The dimensions that separate the tools
Tashkeel: preserved or stripped?
Footnotes: in place or merged into the matn?
Text direction after export to Word
Arabic-Indic numerals and manuscripts
How to test any OCR tool yourself
- 1
Pick a hard page from your text
A vocalized page, a footnote-heavy layout, or one with Arabic-Indic numerals and hijri dates — not a clean page every tool reads well.
- 2
Run it and inspect the output
Did the tashkeel survive? Are the footnotes separated from the body text? Did ١٢٣ stay ١٢٣ or silently become 123?
- 3
Export to Word and edit
The decisive test: delete a word and type a new one. If the line reverses or the letters disconnect, the text layer is broken no matter how good it looked on screen.
What usually gets lost — and what survives
A vocalized classical page with a footnote: how it comes out of a generic tool, and out of a tool built for Arabic.
ﻢﻠﻌﻟا ﺐﻠﻃ ﻞﻌﺟو ،تﺎﺟرﺪﻟا ﻢﻠﻌﻟا ﻞﻫﻷ ﻊﻓر يﺬﻟا ﻪﻠﻟ ﺪﻤﺤﻟا (1) تﺎﺑﺮﻘﻟا ﻞﻀﻓأ ﻦﻣ ﻲﻓ (١) ﺎﻬﺸﻣﺎﻫ ﻲﻓ ﺎﻣ ﺎﻨﺘﺒﺛأو ،«ﻰﻠﻌﻟا» ﻞﺻﻷا ﺔﺨﺴﻧ
Reversed direction · diacritics stripped · footnote merged into the body
الْحَمْدُ لِلَّهِ الَّذِي رَفَعَ لِأَهْلِ الْعِلْمِ الدَّرَجَاتِ، وَجَعَلَ طَلَبَ الْعِلْمِ مِنْ أَفْضَلِ الْقُرُبَاتِ(١).
(١) في نسخة الأصل: «الدَّرَجَاتِ الْعُلَى»، وأثبتنا ما في هامشها.
The comparison, on what matters for Arabic
A fair comparison: free tools do real work in many cases. The right question is not "what is best overall?" but "what is best for your text?"
| Dimension | Free tools (Google Docs, Tesseract, PDF24) | Generic cloud engines | ScribeTools |
|---|---|---|---|
| Clean modern print | Genuinely good — fast, free, often all you need | Very good | Very good — but you may not need us here |
| Tashkeel (diacritics) | Usually stripped or scrambled — marks treated as noise | Inconsistent; degrades as vocalization gets denser | Built for vocalized text — marks stay on their letters |
| Footnotes vs. body text | Footnotes merge into the matn or vanish | Read as one continuous block, no layer separation | Footnotes separated in place, with their numbering |
| Text direction in Word export | Often reversed; letters disconnect when you edit | Needs manual cleanup after export | Clean RTL Word file that survives editing |
| Arabic-Indic numerals & hijri dates | Converted to Western digits or scrambled | Varies by engine | Kept exactly as in the original |
| Manuscripts & handwriting | Weak in general | Inconsistent; designed for modern documents | Supported, with results that vary by hand clarity — we don’t claim it’s solved |
| Speed & cost | Instant and free — their strongest suit | Cloud account and developer-style pricing | 20 free credits monthly, then packs from $3.99 for ~100 pages |
If your text is clean, modern, unvocalized print — try the free tools first; they may be all you need. The table above is descriptive, from working with these tools, not a quantitative measurement.
Accuracy measurements: coming soon
We are building a public benchmark for Arabic OCR accuracy on vocalized classical texts and legal documents, with a published methodology and reproducible samples. Until it ships we publish no accuracy numbers — ours or anyone else’s. When it does, the results table will live here.
What scholars and researchers say
“I just used this for a file that wouldn't OCR properly at all. It's worked perfectly… it did a better job than other tools out there.”
“This is such an amazing resource!!! …Checked it and worked great!”
“This could be a game changer in the future study of Arabic font texts…”
Frequently asked questions
- What is the best OCR for Arabic?
- There is no single best tool — the honest answer starts from your text. Clean, modern, unvocalized print is handled well by free tools like Google Docs, which cost nothing and run instantly. Vocalized classical texts, footnote-heavy pages, documents with Arabic-Indic numerals and hijri dates, and manuscripts need an Arabic-specialist tool like ScribeTools, built for exactly those cases and exporting to Word with the text direction intact and editable. Run a hard page from your own text through more than one tool and compare the results yourself.
- Are free OCR tools good enough for Arabic?
- Often yes, and pretending otherwise would be unfair: Google Docs, Tesseract, and PDF24 read clean modern print well, free, instantly, with no signup. They break on what makes classical and official Arabic text distinctive: diacritics are treated as visual noise and stripped, footnotes merge into the body text, Arabic-Indic numerals get converted, and the text direction reverses when you export to Word. If your text is the first kind, save your money. If it is the second, a specialist tool saves you the hours of manual cleanup instead.
- Why do OCR tools strip Arabic diacritics (tashkeel)?
- Because most engines are trained on unvocalized text, so they treat fatha, damma, kasra, and shadda as visual noise around the letters rather than meaningful marks — and either drop them or misread them as stray characters. Vocalized text needs models that recognize each mark as part of the word and encode it as proper Unicode attached to its letter. That is what ScribeTools is built around: vocalized classical text is the core design case, not an edge case, so the marks come out in Word editable and searchable.
- What is the best OCR for Arabic manuscripts and handwriting?
- Manuscripts are the hardest OCR case there is, and we know of no tool — ours included — that makes handwriting a solved problem. Generic free tools are weak here because they are designed for print. ScribeTools supports manuscripts with results that vary by hand clarity, scan quality, and marginalia density; the premium engines (2–4 credits per page) improve accuracy on complex layouts and handwriting without reaching perfection. Try a few pages of your manuscript on the free credits first, and collate the output against the original before relying on it.
- Do you publish accuracy numbers?
- Not yet — deliberately. The accuracy percentages common in this market are measured on undisclosed samples with unpublished methodologies, so they neither support comparison nor survive serious scrutiny. We are building a public benchmark for Arabic OCR accuracy on vocalized classical texts and legal documents, with published, reproducible samples and methodology. When it ships, its numbers will be published here exactly as measured — and until then we publish no percentages at all. In the meantime, test the tool on your own text with the free credits: your text is the benchmark that matters to you.
- How do I try ScribeTools on my own documents?
- Create a free account and you get 20 credits every month with no credit card — a page costs roughly one credit on the default engine. Upload hard pages from your own text (a PDF, or JPEG/PNG/TIFF images), choose "Editable Word document", and inspect the result: the tashkeel, the footnotes, and how the text behaves when you edit it in Word. If the output convinces you, one-time credit packs start at $3.99 for 100 credits and never expire.