Teaching Your AI Writer to Sound Like Your Last Fifty Posts
How Brand Voice sync actually works: a hash-diffed re-index of your own published articles, feeding a retrieval layer your AI drafts pull from.
"Make it sound like us" is the hardest instruction to give an AI writing tool, because "us" isn't a style guide most teams have actually written down — it's an accumulated pattern across everything you've already published. Brand Voice sync is built around reading that pattern directly from your own archive instead of asking you to describe it.
What actually happens when you sync
Triggering a sync reads every article you've published, compares each one against what's already indexed using a content hash, and only re-processes what's actually changed since the last sync. An article that hasn't been touched since it was last indexed gets skipped entirely — no wasted work, no redundant re-indexing of your whole archive every time you publish one new post.
Whatever is new or changed gets ingested into a workspace-scoped index, one article per ingest call rather than batched together. That's a specific, deliberate choice: keeping the ingest one-to-one with the article means the number of chunks that come back from indexing maps cleanly to exactly that article, which matters later if that article needs to be re-synced or removed — the system can find and clean up precisely the right pieces without guessing.
Where that indexed material actually gets used
This isn't indexing for its own sake. The Brand Voice index feeds retrieval at generation time — when you're drafting something new, or when an article is being repurposed into social posts and an email newsletter, relevant chunks from your own published voice get pulled in as context for the model. The output isn't styled generically; it's grounded in passages you've actually published before.
Why this beats a style-guide prompt
A written style guide is someone's summary of your voice, which is already one layer removed from the voice itself, and it goes stale the moment your actual writing drifts from it. Reading directly from your published archive, kept current every time you sync, means the model's reference material updates itself as your body of work grows — no one has to remember to update a document describing how you write.