Make AI Write Blog Posts That Sound Like Your Brand
Quick answer: You can't retrain ChatGPT, Claude or Gemini, but you can give them a one-page voice file: 5 to 8 style rules, a list of banned words and 3 real samples of your best writing. Save it in a Project or custom instructions, write in four steps (brief, outline, draft, voice check) and edit the result.↗ Share on X
You can't "train" ChatGPT, Claude or Gemini on your brand voice the way a developer trains a model. What you can do works almost as well: build a one-page voice file (5 to 8 rules plus 3 real samples of your best writing), paste it into the tool's custom instructions or a saved project, and make the AI rewrite its first draft against that file before you edit. Most small teams can set this up in an afternoon. The rest of this article shows exactly what goes in the voice file, which settings to use in each tool, and how to check that the output really sounds like you.
Why does AI writing sound generic in the first place?
Writing Blog Posts With AI: 11 Mistakes That Sink Rankings →
Free AI Writing Tools: 7 Mistakes That Make Your Posts Worse →
Are Free AI Writing Tools Good Enough For Blog Posts? →AI writing tools learn from a huge mix of public text. When you give them a short prompt like "write a blog post about email marketing," they fall back on the safest, most average version of that topic. That's why so many AI posts open with "In today's fast-paced digital world" and close with "In conclusion."
The tool isn't bad. It just has nothing specific to copy. Your job is to give it something specific: your words, your rhythm, your opinions, and the phrases you never use.
There are three common mistakes that keep the output generic:
1. Describing the voice with adjectives only. "Friendly, professional, witty" means something different to every writer, human or machine.
2. No real examples. Rules without samples are like a recipe without a picture of the dish.
3. Asking for the final post in one shot. One long prompt that asks for research, structure and voice at the same time gets you a little of each and a lot of nothing.
Smart software picks in your inbox
What goes into a brand voice file?
A voice file is a plain document, about one page long, that you reuse every time. Here's the structure that works best:
| Section | What to write | Example |
|---|---|---|
| Who we write for | One sentence about the reader | "Owners of small online shops who do their own marketing at night." |
| How we sound | 3 to 5 rules, each with a do and a don't | "Use short sentences. Don't stack three adjectives." |
| Words we use | 5 to 10 phrases that are typical for you | "Here's the thing," "in plain English," "real numbers" |
| Words we never use | 10 to 20 banned words | "leverage," "delve," "game-changer," "unlock," "seamless" |
| Point of view | First person or third? Opinions allowed? | "We say 'we.' We take sides and say why." |
| Format habits | Length of paragraphs, use of lists, headings | "Max 3 sentences per paragraph. One table per post." |
| Samples | 3 excerpts of 150–300 words from your best posts | Pasted as-is, not summarized |
The banned words list does more work than people expect. AI tools lean on a small set of filler words. Removing them alone makes the text sound noticeably more human.
How do I pick the right writing samples?
Pick the posts that got the best reaction from real readers: the most replies, shares, or time on page. Don't pick the post you think is "most professional." Pick the one that sounds most like you on a good day.
Choose samples that show different jobs:
- one intro that hooks the reader,
- one section that explains something technical in simple words,
- one ending with a clear next step.
If you're a new brand with no archive yet, write those three samples yourself, by hand, before you involve AI. It's the best hour you'll spend on this whole project.
Where do I put the voice file in each AI tool?
How to Write Blog Posts With AI Without Sounding Like AI →
AI Blog Posts That Rank: 13 Mistakes to Stop Making →
How to Use AI to Research Blog Topics That Rank on Google: A Practical Guide →Every major tool now has a place to store instructions so you don't paste them every time. Names change often, so check the current menu, but the logic is the same.
1. ChatGPT: create a Project (or a custom GPT) and paste the voice file into its instructions. Upload your samples as files inside the project.
2. Claude: create a Project, put the voice file in the project instructions, and add your best posts to the project knowledge.
3. Gemini: create a Gem with the voice file as its instructions.
4. Jasper, Copy.ai and similar marketing tools: look for "Brand Voice" or "Brand Kit." You usually paste text or a URL, and the tool builds a voice profile. Review what it builds and fix anything vague.
