How Long Does It Take to Translate a Document? Realistic Timelines by Page Count
How long does it take to translate a document? Realistic turnaround times by page count for AI, freelancers, and agencies, plus what quietly adds days.
Someone in procurement forwards a supplier manual on Thursday afternoon and asks whether the Spanish version can be ready by Monday. Nobody in the building reads Spanish. There is no translation department. The only reference point anyone has is a vague sense that translation takes "a few days." How long does it take to translate a document is the question we get most often from companies that own documents but have never bought translation before, and the useful answer starts with a correction: page count is a bad unit of measurement, and most of the calendar time is not translation at all.
We've watched a lot of business documents move through a translation workflow, and the same pattern keeps showing up. The language work is rarely the bottleneck. Waiting is. What follows are the numbers we'd plan with, where they come from, and the specific things that quietly turn a two-day job into a two-week one.
A page is not a unit of work
Ask three suppliers how long a 40-page document takes and you'll get three answers, because none of them can tell what a page means until they open it. A page of a legal contract might hold 600 words of dense text. A page of a product brochure might hold 60 words and four images. A page of an assembly manual might hold 25 words of labels wrapped around a diagram.
Everyone in the industry quotes and schedules by word count, not page count. If you need a rough conversion for a first estimate, a typical DOCX page of running text lands somewhere between 250 and 400 words. Legal and financial documents run higher. Slide decks and marketing material run much lower per page but take longer per word, because short text carries more decisions.
The fastest thing you can do before asking anyone for a deadline is get the actual number. In Word, it's under Review, then Word Count. In Excel the count is harder, because text hides in cell comments, hidden sheets, and dropdown lists, so open every tab before you trust it. In PowerPoint the built-in count usually misses speaker notes and text inside grouped shapes, which is exactly the content that surprises people later.
That number is what a translator, an agency, or a tool will actually schedule against. A 40-page contract at 550 words per page is 22,000 words. A 40-page illustrated manual at 90 words per page is 3,600. Those are not the same job, and no honest supplier will give you the same date for both.
How long does it take to translate a document: the working numbers
The baseline the industry plans with is roughly 2,000 to 3,000 words of finished translation per translator per day. The American Translators Association uses about 2,000 words a day in its client-facing guidance, and that figure has held up as a planning number for years. It assumes a specialist working in a familiar subject with a clean, editable source file.
Three things move it.
Subject matter moves it down. A translator who works in pharmaceutical regulatory text every week might do 2,500 words a day there and 1,200 in a mechanical engineering manual, because research time swamps typing time.
Review moves it down again. Professional delivery normally includes a second linguist checking the work, and that reviewer handles roughly 5,000 to 8,000 words a day. On a mid-sized document, review adds most of another day.
AI moves it up, though less than the marketing suggests. The machine produces a full draft of a 20,000-word document in minutes. A human then post-edits it, and post-editing runs somewhere around 4,000 to 7,000 words a day depending on how good the raw output is and how much the editor trusts it. Faster than translating from scratch, not instant.
So a 10,000-word document takes a single human translator about four working days plus a day of review. The same document under a post-editing workflow takes about two days total. The same document with no human review at all takes under an hour of processing, and whether that's acceptable depends entirely on what the document does.
Realistic timelines by page count
Here are the ranges we'd actually give someone, assuming standard business documents of running text, one language pair, and an editable source file. Working days, not calendar days.
| Length | AI only, no human review | AI plus human post-editing | Full human translation and review |
|---|---|---|---|
| 1–5 pages (~1,500 words) | Same day | 1 day | 2 days |
| 10 pages (~3,000 words) | Same day | 1 day | 2–3 days |
| 25 pages (~8,000 words) | Same day | 2 days | 4–5 days |
| 50 pages (~16,000 words) | Same day | 3–4 days | 8–10 days |
| 100 pages (~32,000 words) | 1 day | 6–8 days | 15–20 days, or 8–10 split across translators |
Two things about that table are worth saying out loud. Splitting a long document across several translators buys you speed and costs you consistency, because two people will render the same term two ways unless someone enforces a shared glossary. And the AI-only column is almost never the real timeline for a document that leaves your company, because somebody still has to look at it, and that somebody has a calendar.
This breaks down in a few situations. Rare language pairs have thin translator pools, so a Norwegian to Vietnamese job may sit in a queue for days before anyone starts. Certified and sworn translation carries a formal signing and stamping step that adds turnaround regardless of length. Scanned PDFs are a different job entirely, since the text has to be recovered before anything can be translated, and we'd budget half a day to a full day for that on top of everything else.
The parts of the schedule nobody counts
A client sent us a 42-page maintenance manual for a packaging line, English to Spanish, and wanted it in three days. The word count was 11,400. On the numbers above that's a two-day post-editing job. It took nine days. Here's where the seven extra days went.
