Who signs off on the copy

Nielsen Norman Group published its own editorial process in September 2026: AI assists with clarity, formatting and critique, and a named human keeps editorial judgment and responsibility.

The division is worth borrowing, because interface copy has the same structure as editorial copy and almost never has the same accountability.

Separating the drafting from the deciding

Microcopy is a natural fit for generation. A tool will produce ten button labels faster than a person and several will be usable.

What a tool cannot do is decide which of the ten matches the vocabulary the product has already taught. That decision requires knowing what the same action is called on the other four screens, what the support team hears people call it, and what the legal team has said about a particular verb.

So the workable split is that generation proposes and a named person decides, which is the same arrangement the newsroom describes.

Giving the name somewhere to live

The accountability half is the part that usually goes missing, because interface copy rarely has an owner. It is written by whoever is building the screen, reviewed by whoever notices, and changed by anyone.

The minimum fix is a single person who owns the product's vocabulary and a place where it is written down. Not a style guide, which nobody opens, but a term list: the canonical noun for each object, the canonical verb for each action, and the terms that are forbidden because they were tried and confused people.

A tool will produce ten button labels faster than a person. It cannot decide which one matches the vocabulary the product already taught.

BrilliantUX editorial principle

Reviewing the copy where it lives

The last piece is where review happens. Copy reviewed in a spreadsheet reads differently from copy reviewed in the interface, and the difference is not small. Length, truncation, and the relationship between a button label and the heading above it are all invisible in a cell.

Reviewing in the build, with the real component and the real constraints, is what catches the cases where a perfectly good sentence is the wrong sentence for that particular 180 pixels.