Voice Brain: a memory of everything you have written

By Jack Stovell X · Instagram · LinkedIn · published 2026-09-22 · updated 2026-10-02

Voice Brain reads what you have written with ScriptGrain, and anything you add, and remembers it properly: the ideas, the claims, the stories, the frameworks, each tied to the exact words it came from. Every new brief gets checked against that memory first, so drafts build on what you have said instead of repeating it or quietly contradicting it.

Try it on your own posts

Here's the quickest way to see what Voice Brain does. Paste or link two to five of your own pieces into the free content ideas tool, right below this section. No account. Nothing stored.

It reads your writing with the same extractor Voice Brain uses. Then it hands back five things to write next that build on what you've already said, the threads you keep coming back to, and the claims that pull against each other. Each one comes with your own quoted words, so you can see exactly where it came from.

One run is a snapshot. Voice Brain keeps the memory going across everything you write.

Why we built it

I'm Jack Stovell, founder of ScriptGrain. We didn't design this feature from a whiteboard. We built it because we were already running a version of it ourselves.

We have run a version of this memory inside our own setup for more than three months. It is how every AI agent we run already knows the product, and it is how they write content for us through ScriptGrain. That's not a small detail. It made a clear difference to new content and to improving older pieces, so we built it into the product.

So the origin story is boring, in the best way. We needed it, we used it, it worked. That's why it exists.

What it remembers

Voice Brain reads your finished drafts, including your edits. It reads your writing samples. It reads pieces you add yourself, pasted or by URL, as long as they run to 60 words or more. And it reads the pages your voice monitor watches. Every half hour it checks for new and changed pieces and reads only those.

From each piece, it keeps the ideas, claims, stories and frameworks it contains, and it keeps them with the exact words they came from. No verbatim quote, no memory. Anything that can't be traced back to the text gets thrown away. That's a deliberate constraint, not a limitation someone forgot to fix.

Pieces then get grouped into topics, and memories get linked to each other: 'repeats' when you have said this before, 'builds on' when a new piece extends an old one, 'contradicts' when two pieces disagree, and 'refines' when a later piece sharpens an earlier claim. Contradictions surface as 'you changed your mind', which sounds blunt because it is. You can mark one as not a change of mind if it isn't, but the system doesn't pretend the tension away by default.

Better new pieces

Here's where it earns its keep. Before you write, the novelty check reads your brief against the memory and tells you plainly whether it repeats something, builds on something, or is genuinely new. It also suggests angles grounded in what you have already argued, not generic ones pulled from thin air.

While it writes, once a voice has 10 or more pieces in its memory, every outline and draft gets handed two things: what is established, to build on, and what is already argued, so it isn't repeated. There's a setting called 'Avoid repeating myself', on by default. Turn it off when similar content is genuinely fine, a landing page variant, say, where repetition is the point.

Say the brief is "why founders should write their own content." The novelty check might flag that you have made this argument twice already, once as a claim about authenticity and once as a story about a client who outsourced it and regretted it. It won't stop you writing a third version. It'll tell you what angle hasn't been used yet, so the third piece adds something instead of echoing the first two.

What next asks the memory for five angles that build on what you keep saying and go where you haven't been. Not five random topics. Five extensions of an argument you have already started, aimed at the gaps.

Better older pieces

New pieces aren't the only thing that benefits. Page rewrites use the same memory, so an old page rewritten in your voice builds on everything you have said since it first went live. So an old page comes back in your current voice, building on what you have argued since.

Older pages can also carry a register that was never yours. Everyone has heard the shared register that spreads when enough writing comes from the same few places: stock openers, stock closers, nothing you'd actually say. We wrote about the AI accent for exactly that reason. You'll learn to spot it, see why it arrived, and check your own pages for it.

And the memory tells you where to look. The 'changed your mind' list surfaces genuine contradictions worth resolving. Topics you haven't written about for a while surface too, which is a quieter but useful signal: if a subject has gone cold, that's often exactly where an old page needs a second pass.

One brain or several

Every voice starts with its own brain. That's the default, and for most people it's also the right answer.

Share a brain when the voices genuinely overlap: a founder and their company is the clear case, or a team writing on a single subject where consistency matters more than individual flair. Each voice still writes in its own style; sharing memory doesn't flatten anyone into sameness.