Two days went to getting a usable source file. What arrived was a PDF exported from InDesign by an agency that no longer worked with the client. The original layout file had to be tracked down, and when it couldn't be, someone rebuilt the tables by hand.
One day went to a question nobody could answer. The manual gave a torque specification in two conflicting units. The translator flagged it rather than guessing, which was correct, and the engineer who could resolve it was travelling.
Three days went to internal review. The client's Mexico City distributor wanted to check terminology before print. Nobody had told the distributor this was coming, so the file sat in an inbox over a weekend.
One day went to layout repair. Spanish ran about 20% longer than English, table cells overflowed, and page breaks moved.
None of that is translation. All of it is turnaround. When you estimate how long a document will take, the language work is the part you can predict most accurately, and it's usually the smallest slice.
Where AI actually saves time and where it doesn't
The saving from AI translation is concentrated in one place: producing the first complete draft. That step goes from days to minutes. Everything downstream stays roughly where it was.
It helps most with high-volume, low-stakes, repetitive content. Internal policies, supplier correspondence, product registers, training material refreshed every quarter. A six-tab XLSX product register with 4,300 words spread across cells is the clearest case we see. Translating that by hand means a person moving cell by cell for two days, and the risk of a misplaced row is real. An automated pass returns the whole workbook with its structure intact in minutes, and someone who knows the product spot-checks terminology in an hour.
It helps least with short, high-consequence text. A one-page press release, a set of warranty terms, a tagline. The draft arrives instantly and then three people argue about it for a week. Automation didn't compress that schedule at all, because the schedule was never about production speed.
There's also a category where speed is the wrong goal. If the document creates legal exposure, goes to a regulator, or could hurt someone when wrong, the review step is the product and the draft is scaffolding. We've seen a company compress a safety document from ten days to two, publish it, and then spend a month on a correction cycle. If you can't judge the output yourself, a QA report or a quality rating tells you where to look, but it doesn't replace a person who reads the target language reading the finished file. Our walkthrough on how to check translation quality when nobody on your team speaks the language covers what to ask for.
Document type changes the estimate more than page count does
A 30-page Word document, a 30-slide PowerPoint, and a 30-tab Excel workbook are three different schedules even at identical word counts.
DOCX is the friendliest. Text flows in one direction, formatting is recoverable, and length changes between languages mostly resolve themselves. A DOCX estimate usually lands close to whatever the word count implies.
PPTX is slower per word than anything else. Slide text is short, which means every line is a decision about tone and fit. It also expands badly: German or Russian running 20% to 30% longer than English will push text out of a box that was already tight. Budget extra time for slide cleanup, and budget it separately from translation, because it's a different person's job.
XLSX is unpredictable in a specific way. The translation itself is fast, since cells are short and repetitive. The risk sits in structure. Formulas referencing text values, dropdown lists tied to named ranges, conditional formatting keyed to specific strings. Any of it can break silently, and nobody notices until a report returns the wrong number three weeks later. We budget review time on spreadsheets by tab count rather than word count.
Scanned documents sit in their own category. Text recognition on a clean scan is quick. On a fax of a fax with a stamp across the middle it isn't, and someone will retype sections. If your source is a scan, add a day, and treat any estimate given before someone has opened the file as a guess.
How to set a date you can defend
Start with the word count, not the page count. Open the file, get the real number, and check the places where text hides: speaker notes, hidden sheets, headers, image captions, text inside grouped shapes.
Then work backwards through the whole chain instead of just the translation step. Sketch it as four blocks: source file preparation, translation, internal or in-market review, and final formatting. Put a named person and a date on each one. In the packaging-manual case above, three of those four blocks were unowned at the start, which is exactly why the estimate was wrong.
Ask your reviewer for their availability before you promise anything. The most common cause of a missed translation deadline we see is a reviewer who was never told the file was coming. If the person checking the Spanish is a distributor in another time zone with their own workload, that isn't a two-hour task, it's a scheduling negotiation.
Then apply a simple rule for the date: take the translation estimate, add the review window you just confirmed, and add one working day of slack for every 25 pages. That slack absorbs the source-file surprise, the terminology question, or the table that breaks. It isn't padding. It's the part of the job that always happens and never gets scheduled.
If you're weighing cost alongside time, our breakdown of what document translation costs in 2026 sits next to these timelines.
For the routine documents in that middle band, where you need a fast reviewable draft rather than a full agency engagement, this is roughly what SnapIntel is built for. You upload a DOCX, XLSX, or PPTX, pick the language pair, and get back a translated file in the same format along with a QA report and a quality rating you can use to decide where a human should look. There's a one-time trial of 5,000 words over 7 days with no card required, and a Pro plan at $20 a month covering 60,000 words. It compresses the draft-production block on the timeline above. It does nothing about your reviewer's calendar, and neither does anything else.