Most AI writing is prompting: write instructions, see what comes back, then fix it by hand. Vera treats your voice the way engineers treat software. It's specified, tested, measured and versioned. This page is the detail behind the homepage.
Your rules aren't suggestions in a prompt. Every draft is checked against them before anyone sees it, the way code is tested before it ships.
One goal you can measure: how much of each draft you have to change, and how close it sits to your measured voice. Every improvement is judged by whether that number goes down.
Before Vera changes how it writes for you, the new version is tested against posts you've already approved. If it doesn't beat the current version, it doesn't go live.
Every version of how Vera writes for you is saved and scored. If something drifts, it rolls back to the last version that sounded like you.
When the underlying AI models change, your voice can be re-tested against your own posts, not rewritten from scratch.
Measured on every post: see the numbers below.
The optimizer's job is to find the writing instructions and examples that reproduce your approved posts with the fewest edits.
Vera compares each draft with what you actually publish. All three numbers should fall as it learns you.
Changes you made to a draft before posting it.
Average share of each post you changed before publishing it.
AI cost to turn one idea into drafts for every channel.
Example monthly report · illustrative data, not client results.
Today Vera adapts by improving its instructions and examples from your decisions. Once there's enough of your own writing, the plan is to fine-tune a model on it.