How to Make ChatGPT Remember Your Brand Voice

To make ChatGPT remember your brand voice, write it down as a short file of rules and examples, then store it where every chat reads it: custom instructions, a Project, or a connector that loads your brand file. Don’t leave it to memory. Memory summarizes, and voice is the first thing it loses.

Why does ChatGPT forget your brand voice?

ChatGPT starts every chat from its own default style. It only writes like you when something in the conversation tells it how, and the usual way is pasting a paragraph at the start of a chat. That works for one chat. The next day, in a new chat, the paragraph is gone and the default comes back: the same openers, the same lists of three, the same upbeat tone.

Memory looks like the fix, but it isn’t built for this. ChatGPT writes its saved memories itself and prunes them when it runs out of room. A voice guide of fifty careful lines tends to become one line like “prefers a casual tone”, which is too vague to change anything.

Where can you store your brand voice in ChatGPT?

There are four places, and they do different jobs.

Where	What it holds	Best for	The catch
Custom instructions	A short text applied to every new chat	One brand, one voice	Short, and it applies everywhere, including chats that aren’t about your brand
A Project	Instructions and files that apply only inside that project	A client, a second brand, a campaign	Only works when you start the chat inside the project
Memory	Facts ChatGPT chooses to save	Small personal details	You don’t control what it keeps or how it words it
A connector (MCP)	Your brand files, read by a tool when needed	Several apps, several brands, files that change	Needs a connector that stores files, and a plan that allows connectors

Most creators end up using two of these: custom instructions for the basics, plus a Project or a connector for the full brand file.

What should your brand voice file say?

A voice file is useful when it gives the model decisions, not adjectives. “Friendly and professional” tells ChatGPT nothing it wasn’t going to do anyway. Write rules it can follow and examples it can match:

Who you write for, in one sentence, with the situation they’re in.
Words you use and words you never use. Ten of each beats a paragraph of tone.
Sentence habits. Sentence length, how you open a post, how you end one.
Two or three real samples of posts you wrote and liked, labeled as samples.
What to avoid: the phrases that make a draft sound generic to you.

No file yet? Start from our brand voice guidelines template. Pair it with your content pillars so the voice has topics to work on.

How do you set up custom instructions for your voice?

Open Settings → Personalization → Custom instructions.
In the field about how ChatGPT should respond, paste the short version of your file: reader, five rules, words to avoid.
Keep the samples out. They’re long, and custom instructions apply to every chat, including ones that have nothing to do with your brand.
Start a new chat and ask for one post. Compare it with a post you wrote. Fix the rule that caused the biggest gap, then test again.

Custom instructions are the right tool when you run one brand and want its basics everywhere.

When should you use a ChatGPT Project instead?

Use a Project when the voice belongs to one brand among several, or when it needs the full file with samples. A Project keeps its own instructions and files, and they apply only to chats inside it.

Create a Project named after the brand.
Put the rules in the Project’s instructions.
Upload the full voice file and your samples as Project files.
Start every chat for that brand from inside the Project.

The weak point is discipline: a chat started outside the Project gets none of it. Freelancers and agencies with several clients usually need one Project per client.

How does a connector keep your voice across chats and apps?

A connector (an MCP server) is a tool ChatGPT calls while it works. Instead of you pasting the file, the AI reads it from the connector when a task needs it. Because the file lives outside ChatGPT, the same voice works in Claude too, and when you update the file, every app reads the new version.

The Nexin connector works this way. Its brand tool saves the files the Core interview writes, starting with your brand and speaker files. They’re stored on your account, encrypted, and read back in any new chat. You can keep separate sets for up to 50 brands. Setup is one URL in Settings → Apps & Connectors; custom connectors depend on your ChatGPT plan and workspace settings.

How do you check that ChatGPT kept your voice?

Don’t judge by feel after one post. Ask for five short posts on different topics and look for the same three things each time:

Openers. Do they start the way you start, or with ChatGPT’s stock openers? Compare against these hooks for social media.
Banned words. Search the drafts for the words your file forbids.
Rhythm. Read two posts aloud. If every sentence has the same length, the voice slipped.

If the drafts drift, the rule that failed is usually too vague. Rewrite it as something checkable (“never open with a question”) rather than adding more adjectives.

Does this work the same in Claude?

The ideas carry over with different names. Claude has its own Projects, and skills that load a method when a task needs it. Read Claude Projects vs skills for the split and how to use Claude for social media for the setup. Claude vs ChatGPT for content creation covers which one to write in.

FAQ

Can ChatGPT learn my voice from my old posts?

It can imitate samples you give it in the chat or the Project. It won’t study your account on its own, and imitation without rules drifts within a few posts, so give it both samples and rules.

How long should my brand voice instructions be?

Short enough to fit in custom instructions for the basics, with the full file kept in a Project or a connector. Five sharp rules beat thirty vague ones.

Should I turn memory off?

Memory is fine for small facts, so you can leave it on. Just don’t rely on it to hold the voice itself, and check the saved memories now and then for wrong summaries.

Can I use the same voice in ChatGPT and Claude?

Yes. Keep the voice in one file you own. Paste it into each app’s instructions, or use a connector that both apps read from.

What if I run more than one brand?

Use one Project per brand in ChatGPT, or a connector that keeps a separate set of brand files for each brand, so the voices never mix.

Keep one brand voice in every chat. Nexin Core runs a brand interview and writes your brand and speaker files. The Nexin connector saves them, so ChatGPT and Claude read them in every new chat. Every other module, from Instagram to Copy, writes from those files. Nexin writes; it doesn’t post or schedule. See pricing.