Watch the full guide on YouTube:
Want to turn your AI knowledge into paid work?
Join my free live AI workshop. I will show you what businesses pay for, how I structure a workshop and how to get started.
The internet is filling up with the same bloody post.
One grand statement. A neat little contrast. Slop comments underneath calling it "high-signal truth".
Yes I’m looking at you LinkedIn!
But it’s a wider problem. And getting worse. AI sloppifaction of writing.
After a while you start wondering whether anybody is actually talking to anybody. Bots talking to bots, speedrunning the dead internet theory.
In the midst of all of this, this week OpenAI has added a writing-style feature to ChatGPT Work. It can study emails, messages and files from Gmail, Google Drive, Slack and SharePoint, then use your phrasing, capitalisation and sign-offs in the next thing it writes.
Pretty cool! It helps our AI writing sound, well, less AI.
Turn on ChatGPT's writing style
Open ChatGPT and go to Settings → Personalisation → Reference my writing style → Set up writing style.
Choose the services you are comfortable connecting. OpenAI's own guidance focuses on recent, human-written sent email and excludes forwarded or generated messages which do not represent your voice.
That saves you pasting the same tone instructions into every chat. It also means the material you give it needs some care. An inbox or Drive full of old AI copy you never liked will teach it to produce more old AI copy you never liked.
And yes, giving OpenAI access to your email and Slack is a fairly big privacy decision. You can use the other five methods without connecting any of them. I covered them ALL on my Youtube channel:
Give it your actual voice
Briefly though:
Give AI copies of your actual writing. Ideally from pre 2022 (ie pre AI) to make sure uncontaminated!
Have AI interview you and capture your spoken tone of voice. Save the transcript of how you actually talk.
Feed your writing samples and spoken transcipt into AI and have it create a Tone of Voice document (ideally a Skill file)
Also use an AI Tells checklist or Skill to do a pass on any outputs to remove heinous AI language.
Everytime you do a significant piece of writing and and then human edit it GIVE the AI back the human edit so it can see the difference between its draft and your final version!
Save all of this work in your AI brain, where ChatGPT, Claude and Codex can all reach it.
Show it what you changed
The final step is so important but very few people do it.
If I was an AI I’d say “Here’s the part everyone misses”… unfortunately that’s true here! Ha!
Let’s say AI writes a draft. You fix the opening, cut two paragraphs, put your opinion back in and publish it. Normally the AI never sees those changes. Next week it makes the same mistakes and you get annoyed again.
But…how would it know how to do better? You have to show it!
Give it both versions and say:
Compare this draft with my published version. Separate factual corrections and one-off edits from writing preferences we should keep. Propose updates to my voice guide.
I use this loop on this newsletter. Draft, human edit, publish, compare, update the guide. The setup gets better because it learns from the version I was actually willing to send you.
Doesn’t mean I stop editing it though. Same with all writing. And I recommend the same for you.
Or at least read bloody thing before you post it! Ha!
Here’s the full guide:
To the Task,
Kyle