AddThisFeature

AI Translation

Translate app content while preserving structure, placeholders, and product terminology.

involved AI Content

What it adds

A translation pipeline over the app's translatable content that keeps source and target linked, protects non-translatable tokens, and routes uncertain output to review.

What your agent is told to do

5
  1. 1

    Identify what is genuinely translatable and separate it from what is not. Extract placeholders, markup, URLs, identifiers, code, and proper names into protected tokens before generation and restore them afterwards.

  2. 2

    Store every translation with a reference to the exact source version it came from. When the source changes, mark the translation stale and queue it for retranslation rather than leaving a silently outdated string in place.

  3. 3

    Hold a per-language glossary of product terms and their approved renderings, and apply it to every request. A term that translates three ways across the app is worse than leaving it in the source language.

  4. 4

    Run translation through the app's existing background-job system in batches, with a token ceiling per batch and backoff on provider errors. Do not translate on page render or on a user's request path.

  5. 5

    Require review before a translation becomes visible to end users, and let a reviewer approve, edit, or reject per string. Machine output published unreviewed will be found by a customer before it is found by the team.

Edge cases it handles

8
  • Placeholders, markup, URLs, and code fragments come back reordered, translated, or dropped. Verify every protected token is present and unaltered after generation, and reject the string when one is missing rather than shipping a broken interpolation.
  • Glossary terms must be enforced after generation as well as requested before it, because a model will happily ignore a term list mid-sentence.
  • Word-for-word output loses register. Formality, address form, and politeness level must be specified per language, since the correct choice differs by locale and cannot be inferred from the English source.
  • Ambiguous source text — a bare noun that is also a verb, a string with no context, a fragment reused in several places — must be flagged for a human rather than resolved by guessing. Give translators the source context and where the string appears.
  • Source and target must stay linked in both directions, so an edit to the source can find every affected translation and a reviewer can always see what a translation was made from.
  • Text expansion breaks layouts. Translated strings will run longer than the source in several languages, and the review surface should show where a string is used so the reviewer can catch overflow.
  • Right-to-left languages need direction and alignment handled at the layout level; a correct translation rendered in the wrong direction is still broken.
  • A refusal, a timeout, or truncated output must leave the previous approved translation in place and mark the string as failed, never fall back to showing the untranslated source where a translation already existed.

Definition of done

9
  • Placeholders, markup, URLs, and code are protected before generation and verified intact after it.
  • Each translation records the source version it derives from and is marked stale when the source changes.
  • A per-language glossary is applied and enforced on the output.
  • Translation runs in background batches with token ceilings and backoff, never on a request path.
  • No translation reaches end users without a reviewer approving it.
  • Ambiguous source strings are flagged with their usage context for human resolution.
  • A failed translation leaves the last approved version visible.
  • The feature matches the existing design system.
  • No existing functionality is broken.

Related features

How it works

  1. 1

    Copy the link

    Grab the Markdown instruction URL for this feature.

  2. 2

    Give it to your AI

    Paste it into Claude Code, Cursor, v0, Lovable — whatever you build with.

  3. 3

    It inspects, then implements

    Your agent reads your existing app first, then adds the feature to fit it.

Works with your stack

These instructions are written to adapt. They tell the agent to detect your framework, match your existing design system, and reuse what you already have — rather than assuming a particular stack.

Need it tighter than that? Customize the feature and tell it exactly what you're running.