Tip: keep the master voice file in a shared doc, not only inside the AI tool. When you switch tools, or a teammate starts writing, you copy it over in one minute.
What's the step-by-step process to write a post in your voice?
Don't ask for the finished post at once. Break the job into four short steps. Each step gets better results than one giant prompt.
Step 1 — Brief. Give the topic, the reader, the one thing they should do after reading, and 3 to 5 points you want covered. Add your own opinion in one line, even if it's rough.
Step 2 — Outline. Ask for an outline with headings only. Fix it yourself. This is where you add the stories, numbers and examples only you know. The AI cannot invent your customer stories, and you don't want it to try.
Step 3 — Draft. Ask for the draft section by section, reminding it to follow the voice file. Shorter sections stay closer to your style than one 1,500-word request.
Step 4 — Voice check. Paste the draft back and ask: *"Compare this draft to the voice file. List every sentence that breaks a rule or uses a banned word, then rewrite only those sentences."* This self-review step catches most of the generic phrases before you even start editing.
Then you edit. Plan on spending real time here. A good rule of thumb is that the AI saves you the blank page, not the editing.
How do I make AI writing sound more human?
Even with a solid voice file, some AI habits slip through. Watch for these patterns while you edit:
| AI habit | What to do instead |
|---|---|
| Every paragraph the same length | Mix one-line paragraphs with longer ones |
| Lists of exactly three adjectives | Keep the one adjective that matters |
| "It's not just X, it's Y" | State Y directly |
| Vague claims like "many experts agree" | Name your source or cut the line |
| Generic examples ("a company might...") | Use a real case from your work |
| Summary at the end that repeats the post | End with one concrete action |
The biggest single fix is adding things only you can know: a mistake you made, a number from your own account, a customer question you got last week. That's what readers remember, and it's also what search engines are trying to reward.
Will Google penalize AI-written blog posts?
Google's public guidance says it rewards helpful content no matter how it was produced, and acts against content made mainly to manipulate rankings. In practice, the risk isn't "AI was used." The risk is publishing thin, generic posts that look like thousands of other pages.
So the voice file isn't only about style. A post that includes your real experience, clear answers and specific details is more useful, and that's what gets and keeps rankings. A post that could have appeared on any site usually doesn't.
As for AI detectors: they are unreliable and often disagree with each other. Don't edit for a detector score. Edit for the reader.
How do I test whether the voice is working?
Run a simple blind test once you have three or four AI-assisted posts:
1. Take one paragraph from an old post you wrote by hand.
2. Take one paragraph from a new AI-assisted post.
3. Show both, without labels, to two or three people who know your brand: a coworker, a loyal customer, a friend.
4. Ask which one sounds more like you and why.
If they can't tell, your voice file works. If they pick the old one every time, ask what felt off. Their answer usually points to one missing rule. Add it to the file.
Also track a few simple numbers on new posts compared with your older ones: average time on page, scroll depth if your analytics shows it, and replies or comments. If those drop, the voice is drifting.
How often should I update the voice file?
Treat it as a living document. Review it every month or two, or whenever you notice the same edit over and over. Each repeated edit is a missing rule. For example, if you keep deleting the word "robust," add it to the banned list. If you keep adding a personal story to the intro, add "open with a real situation from our work" as a rule.
After a few rounds, most teams find their editing time shrinks a lot, because the file now captures what they used to fix by hand.
Your next step
Today, open a blank doc and create your voice file with the seven sections from the table above. Start with the easy wins: write 15 banned words and paste three of your best past excerpts. Then save it into a Project or custom instructions in the AI tool you already use, and write your next post with the four-step process. Edit it, run the blind test, and add one new rule to the file based on what you changed.
FAQ
Can I actually train ChatGPT on my writing style?
Not in the technical sense. You can store your voice rules and writing samples in a Project, custom GPT or custom instructions, and the tool will use them every time you write.
Does Google penalize AI-written content?
Google's public guidance focuses on whether content is helpful, not on how it was made. Thin, generic posts are the real risk, so add your own experience and specific details.
How many writing samples should I give the AI?
Three excerpts of 150 to 300 words usually work well: one strong intro, one clear explanation and one ending with a next step.