How to Translate a Scanned PDF in Smartcat: 3 Routes Compared

A client sends a scanned birth certificate and a 14-page contract that went through an office copier, and both have to be in Smartcat by tomorrow. This guide shows how to translate a scanned PDF in Smartcat without spending OCR budget on the wrong workflow, and what you get back at the end: a PDF or a Word file.

Smartcat reads scans itself. In Trados Studio and memoQ an image-only PDF gives the CAT tool nothing to extract (here’s why), so you convert it elsewhere first, as the Trados Studio and memoQ guides show.

The catch with Smartcat is threefold: the default PDF workflow doesn’t take scans, the scan workflow has to be picked by hand, and recognition is billed per page. Below are three ways through - Smartcat’s own OCR, OCR to Word first, and a scan-to-Word service - with what each one returns and where each one breaks.

What Smartcat actually does with a scanned PDF or image¶

Smartcat runs OCR (optical character recognition, the step that turns a picture of letters into text a program can split into segments) on scanned PDFs and image files. Its help article on OCR pages for freelancers, updated 24 July 2026, says recognition is required for scanned PDFs and image-based files and that each page needing it uses one OCR page.

The OCR has a history. A MultiLingual article from 3 April 2015 reported that ABBYY built its OCR into what was then called SmartCAT, so PDFs could be read “no matter whether they are text- or image-based”. Smartcat’s current help pages don’t name the OCR engine it uses today, so don’t assume it’s still ABBYY.

Three PDF workflows, and only one is built for scans¶

Smartcat’s guide to choosing a PDF translation workflow, updated 17 September 2026, lists three workflows. The table copies the guide’s own ratings.

The default workflow is the one that can’t handle a scan. In the same guide, Smartcat describes AI PDF Translation like this:

Does not support scanned PDFs (documents that are images of text)

So if you upload a scan the way you upload any other PDF, you’re on the wrong workflow before a single segment exists. The AI PDF Translation article, updated 31 August 2026, asks for “Documents with selectable text (not scanned images)”, accepts PDF files up to 100 MB, warns that documents over 60 pages take much longer, and states: “Handwritten text within PDFs is not supported.”

PDF via Image Pipeline is the scan workflow, and its parsing method is called pdfimageonly (the name says what it expects: a PDF made only of page images). The guide rates its layout “Very good” and its output is a translated PDF.

Standard (OCR to Word) is the workflow the guide points to when you need an editable Word file. Its layout is rated “Approximate”, its scanned-PDF support “Limited”, and the guide warns that tables and multi-column pages frequently need rework. All three workflows support your glossary and translation memory (TM), so the choice comes down to the input you have and the output the client wants.

Auto-detection: Smartcat’s own pages disagree¶

Smartcat’s PDF translator page says “Scanned files are detected and OCR’d automatically.” The same page then gives the most useful warning in this whole topic:

a misread source produces a confidently wrong translation

The warning matters more than the promise: a misread date or ID number doesn’t look broken after translation, it looks like a fluent sentence with a wrong number in it.

Smartcat’s page on translating large PDFs contradicts the auto-detection claim: it says pdfimageonly is “not detected for you”, and adds that “300 pages of OCR means 300 chances to misread”. The help center sides with the second page, since the default workflow simply doesn’t support scans. Pick pdfimageonly by hand every time you upload a scanned PDF.

Paying for OCR: OCR pages or Smartwords¶

Freelancers pay for recognition in OCR pages. As of July 2026, the OCR pages article says you buy them under “Productivity Services”, on a balance separate from MT pages, and a 20-page scan uses 20 OCR pages. The article gives no price per OCR page, so check the current price in your account before you quote the client.

Business workspaces pay in Smartwords, Smartcat’s usage unit for AI work. The workflow guide (September 2026) prices the Image Pipeline in two places, and the two per-page figures don’t match: the comparison table says one thing, the FAQ another. Don’t build a quote on either number; run one representative file and check the cost in your workspace before you send a large scan through.

Quoting the client is a separate problem, because a word count finds nothing in a scan. The guide to quoting a scanned document covers that, and the Smartcat, memoQ and Crowdin pricing breakdown and the free cloud CAT tools comparison cover the plans themselves.

Images and phone photos¶

Smartcat’s supported file formats page, updated 12 October 2026, lists the image formats it accepts through OCR: JPG/JPEG/JFIF, TIF/TIFF, PNG, BMP, GIF, DCX/PCX, JP2/JPC, DJVU/DJV, JB2 and image-only PDF. The same page says “Layout preservation depends on document complexity.” It gives no OCR language list and no image size limit.

