Free OCR tools — feature comparison

Step by step — OCR a scanned PDF

1. Pick the right tool for the document. Sensitive content → local (Tesseract / ScoutMyTool / Apple Live Text); public content where best accuracy matters → cloud (Google Lens / Drive).
2. Pre-process the input. Ensure 300+ DPI, correct orientation, even contrast. Phone-camera scans should use a scan-mode app (Apple Notes, Adobe Scan) rather than raw photos.
3. Run OCR. ScoutMyTool: drop PDF, pick language, generate. Tesseract command: `tesseract input.pdf output -l eng`. OCRmyPDF: `ocrmypdf input.pdf output.pdf -l eng`.
4. Verify accuracy. Cmd-F search a word you can see on the page — search should find it. Spot-check three pages by selecting text and comparing to the visual.
5. Save the searchable output. The OCR result is a new PDF (or text file) with text layer added; archive alongside the original.

Workflow-by-workflow recommendations

One-off receipt or business card capture: iOS Live Text or Google Lens — open camera, point, copy text. Zero setup, near-instant. For a batch of scanned PDFs (research papers, archived documents): OCRmyPDF on the command line — scriptable, handles dozens of files in one command, produces searchable PDFs in the same directory. For confidential single PDFs (legal, medical, financial): ScoutMyTool in the browser — client-side, no upload, fast enough for one-document workflows.

For a research corpus you query repeatedly, OCR once then index with a search engine (Recoll, DocFetcher) — the OCR time is one-time, the query speed is permanent. For occasional one-off OCR on a phone, the built-in tools (Live Text, Lens) are the right answer. For mass conversion of scanned archives, the command-line Tesseract path scales best. The right tool varies by volume, sensitivity, and how often you query the output.

OCR accuracy in practice — what to expect

Real-world OCR accuracy is dominated by input quality, not engine quality. A 300 DPI flat scan of crisp printed text gets 98–99% across all current engines. A phone photo of a slightly skewed receipt drops to 88–94%; a faded carbon-copy receipt drops to 75–85% even on the best engines. The way to improve OCR results is rarely to switch engines — it is to improve the input: rescan at higher DPI, apply contrast correction, deskew, crop to the document area. Most "OCR is bad" complaints I have seen trace to suboptimal input, not the engine itself.

For high-volume workflows where OCR feeds downstream processing (LLM input, data extraction, search indexing), accept that 1–5% of characters will be wrong even at best-case accuracy. Build the downstream pipeline tolerant of OCR errors — fuzzy matching, multi-character-class checks for amounts and dates, manual review of high-stakes extracts. The strategy of "perfect OCR" is unachievable; the strategy of "good-enough OCR with downstream error tolerance" is the working one.

Related reading

• Searchable PDF: making OCR results discoverable.
• Make a scanned PDF searchable: companion deep-dive on OCR mechanics.
• Scanned PDF to Word: OCR + paragraph reconstruction.
• Handwriting to text: the specifically-hard case of OCR.
• Multi-language PDF: OCR across scripts and writing systems.

FAQ

Which free OCR tool is most accurate?

For clean printed Latin script (English, French, German, Spanish, Portuguese): all of the leading free tools are within 1–2 percentage points of each other — 95–99% accuracy. Apple Live Text and Google Lens have a slight edge on hard cases (slanted text, faded paper, handwriting) because they use larger neural models. Tesseract is most accurate on clean print at the cost of being weaker on degraded inputs. For non-Latin scripts (Arabic, CJK, Devanagari), Google Lens is generally the best; Tesseract is close with the right language pack. The accuracy differences are mostly visible at the margins — for typical office documents, any of these works.

When should I use cloud OCR vs local OCR?

Cloud (Google Lens, Drive OCR, OneNote sync) is fastest and most accurate, especially for difficult inputs and non-Latin scripts. Local (Tesseract, OCRmyPDF, ScoutMyTool, Apple Live Text) is privacy-safe — your file never uploads. The trade-off is straightforward: use cloud for public or non-sensitive documents where you want best accuracy; use local for confidential content (HR records, financial statements, client work under NDA). Both can run in the same workflow — local for the sensitive batch, cloud for the routine batch.

How do I install Tesseract on Mac / Windows / Linux?

Mac: `brew install tesseract` (with Homebrew) or `port install tesseract` (MacPorts). Windows: download installer from the Tesseract GitHub releases. Linux: `apt install tesseract-ocr` (Debian/Ubuntu) or `dnf install tesseract` (Fedora). After install, add language packs as needed: `brew install tesseract-lang` (all languages) or download specific language data files from the tessdata GitHub repo. Test with `tesseract input.png output -l eng` — produces `output.txt` with extracted text.

Can I run OCR in the browser without uploading my PDF?

Yes. ScoutMyTool Make PDF Searchable uses Tesseract via WebAssembly — Tesseract compiled to run in browser JavaScript engines. The first OCR run downloads the ~10–30 MB language pack to the browser cache; subsequent runs are fast. Your PDF is read and processed entirely in the browser tab; the tool has no server-side processing for OCR. Memory usage scales with document size — laptops handle 200-page PDFs comfortably; very large documents (1,000+ pages) may exceed browser tab memory and need to be chunked.

How do I OCR a handwritten document?

Handwriting OCR is genuinely harder than printed-text OCR and Tesseract is weak at it (50–80% accuracy depending on handwriting style). For handwritten content, Apple Live Text, Google Lens, and dedicated services (Microsoft Azure AI Vision, Google Document AI) significantly outperform Tesseract. The neural models behind cloud OCR were trained on far more handwriting examples than Tesseract. For occasional handwriting OCR, snap a photo with iOS or Android, let Live Text or Lens extract — accuracy will exceed any Tesseract-based local tool. For batch handwriting OCR at scale, the Azure or Google Document AI APIs (paid) are the right tools.

Citations

1. Tesseract OCR documentation — open-source OCR engine maintained by Google.
2. OCRmyPDF documentation — open-source command-line wrapper around Tesseract.
3. Apple — Live Text documentation in macOS Sequoia and iOS.
4. Google Lens — official Google Lens product documentation.
5. Microsoft Azure AI Vision — paid OCR service documentation (for handwriting comparison).