Quick answer
To teach AI your style, collect 10 to 30 of your best-performing pieces of content, ask an AI tool to list the patterns they share (structure, length, tone, hooks, visuals and calls to action), and turn those patterns into a written style guide with real examples. Then test the AI’s new drafts side by side with your originals, and add every correction you make back into the guide.
Key takeaways
- Use content that actually performed, not what you like best
- Ask AI for specific, measurable patterns, not vague adjectives
- Write the patterns into one reusable style guide with examples
- Test drafts side by side with a proven original
- Every correction you make should go back into the guide
In this guide
- Why teach AI your style from your best-performing content?
- Step 1: How do you pick your best examples?
- Step 2: How do you gather them in one place?
- Step 3: How do you get AI to find the patterns?
- Step 4: How do you turn patterns into a style guide?
- Step 5: How do you test it against the originals?
- Step 6: How do you keep improving it?
- What can’t AI copy well yet?
- Frequently asked questions
- The bottom line
- Related posts
This guide supports our main article, 3 Claude Code AI Workflows Nate Herk Uses Every Day (and How to Copy Them). In it, Nate Herk explains how his AI video editor learned his house style by studying 27 of his best human-edited reels. Here we answer one practical question: how can you teach AI your own style from your best content, without writing code? The method works for emails, social posts, blog articles and video scripts alike.
📘 Main article: See how Nate applied this to video editing, plus two other workflows, in 3 Claude Code AI Workflows Nate Herk Uses Every Day (and How to Copy Them).
Why teach AI your style from your best-performing content?
Because AI on its own drifts towards the average. Nate puts it simply: a language model predicts the most likely next word, so its default ideas are what most people would write. Your best-performing content is the opposite of average. It is proof of what works for your audience.
AI teams call a set of known-good examples a golden dataset and test their systems against it. You can borrow the idea: your top pieces become the target every AI draft is measured against.
Step 1: How do you pick your best examples?
Choose by results, not by what you are proudest of. Pick one channel at a time and sort by the measure that matters to you.
| Content type | Measure to sort by | How many to pick |
|---|---|---|
| Marketing emails | Click rate or sales, not just opens | 10–20 |
| Social posts or reels | Saves, shares or comments per view | 15–30 |
| Blog articles | Search traffic and time on page | 5–10 |
| Video scripts | Average view duration | 5–10 |
💡 Tip: Add three of your weakest pieces too, clearly labelled. Showing the AI what did not work helps it understand what did.
Step 2: How do you gather them in one place?
Put everything in one folder or one document. For text, paste each piece in full with its result written above it, for example “Email: 9% click rate”. For video, add the transcript plus a few screenshots, because many AI tools cannot watch video directly. Nate’s editor gets round this by taking screenshots every few seconds.
- Name files clearly, such as best-01-email-launch.txt
- Remove customer names and private details
- Keep the original dates, in case your style has changed over time
Step 3: How do you get AI to find the patterns?
Ask for specific patterns you could count or check. Upload or paste your examples and use a prompt like this:
These are my best-performing [emails/posts/scripts], each with its result. Study them and list the patterns they share. Be specific and measurable: opening line style, length, sentence length, structure, tone, words I use often, how I use examples, call to action, formatting and visuals. For each pattern, quote one example. Then list what the weak examples do differently.
Nate’s editor went as far as noting how often scenes change and the relative volume of voice and music. You do not need that detail on day one, but the more concrete the patterns, the more useful the guide.
Step 4: How do you turn patterns into a style guide?
Ask the AI to rewrite its findings as a style guide, then edit it yourself. Keep it to one or two pages so any tool can use it. A good structure:
- Who it is for: your audience in two sentences
- Voice: three to five rules, each with a “do” and “don’t” example
- Structure: the shape your best pieces follow
- Must-haves: for example a clear next step in every piece
- Never: phrases, claims or habits to avoid
- Two full examples of your best work
Save it where your AI tool can reuse it: a custom GPT, a Claude project, or a skill file if you use Claude Code. If you build several, our guide How to Load AI Skills Only When You Need Them, Not Every Session keeps them tidy.
Step 5: How do you test it against the originals?
Give the AI the brief for one of your proven pieces and compare its draft with the original. Nate did exactly this, putting the AI edit next to the human edit of the same reel. Score the draft out of 10 on voice, structure and clarity, and note exactly where it falls short.
⚠️ Warning: Do not publish AI drafts unchecked, even when they look right. Check facts, links and any claims about results. Honest, accurate content keeps your audience’s trust.
Step 6: How do you keep improving it?
Every time you fix something, add the fix to the guide as a rule. Nate calls each piece of feedback an investment: the skill gets better every time. Once a month, add your newest top performers and remove rules that no longer hold.
If several people approve content, agree who signs off each version. Our guide How to Version and Approve AI Campaign Assets Before You Export Them covers a simple approval process.
What can’t AI copy well yet?
The finer, more human touches. Nate says AI still lags behind a human editor on music and sound design. In written content the equivalents are humour, timing and genuine personal stories. Keep those parts for yourself, and let the AI handle structure, first drafts and consistency.
Your next step: pull your ten best emails or posts into one document today, write each result above it and run the pattern prompt. For more on where this idea came from, read our main article on Nate Herk’s Claude Code workflows.
Frequently asked questions
How many examples does AI need to learn my style?
Ten to thirty strong examples is a good range. Nate Herk’s editing skill learned from 27 top reels. Quality and real results matter more than volume.
Can ChatGPT learn my writing style?
It can follow your style closely if you give it examples and a written style guide each time, for example in a custom GPT or project. It does not permanently learn from you unless you save those instructions.
What is a golden dataset?
A set of known-good examples used to guide or test AI output. Your best-performing content can play that role for your brand.
Should I include bad examples?
A few, clearly labelled. Contrast helps the AI see what makes your best work different.
The bottom line
The fastest way to teach AI your style is to show it what has already worked for you, write the patterns into a style guide and keep improving it with every correction. Your next step: gather your ten best pieces today.
Related posts
- 3 Claude Code AI Workflows Nate Herk Uses Every Day (and How to Copy Them)
- How to Load AI Skills Only When You Need Them, Not Every Session
- How to Version and Approve AI Campaign Assets Before You Export Them

[…] Main article: You can use this idea without writing any code. Our guide How to Teach AI Your Style Using Your Best-Performing Content shows how to turn your best emails, posts or videos into a style guide any AI tool can […]