The AI Image Translator is a separate tool with a different result. Per its help article, updated 24 July 2026, it takes PNG, JPEG, JPG, BMP, PCX, JP, JPC, JFIF and GIF (first frame only), removes the text from the image, translates it and puts it back. You download the result “in the same format as it was uploaded”, so a JPG of a diploma comes back as a JPG, not as a DOCX you can edit.

Check for a text layer first¶

Select a sentence in the PDF and paste it into a plain-text editor. Real words mean a text layer exists and the default workflow may fit; nothing means it’s a scan. The scanned PDF to Word comparison covers the test and mixed PDFs in detail.

Why scans break in a CAT tool: the mechanism¶

A scan is a photograph of a page. A CAT tool works on text: it splits sentences into segments, matches them against the TM and writes the translation back into the same structure. A scan has neither text nor structure, so two separate jobs have to happen before the first segment appears.

The first job is recognition: OCR reads letters from pixels. Every misread digit, every Latin “o” swapped for a Cyrillic “о”, every word broken across a fold becomes the source text your TM and machine translation work from. Smartcat’s segment editor can’t tell a misread number from a real one.

The second job is layout reconstruction: deciding which lines form a paragraph, which boxes form a table and in what order to read two columns. OCR can get every letter right and still return a table as loose lines or stitch two columns into one sentence. The article on why CAT tools break PDF tables and multi-column layouts shows how that turns into broken segments.

Official documents add parts that aren’t running text at all: stamps, seals, signatures, handwritten dates, photos and QR codes. OCR skips them, keeps them as pictures, or reads the ring of a round seal as a garbled line in the middle of a paragraph. Smartcat’s help pages on its three PDF workflows don’t mention stamps or seals anywhere, so there’s no documented behaviour to rely on.

Rebuilding a scan as an editable Word file is its own job. In the ChatsControl pipeline (more in Route 3), a vision model looks at each page and writes its Word copy. Text is typed as printed, tables become real tables, fields sit next to their labels, and stamps, seals, signatures and photos are cut from the page and set where they stand. QR codes and barcodes are read and drawn again clean. A second check compares the rendered copy with the page line by line and sends what it finds back to be fixed. Each scan gets a quality grade (good, fair or poor), and handwriting and very degraded pages are read unreliably, which the grade reflects.

Route 1: Smartcat’s own OCR, step by step¶

Use this route when the client wants a translated PDF back and the scan is readable printed text. The steps follow Smartcat’s workflow guide; the guide names no menus beyond the upload dialog, so neither does this list.

1. Sort the files. Test each PDF for a text layer and split the batch. The guide advises keeping the “digital native and scanned ‘piles’” apart, because they go through different workflows.
2. Check your recognition budget. Count the pages that need OCR. On a freelancer account, compare the count with your OCR page balance; in a business workspace, check what the job will draw in Smartwords.
3. Create the project without a file. The guide’s scan route starts with an empty project.
4. Upload the PDF and pick the parsing method. In the upload dialog, choose pdfimageonly.
5. Attach your TM and glossary. All three PDF workflows support both.
6. Check the recognised source before translating. Put the segments next to the scan and note every misread name, date, number and ID. A wrong digit in the source becomes a wrong digit in every target language.
7. Translate and review as in any other Smartcat project.
8. Download the translated PDF. The Image Pipeline’s output is a PDF, per the guide.

Step 4 is the one that decides everything. The guide puts it in one line:

choose pdfimageonly as the parsing method

Miss that line and the scan lands in a workflow that doesn’t support scans. Write the choice into your intake checklist so a project manager or a new freelancer can’t skip it.

When to pick Standard (OCR to Word) instead¶

Pick Standard when the deliverable must be an editable Word file, for example when the client’s notary or a sworn translator will add a certification page. Smartcat’s guide is honest about the cost of that choice: layout “Approximate”, scanned-PDF support “Limited”, and tables and multi-column pages frequently need rework.

For a one-page letter that’s fine. For a birth certificate built as a table of fields, you’ll rebuild the table in Word after export, and that’s layout work you either bill or absorb (see DTP charges for scanned documents).

Limits of route 1¶

• The output is a PDF. Any later correction means another round in Smartcat rather than a quick edit in Word.
• Every page costs. Each page that needs recognition draws an OCR page or Smartwords before translation starts.
• Layout varies. The formats page says layout preservation “depends on document complexity”, so a form with nested boxes needs a closer check than a letter.

Route 2: OCR to DOCX first, then import into Smartcat¶

Route 2 moves recognition out of Smartcat: you turn the scan into a Word file with an OCR tool, clean it, and import it as an ordinary DOCX, which gives you a checkable source and an editable Word target.

