Between September 2025 and January 2026, translation’s share of Claude usage increased from 0.63% to 0.71%, pointing to a modest increase in translation-related activity.
How Translators Use AI in Practice
More telling are the task-level interaction patterns within translation workflows, which show how translators are using AI across different types of work.
Core translation execution tasks — such as translating written or spoken content from one language to another or adapting translations — remain mostly automated, meaning AI directly executes tasks with minimal human involvement. These tasks are dominated by directive, one-shot interactions (i.e., “do X,” with little back-and-forth), indicating that Claude is mainly used as an execution layer when the task is clearly defined and tightly scoped, rather than an autonomous decision-maker.
That said, even execution tasks show early signs of becoming more iterative. For example, reading written materials and rewriting them into specified languages remains mostly automated, but task iteration increases slightly from 26% to 28%, while validation rises from 1% to 2% between September and January.
A more pronounced shift appears in interpreting. Simultaneous or consecutive interpreting remains mostly automated, but moves from 100% directive interaction in September to a more mixed pattern in January, with 52% directive, 28% task iteration, and 20% learning. This suggests that even highly execution-heavy language tasks are becoming less purely one-shot and more conversational over time.
By contrast, review, validation, and meaning-related tasks show consistently stronger signals of augmentation, with higher levels of task iteration, learning, and human oversight.
Tasks such as proofreading, editing, revising translated materials remain mostly augmented in both periods, with validation increasing from 11% in September to 17% in January, alongside sustained task iteration.
Other meaning-level tasks, such as identifying and resolving conflicts related to meaning, continue to be classified as fully learning-driven, underscoring Claude’s role as a support tool for understanding and interpretation rather than decision-making.
January’s data also surfaces more workflow-oriented translator tasks that showed very limited observed usage in September. These include checking original text or conferring with authors — which shows 65% task iteration — and compiling terminology and reference information — which is mostly augmented with 35% learning. This points to AI being used more in the work around translation — clarifying meaning, supporting decisions, and handling terminology — not just translating text.
Low AI Autonomy, Strong Human Control
In the latest report Anthropic also distinguishes automation from AI autonomy — defining autonomy as “the degree to which users delegate decision-making to Claude.” Translation provides a clear illustration of this distinction, according to Anthropic.
A prompt such as “Translate this paragraph into French” is considered high automation but low AI autonomy, because the task is tightly specified and requires limited decision-making from the model.
The task-level data supports this framing: even where translation tasks are automated, autonomy remains low, with humans continuing to define goals, resolve ambiguity, and carry responsibility for meaning, tone, and risk.
Claude’s strongest foothold, therefore, is not in replacing translators, but in supporting verification, auditing, and iterative refinement — particularly in higher-responsibility language work.
Taken together, the September-to-January comparison suggests that AI’s role in translation is stabilizing around this pattern:
- Automation for clearly bounded execution tasks
- Augmentation for review, validation, and meaning-level work
- Human-in-the-loop control where context, judgment, and risk matter
Rather than moving toward full autonomy, Claude is becoming more embedded as a review and quality layer in professional translation workflows.
