to follow your brand
Your team already drafts with AI tools, and the output drifts toward everyone else’s voice. We write your brand in a form each tool follows and add a checker for the signs of machine-written copy. We also run your first two experiments with the new set-up.

How we configure the tools
We run the same rules on every draft we produce ourselves.
Five steps from brand file to published work
AI set-up sprintBrand file
We write your position, voice, banned phrases and visual rules as instructions.
One sourceTools
We load the file into Claude projects and skills, custom GPTs, Lovable knowledge and Figma libraries.
Per-tool set-upDraft
Your team drafts as before with the brand already in the context.
DraftsCheck
We scan every draft for the six signs before an editor reads it.
Flags per draftPublish
Your editors spend their time on the concept instead of the cleanup.
Published workBefore and after our editor
Our checker marks the signs of machine-written copy. Our editors rewrite the line around a fact the buyer can check.
Fictional example. Five signs flagged in the draft, none in the edit.
Six signs of machine-written copy we check for
We wrote these rules after reviewing thousands of generated drafts. Our own copy passes them before we publish anything.
Dashes used as pauses
An em dash where a comma or a full stop belongs. Readers spot it first.
Runs of three-word sentences
Short fragments stacked for drama, in the rhythm of a model imitating a copywriter.
Dramatic phrasing
Emotional or theatrical lines that sound out of place in a business conversation.
Personification
Abstract nouns given human actions, such as a story told by data or a market that wakes up.
Contrast headlines
Every title built as this versus that. Five in a row look like a template.
Generic claims
Benefits any competitor could print, with no number, name or date behind them.
What your team receives
The brand file
Your position, voice and visual rules written once for people and models.
Tool set-up
Claude, ChatGPT, Lovable and Figma configured and shared with the whole team.
The checker
The six signs and your banned phrases, run on drafts before editing.
Team session
Two hours with your marketers on prompts, review and the decisions that stay human.
Two experiments
We run your first two tests with the new set-up so your team sees results in market.
Before and after scores
We score your last 50 published pieces against the checker and then the next 50.
Three example test designs
How a brief for this service usually looks. We write yours with you.
The same 20 briefs drafted with and without the brand file
- Arms
- Two arms
- Channel
- Claude and ChatGPT
- Measure
- Checker flags per draft and editing time
A Claude Skill for proposals against the current template
- Arms
- Two arms
- Channel
- Claude
- Measure
- Checker score and review rounds
Lovable prototypes with the design tokens loaded
- Arms
- Two arms
- Channel
- Lovable
- Measure
- Review rounds to sign-off
These are example designs, and their numbers describe the setup. Real results publish on the experiment log.
Engagement terms
Questions about AI tools
Do you sell AI software?
No. We configure the tools your team already pays for. Your company keeps the brand file in plain text.
Will our copy stop sounding machine-written?
The checker removes the common signs. The brand file adds what is specific to your company, so the copy sounds like your team wrote it.
Can you set up agentic marketing workflows?
Yes, after the brand file and the checker are in place. Agents that draft and publish on their own need both, because nobody reads every output.
Request the AI set-up sprint
Send us the tools your team uses and the content it publishes each month. We reply with the scope and the fee.