Microsoft Word, Adobe Acrobat, ABBYY FineReader PDF and Google Docs (through Google Drive) can all produce a DOCX from a scan. Microsoft says Word’s PDF conversion works best with files that are mostly text, and Google warns that Drive OCR is unlikely to detect tables, columns, lists and footnotes. Acrobat lets you set the recognition language, and FineReader is a dedicated desktop OCR application. The comparison of scan-to-Word options has the limits and sources for each.

Per Smartcat’s OCR pages article, OCR pages go on pages that need recognition, and a Word file with real text has none.

Step by step¶

1. Run OCR with the document’s language set, where the tool lets you set it. A Ukrainian certificate read with an English-only setting can fill the text with lookalike Latin letters.
2. Export to DOCX and keep the scan open next to it.
3. Check the text against the scan: names, dates, numbers, IDs, addresses. Mark anything you can’t read as unreadable instead of guessing.
4. Fix the structure before import: - Join lines broken inside sentences. OCR puts a hard return after every printed line, and a CAT tool ends a segment at every paragraph mark, so one sentence turns into three segments. - Pull text out of floating text boxes into the body or into table cells, or it may come into Smartcat out of order. - Fix the reading order of columns so the left column ends before the right one starts. - Rebuild tables as real Word tables, not tab-separated lines. - Clear the formatting noise. OCR output that changes font, size or character spacing in the middle of a word produces extra tags in a CAT tool, so select the body text and clear direct formatting where the layout allows.
5. Decide on stamps, seals and signatures. Leave them as images where they stand, and plan a bracketed note such as [Seal: …] in the target if the receiving authority wants the wording rendered. The stamps and seals guide covers the conventions.
6. Import the DOCX into Smartcat like any client Word file and look at the segments and the tag view before you assign the job.
7. Translate and download the Word target. If the client wants a PDF too, save it as PDF from Word yourself.

Route 2 costs you time rather than per-page fees: an OCR licence or a free tool, plus your cleanup. On a clean single-column letter the cleanup is short; on a form with tables it can approach the cost of retyping, which the real cost of retyping scans puts in numbers.

Route 3: a scan-to-Word service, then Smartcat¶

Route 3 replaces the manual OCR and cleanup of Route 2 with a service that rebuilds the page as Word. One option is ChatsControl: you upload the scan or photo (PDF, PNG, JPG, TIFF, WebP), it reads the page and returns an editable Word copy that follows the page layout. In that copy the text is typed as printed, tables are real tables, stamps, seals, signatures and photos are cut from the page and placed where they stand, and QR codes are redrawn clean. You then import that .docx into Smartcat like any client Word file (or have it translated directly in the same service).

The CAT route looks like this:

1. Upload the scan or photo to ChatsControl.
2. Wait for the job to finish.
3. On the job page, download the Word copy of the original with the “Original” button. It sits next to the translated Word file and the raw scan.
4. Check the copy against the scan: look at the quality grade, then names, numbers and dates.
5. Import the .docx into Smartcat. The result is an ordinary Word document, and Smartcat imports it the way it imports any client .docx. Check the tag view after import, as with any converted file.

The job also produces a translation of the document, so you can compare it with your own work or ignore it. Prices as of October 2026, from ChatsControl: a free tier of 6,000 characters per month, then pay as you go at $1.50 per 1,000 characters ($2.70 per 1,800-character page). Scans are billed by recognised characters, at least 1,000 characters per page, and uploads are deleted automatically after 30 days by default.

The limits matter just as much. ChatsControl isn’t a CAT tool or a TMS: no translation memory, no segment-level project management, so the TM work stays in Smartcat. It runs in the cloud only, with no offline or on-premise mode. Handwriting and very degraded scans are read unreliably, and the quality grade says so. Stamps, seals and signatures stay as pictures where they stand, so if the authority wants the seal’s wording rendered, you add a bracketed note such as [Seal: …] in the translation; the text inside the stamp image isn’t translated.

To judge the Word copies before uploading anything, the scan examples page shows real scans next to the Word files the product returned.

The three routes side by side¶

The table compares what each route returns and what survives from the page. Smartcat’s ratings come from its workflow guide (September 2026); ChatsControl prices are as of October 2026.

Pre-import checklist¶

Run these checks before the file goes anywhere near a Smartcat project. Each line points to a fact in Smartcat’s own docs or to a failure described above.

What goes wrong with scans in Smartcat¶

The scan goes into the default workflow. AI PDF Translation is the default and doesn’t support scans, whatever the marketing says about automatic detection. Choose pdfimageonly by hand for every scan.

The OCR balance runs out mid-batch. Freelancer OCR pages sit on their own balance, apart from MT pages. When they’re gone, the OCR pages article quotes the message you’ll see:

You’ve used all your OCR pages.

