ChatGPT, Claude, DeepSeek, Grok — all of them can translate text from one language to another, and for a paragraph or an email that often works surprisingly well. For a whole 300-page book, the math looks different. Here's an honest assessment, without putting any of them down.
What speaks for it
Instantly available, often free or cheap. No specialized service, no waiting — paste a paragraph in and you have a translation in seconds. For testing, individual chapters, or getting a feel for quality, that's a low barrier to entry.
Flexible for follow-up questions. You can directly ask "why did you translate it that way?" or "is there another option?" — a dialogue instead of a one-way result.
What speaks against it
Context window limits on long texts. General chat tools aren't built for 80,000-word manuscripts. Chapter-by-chapter copy-paste becomes necessary, and in between chunks the model loses track — a character's name might be spelled differently in chapter 3 than in chapter 12, a recurring technical term might suddenly be translated differently. Without a manually maintained glossary, that's left to chance.
No built-in quality check. A second, independent pass that actively looks for omissions, tense errors, or shifts in meaning doesn't exist by default in any of the tools mentioned. What comes out is the result of a single pass — errors aren't automatically cross-checked.
Wordplay and cultural references often get smoothed over, with no indication that a compromise was made at that point. You only notice once you're fluent enough in both languages to catch it yourself — which is usually exactly not the case when translating your own work.
Check the terms of service. Depending on the provider and plan, there can be restrictions on commercial use of generated content, or unclear rules around training-data usage of your own text — worth reading before using it for a book you intend to publish, especially on a free tier.
API cost estimation is similarly tricky. Using the API instead of the chat interface means paying per token — total cost for a whole book is hard to estimate accurately in advance, and long documents cost noticeably more than the first test paragraph would suggest.
When does the DIY route still make sense?
For short texts, individual chapters as a test, or if you're fluent enough in both languages yourself to check every line — then a general AI chat tool is a legitimate, low-cost path. For a complete book intended for publication, where consistency across 300 pages and a quality check matter, the effort of orchestrating that yourself (chunking, glossary maintenance, manual consistency checking) often outweighs the price difference to a dedicated translation service.