What's the best OCR for Arabic manuscripts?
For clean modern Arabic print, free tools like Google Docs are often enough. For manuscripts and classical books — with tashkeel, footnotes, and fragile layouts — use an Arabic-specialist tool such as ScribeTools: it preserves diacritics, keeps footnotes out of the main text, and exports RTL Word files that survive editing.
Why there is no single answer
“Arabic manuscripts” covers two quite different jobs: scanned copies of printed classical books (the most common case in taḥqīq and research work), and true handwritten manuscripts. Tools that handle one can fail badly on the other, so the honest starting point is your text, not a product name.
The practical question is not “which engine has the highest raw accuracy?” but: does the output preserve what makes the text usable — diacritics, footnotes, Arabic-Indic numerals, and right-to-left direction that survives export to Word?
When free tools are genuinely enough
If your source is clean, modern Arabic print — no tashkeel, no footnotes — try Google Docs first: upload the image or PDF to Google Drive, open it with Google Docs, and you get respectable OCR for free. There is no reason to pay for what a free tool already does well.
Free and general-purpose tools break down at four specific points:
- Tashkeel (diacritics): the marks are treated as noise and dropped, or misread as separate letters.
- Footnotes: they bleed into the main text (matn), turning separation back into manual work.
- RTL in Word: text pastes reversed or with broken letter joining, especially where Arabic mixes with numbers or Latin words.
- Arabic-Indic numerals and hijri dates: silently converted or misread — a real problem in dated classical texts.
What to test before committing to any tool
- Does tashkeel survive OCR — and survive export to Word?
- Do footnotes come out separated from the main text?
- Does the Word file open right-to-left with correct letter joining, no manual fixing?
- Are Arabic-Indic numerals and hijri dates untouched?
- Can you process a whole book in one batch, or only page by page?
Run this test on one hard page — dense footnotes or fully vocalized poetry — before feeding in a 400-page book.
Printed classical books vs. handwritten manuscripts
The distinction matters more than any tool choice. Scanned printed classical books — Bulaq and Halabi editions, lithographs, early 20th-century typesetting — are where OCR genuinely works today: the letterforms are regular enough to read, and the difficulty lives in vocalization, footnotes, and aged print rather than in the script itself. True handwritten manuscripts (naskh, taʿlīq, riqāʿ hands) are a different problem: premium engines produce a useful first draft on clear, dotted hands and degrade on dense or damaged ones. A good draft saves you transcribing from zero — it does not eliminate collation against the original.
Where ScribeTools fits
ScribeTools is built specifically for the Arabic that breaks general tools: vocalized classical text, footnote-heavy layouts, and Saudi legal documents. Instead of relying on one engine, it runs each document through multiple OCR engines and vision LLMs together and picks the best result, then exports Word files that open in the correct direction with footnotes and diacritics in place.
On handwritten manuscripts, honesty is due: no tool — ours included — reads handwriting reliably enough to skip review. Premium engines improve results on clear naskh-style hands and complex layouts, but collating against the original remains part of the scholar’s job, not a step software removes.
A practical way to decide today
- Pick the hardest page in your book — dense footnotes, vocalized poetry, or a page with matn and commentary together.
- Run it through the free option first and see what survives of the diacritics and footnotes.
- Run the same page through a specialist tool and compare the two results in Word — not on screen, where direction problems hide.
- Whatever passes on your hardest page will handle the rest; convert the book in one batch and review.
What about cost?
The real comparison is not with free tools but with human transcribers: freelance تفريغ (transcription) plus collation currently runs $1–5 per page and takes weeks per book. ScribeTools starts free with 20 credits monthly (about one page per credit on the default engine), no credit card, and one-time credit packs never expire — so test your hardest pages first and decide from evidence.