Why Generic AI Translation Ignores Your Translation Brief
Every professional translation begins with a brief. Who is this for, what is it for, what register, what tone, what to do with culture-specific items. Chat-based AI translation has nowhere to put a brief. You can paste one into the first message and the model nods, then forgets by the third chunk.
What goes wrong
You ask for a formal academic register and get conversational prose. You ask for British spelling and get a mix. You explain the work is a fantasy novel where invented words must be preserved and the model translates half of them. The output is fluent. It is also not the translation you commissioned.
Why generic AI translation fails here
A brief pasted into a chat is treated as a single user turn. Later turns drown it out. There is no project-level memory and no way to bind the brief to every chunk that follows. Worse, the brief is never recorded in the system, so on retry or resume it is gone.
How TranslationAI solves it
The purpose, mode, and editorial choices you make in the wizard are written to the project row before the first chunk is sent. From that point forward, those values are part of the work order, not a hint. Optional editorial guidelines you supply are injected as binding data into every chunk, every time. A retry uses the same brief. A resume uses the same brief. There is one source of truth for what you asked for, and the translator reads from it on every request.
Further reading: why AI loses narrative continuity, register and why it must be preserved, skopos theory and the role of the brief.