Keep separate brains when the voices shouldn't bleed into each other: different clients, different audiences, anywhere a contradiction in one context shouldn't show up as baggage in another. This matters more than it sounds, particularly for agencies and ghostwriters juggling several client voices under one account: mixing memories there isn't a convenience, it's a liability.

Joining a shared brain brings the voice's memory with it. Leaving takes a copy, so nothing is lost either way. New voices get their own brain unless you pick one when you create them, which means the safe default is always there and the shared option is always a deliberate choice, never an accident.

See it, and change it

Every voice's memory is drawn as a map on the memory page. Each topic owns an area, sized by how many memories it holds. The memories you repeat most sit at the centre of their topic. And the same memory always lands in the same place, so the map reads the same next week as it does today.

Looking for something specific? Search with ⌘K or the slash key.

Select a memory and you see its kind, its quote, its topic and what it links to. Select a topic and you see the memories inside it. Simple.

And you're not stuck with what the extractor found. To be fair, no extractor gets everything right, so you can fix it:

Add a memory by hand. Type it in, without writing a whole piece around it.
Edit its words, its kind or its topic.
Delete it.
Rename a topic, or merge two into one.

Your edits survive the piece being read again. They don't get overwritten the next time something runs. Adding and editing also work through the API and MCP, so the map isn't tied to one screen.

You stay in charge

None of this is a black box you have to trust blindly. You can reword any memory, delete it outright, or forget the piece it came from entirely. Edits and deletions survive the piece being read again, so correcting something once is enough; the system won't quietly resurrect an old version. And a forgotten piece is never read again, full stop.

It's available on every plan, including Free, in the app, through the API, and over MCP. For teams already building their own agents against the AI writing API and MCP server, the memory travels with the voice, not just the interface.

The brain behind this page, today

So what does a working brain look like? Here are the numbers for ScriptGrain's own brand voice, measured on 2 October 2026.

It holds 74 pieces and 796 memories across 109 topics. Those memories break down as 306 ideas, 294 claims, 71 stories and 125 frameworks. Between them sit 1,379 'builds on' links, 147 refinements and 5 contradictions. In September it held 17 pieces and 210 memories.

Then there's the number worth a closer look: 225 places where one piece repeats an earlier one. That isn't there to embarrass anyone. What is already argued goes into every outline and draft as points not to repeat. So the brain doesn't just store what you've said. It keeps your drafts from saying it twice.

Voice Brain is one of several things on what's new, alongside the new tools, and it works next to the loop that learns from your edits. That's the split: the brain remembers what you said, the profile learns how you say it.

That's the whole pitch, really. Not more content. Content that remembers what came before it.

Questions

How do I stop AI repeating what I have already written?

Give it a memory of what you have written and check every brief against it. ScriptGrain's Voice Brain keeps the ideas, claims, stories and frameworks from your pieces, each tied to the exact words it came from, and before you write it tells you whether a brief repeats something, builds on something or is new. Once a voice has 10 or more pieces in its memory, every outline and draft is told what you have already argued, so it is not repeated. The setting is called Avoid repeating myself and is on by default.

Is there an AI that remembers everything I have published?

ScriptGrain's Voice Brain remembers what you write in ScriptGrain, your writing samples, pieces you add by paste or URL, and the pages your voice monitor watches, and checks for new and changed pieces every half hour. It keeps only what it can tie to a verbatim quote from the piece, and you can edit, delete or forget anything it holds.

How do I make ChatGPT remember my writing style?

ChatGPT's memory stores facts about you, not a measured reading of how you write or a record of what you have already said. ScriptGrain splits the job in two: the voice profile measures how you write across 45 attributes, and Voice Brain remembers what you have said. Both work in the app, through the API and over MCP, so the same voice and memory can be used from ChatGPT, Claude and other assistants.

What should I write next, based on what I have already posted?

Paste two to five of your own posts into the free content ideas tool on this page for five angles that build on what you have said, with the quotes they came from. Inside ScriptGrain, What next does the same from the whole memory of a voice, aiming at gaps and topics that have gone quiet.

How do I avoid contradicting myself across posts?

Voice Brain links memories as repeats, builds on, refines or contradicts. When two of your pieces disagree it surfaces them as you changed your mind, so you can resolve it, or mark it as not a change of mind if the difference is deliberate.

Is Voice Brain on the free plan, and does it train on my writing?

It is on every plan, including Free, in the app, through the API and over MCP. ScriptGrain measures your writing and remembers what it says; it never trains a model on it.