A 40-page batch of certificates needs 40 OCR pages before any translation starts, so check the balance before you accept a deadline, not at 11 pm the night before.

The cost estimate doesn’t match the guide. The workflow guide gives two different per-page Smartword figures for the Image Pipeline, so run a sample and read the actual charge before you price a large scanned archive.

A large batch has no bulk shortcut. The workflow guide notes there’s no bulk approve or bulk complete, so 30 separate scans mean 30 files to finish one by one.

A photo goes through the Image Translator. The AI Image Translator returns the file “in the same format as it was uploaded”. A translated JPG of a diploma can’t be edited, certified with a translator’s statement in Word, or checked segment by segment later. Phone photos are better handled as described in the WhatsApp document photos workflow.

Handwriting slips through. The Image Pipeline accepts handwriting via OCR, but a misread handwritten date or surname becomes the confidently wrong translation Smartcat’s own page warns about. For mostly handwritten scans, see digitizing handwritten documents.

Stamps and seals vanish into the layout. None of Smartcat’s PDF workflow docs mention stamps or seals. A certified translation needs the stamp accounted for, so plan a bracketed description ([Seal: …]) in the target and check the output page by page.

The scan is too poor for any OCR. Faded copies, skewed phone shots and fax-quality pages raise the error rate in every route. The poor-quality scan guide lists what to fix before recognition.

The client wanted Word, you delivered PDF. AI PDF Translation and the Image Pipeline return a PDF, Standard returns Word with approximate layout. Ask what the client needs back before you choose the workflow, because switching after translation means re-running recognition. After delivery, a free meaning and numbers check of the translated file against the source catches the misread digits that slip past a read-through.

FAQ¶

How do I translate a scanned PDF in Smartcat?¶

Create the project without a file, upload the scanned PDF and choose pdfimageonly as the parsing method in the upload dialog, as Smartcat’s PDF workflow guide describes. The default AI PDF Translation workflow does not support scanned PDFs, and the scan workflow returns a translated PDF.

Does Smartcat have OCR, and is it free?¶

Yes, Smartcat runs OCR on scanned PDFs and image files, but recognition is paid. Freelancers spend one OCR page per recognised page from a balance bought under Productivity Services, and business workspaces pay in Smartwords.

What is pdfimageonly in Smartcat?¶

pdfimageonly is the parsing method for Smartcat’s PDF via Image Pipeline workflow, the one its guide rates as supporting scanned PDFs and handwriting via OCR. You choose it by hand in the upload dialog, because Smartcat’s large-PDF page says it is not detected for you.

Can Smartcat translate a photo or an image of a document?¶

Yes. Smartcat accepts JPG, PNG, TIFF, BMP, GIF, DjVu and other image formats through OCR, and its AI Image Translator replaces the text inside an image. The Image Translator returns the image in the format you uploaded, so a photo of a certificate comes back as a picture, not an editable DOCX.

Can I get an editable Word file back from Smartcat instead of a PDF?¶

Smartcat’s Standard (OCR to Word) workflow produces Word, but its guide rates the layout as approximate and scanned-PDF support as limited. For a cleaner editable result, convert the scan to DOCX first, check it against the page, and import that Word file into Smartcat as an ordinary document.

Will the segments and tags be messy after I import an OCR’d Word file?¶

Segments and tags can get messy. OCR output with line breaks inside sentences, floating text boxes and frequent font or spacing changes produces split segments and extra tags in any CAT tool. Clean the DOCX before import and check the tag view after import, as with any converted file.

Is it safe to upload client documents to an online OCR service?¶

Smartcat, ChatsControl and the other cloud routes all process the file on someone else’s servers, so check the client’s NDA before you upload. ChatsControl deletes uploads after 30 days by default as of October 2026; if the contract rules out cloud processing, use desktop OCR instead.

Sources¶

1. Choosing the right PDF translation workflow in Smartcat - Smartcat Help Center, the three PDF workflows and the pdfimageonly route (updated 17 September 2026)
2. Understanding OCR pages and MT pages for freelancers - Smartcat Help Center, how OCR pages are counted and bought (updated 24 July 2026)
3. Translating PDF documents with AI - Smartcat Help Center, requirements and limits of AI PDF Translation (updated 31 August 2026)
4. Supported file formats - Smartcat Help Center, image formats accepted through OCR (updated 12 October 2026)
5. Image Translator - Smartcat Help Center, the AI Image Translator and its output format (updated 24 July 2026)
6. ABBYY and SmartCAT - MultiLingual, 3 April 2015, on ABBYY OCR built into SmartCAT