Which translation service do you need? It depends on the artifact you need back, and the honest answer for four very common jobs is "not us": subtitle files, scanned-image PDFs, pixel-perfect layout reproduction, and certified translations each belong to a different specialist. This guide routes each job to the right kind of service — including the jobs where a document AI service like GoodTrans is exactly right.
The 30-second routing table
| Your job | Right kind of service |
|---|---|
| Translate subtitle files (SRT/VTT) with timing intact | Subtitle/localization tools built around video |
| Translate a scanned PDF (image-only, no text layer) | OCR tool first, then a text translation workflow |
| Reproduce an exact page layout (brochures, designed PDFs) | Translation + DTP (desktop publishing) specialists |
| Certified, sworn, or court-accepted translation | Certified/sworn translators, per jurisdiction |
| Long documents where you need clean, reviewable text | Document AI translation service (see the end) |
Subtitles and SRT files
Subtitle work is not just translating the words. An SRT file carries timecodes, line-length limits, reading-speed constraints, and frame-sync rules. A general document translator that outputs Markdown or TXT will flatten exactly the parts that make a subtitle file work.
What to look for instead: a tool that ingests SRT/VTT directly, translates the text while preserving timecodes, and re-exports in the same format. Professional subtitlers additionally handle line breaking and reading speed for broadcast standards. If your video is a one-off for personal use, free subtitle translators exist; if it is client work, budget for a human subtitler to QC the result.
Scanned PDFs (image-only files)
A scanned PDF is a picture of text. Translation tools that work on text need a text layer — so the honest workflow is two steps:
- Run the scan through OCR software (Adobe Acrobat, ABBYY FineReader, Google Drive's OCR, and others) to extract editable text.
- Check the extraction — OCR errors on tables, footnotes, and handwriting are common — and then translate the extracted text.
GoodTrans translates the text layer of PDFs; a pure image scan has nothing to extract, so run OCR first. Be skeptical of any service that promises perfect scan translation with no OCR step: the OCR quality decision happens before translation and it drives everything downstream.
When the layout must survive
Designed documents — brochures, magazines, certificates with seals, slide decks — are layout objects. Translating them "in place" means text expansion and contraction (Spanish runs ~20-25% longer than English), font substitution, and re-typesetting. That is DTP work done in the original authoring files (InDesign, Illustrator, PowerPoint), by specialists who handle right-to-left scripts and CJK line breaking.
GoodTrans deliberately delivers editable Markdown and TXT instead: the content layer, reviewable and reusable. That is the right artifact for manuals, reports, policies, and manuscripts — and the wrong one when a pixel-perfect page is the deliverable.
Certified, sworn, and court-accepted translations
For immigration filings, courts, and many government processes, the translation must come from a certified or sworn translator recognized in that jurisdiction, often with a signed certificate of accuracy attached. No AI service can produce that — the certification attaches to a person's credentials and liability, not to software.
If your job is certification, find a certified translator or agency in the target jurisdiction and expect to pay per-page or per-word professional rates. If the certified part is one page inside a larger stack (a certificate inside a hundred-page report), some people translate the stack with AI and send the single certifiable page to a certified translator — check whether your receiving institution allows that split before relying on it.
What a document AI service is for
The jobs above share a trait: the deliverable is not "the text, translated and reviewable." For everything where it is — manuals, reports, specifications, policies, academic papers, book manuscripts — the workflow looks like this:
- upload the whole document, not pasted fragments
- set terminology once, applied consistently across hundreds of pages
- receive editable files, a bilingual review file, and a quality report
- know the cost before the task starts
That is exactly what GoodTrans does, with the estimate shown upfront on the translate page. For the boundary cases — AI vs human cost, per-word pricing, rights questions — see how much translation costs and whether it is legal to translate a book.
FAQ
Can Google Translate handle a scanned PDF? Google Drive can OCR some scanned PDFs and translate the result, but quality on complex layouts, tables, and handwriting is unreliable. For anything important, run dedicated OCR software and proof the extraction before translating.
How do I translate an SRT file and keep the timing? Use a subtitle-specific tool that preserves timecodes on export. General document translators output prose formats, which destroy the timing structure subtitles depend on.
What is a certified translation? A translation accompanied by a signed statement from a credentialed translator or agency attesting to its accuracy, accepted by courts, immigration authorities, and universities. Requirements vary by country and institution — confirm what your receiving organization accepts before ordering.
How do I keep the formatting when translating a PDF? You generally do not, directly. Either re-typeset the translated text in the original design files (DTP work), or accept a clean editable format like Markdown and rebuild the layout yourself. Pixel-perfect automated PDF translation does not exist yet; be wary of claims that say otherwise.
