What OCR PDF does, and why page selection matters
A PDF can look identical on screen and still work completely differently underneath. Some PDFs carry real, selectable text; others are just a scanned photo or a flattened image with nothing to select. OCR ("optical character recognition") reads the pixels of a page and turns the shapes it finds into actual, computer-readable text, which NexKit then adds back onto the page as an invisible layer - the page keeps looking exactly like the original, but the text underneath can now be selected, copied, and searched.
Most real documents are not uniformly one type or the other. A scanned contract might have a typed cover letter, three photographed pages, and a table that was pasted in as an image. OCRing a page that already has good text is not harmless - it can leave two overlapping text layers on the same page, which shows up as garbled double text when you try to select or search it. That is why NexKit analyzes every page before doing anything else, and only recommends OCR for the pages that actually need it.
How to tell if your PDF needs OCR
The fastest check: open the PDF in any viewer and try to select a line of body text with your mouse. If a normal text selection highlights individual words, the page already has a text layer and does not need OCR. If your cursor just draws a selection box over what looks like text, or selects nothing at all, that page is image-only and OCR is what adds real text to it.
NexKit does this check automatically for every page as soon as you upload a PDF, using the same signal (the PDF's own text layer, read directly - no OCR needed for pages that already have one) plus a check for pages whose text layer exists but decodes to garbled or mismatched characters, which happens with some PDFs that were produced by unusual export tools. You will see this as a per-page breakdown before you commit to running OCR on anything.
Text, scanned, and mixed PDFs
A "text" PDF has a real text layer on (nearly) every page - exported from a word processor, generated from HTML, or previously OCR'd. A "scanned" PDF is pages of raster images with no text layer at all, typically produced by a scanner or a phone camera. A "mixed" PDF combines both within the same document, and sometimes within the same page, such as a scanned photo that already carries an invisible OCR text layer from an earlier pass, or a typed page with an embedded image in the margin.
NexKit classifies at the page level, not just the document level, because a single mixed PDF is common in practice: a cover page typed directly, followed by scanned attachments. Treating the whole file as one category would either skip pages that need OCR or needlessly re-OCR pages that already work.
How NexKit selects which pages to OCR
After analysis, NexKit shows a recommended action - skip OCR entirely if every page already has usable text, OCR just the pages that need it, or OCR the whole document if none of it has real text - along with the page count, how many pages already have text, and any pages flagged for a suspect text layer or a complex multi-column or table layout (complex-layout detection is a rough heuristic and mainly useful as a hint, not a precise guarantee).
You can accept the recommendation, switch to "All pages" to OCR every page regardless of what it already has, select a custom page range, or choose "Force OCR" to deliberately re-run recognition on pages that already have text (useful if an existing text layer is known to be bad). Whatever you choose, pages you do not select are copied through untouched - their existing content and any existing text layer are left exactly as they were.
Chinese OCR accuracy: what to actually expect
Simplified and Traditional Chinese use different trained character models, and NexKit treats them as separate languages you choose explicitly (selecting both when a document mixes scripts, which does happen with some historical or Hong Kong/Taiwan-sourced material, works but is slower and can occasionally confuse similar-looking characters between the two scripts). Recognition quality depends heavily on scan resolution and font: dense, small printed text, ornate or handwritten-style fonts, and low-contrast or skewed scans all measurably reduce accuracy compared to a clean, high-resolution, horizontally aligned scan of a common printed font.
Do not treat OCR output as a substitute for proofreading, especially for Chinese text, where a single misrecognized character can change a name, a date, or a number. Always preview the recognized text and spot-check it against the original page image before relying on the result for anything where exact accuracy matters - contracts, IDs, or financial figures in particular.
OCR vs. plain text extraction
OCR and plain text extraction solve different problems. Text extraction (used by tools like PDF to Word) reads text that a PDF already has and reflows it - it is fast and exact, but it has nothing to work with on a page that has no text layer. OCR is the opposite: it works from the rendered pixels of a page, which means it works on scans and photos, but the result is a best-effort recognition, not a guaranteed-exact readout of what was printed.
If your PDF already has real text everywhere - most PDFs exported directly from software do - OCR PDF will tell you so and recommend skipping it; go straight to PDF to Word or plain text extraction instead. Reach for OCR PDF specifically for the pages that are scans, photos, or flattened images.
How the searchable text layer works
For each page NexKit runs OCR on, the recognized words are positioned and drawn back onto that exact page location as text with zero opacity - invisible to the eye, but present in the page's content stream at the correct coordinates, matching the page's stored rotation. A PDF viewer's search and text-selection features read this layer directly, so "Find" and copy-paste work as expected, while the page still displays your original scan pixel-for-pixel.
This is why file size grows only modestly after OCR (a small amount of positioned text per page, not a re-encoded image) and why the visual page never changes, even if the recognized text itself contains errors.
Processing entirely in your browser
Page analysis and OCR recognition both run locally using your browser's own processing power - your PDF, its rendered page images, and the recognized text are never uploaded to NexKit servers at any step of this workflow. Heavier processing (the PDF renderer and the OCR engine) runs off the main thread so the page stays responsive, and processing happens on real page-by-page progress, not a simulated bar.
Because everything runs locally, results depend on your device: OCRing many pages, or choosing the High accuracy preset on a large file, takes noticeably longer on an older phone than on a desktop. Keep the browser tab open and avoid switching away from large jobs until they finish.
Supported scope and current limits
This tool currently supports PDFs up to 25MB and 30 pages, and English, Simplified Chinese, and Traditional Chinese recognition, in Fast, Balanced, or High accuracy quality presets. Password-protected and encrypted PDFs are rejected with a clear error - remove the password first with a PDF editor, if you have the password, then try again. Corrupted or empty (0-page) PDFs are also rejected with a clear error rather than failing silently.
Handwriting, decorative or highly stylized fonts, very low-resolution scans, and languages outside English/Chinese are outside what this OCR engine reliably supports today.
Troubleshooting
If OCR stalls or fails partway through a large PDF, cancel and retry with the Fast preset first, or split the PDF into smaller batches with Split PDF and OCR each part separately. If a specific page fails, you can retry just that page without redoing pages that already succeeded.
If recognized text looks wrong or garbled, check the source page: very low-resolution scans, heavy skew, or unusual fonts all reduce OCR accuracy - try re-scanning at a higher resolution if possible. If a page you expected to be skipped got flagged for OCR anyway, it likely has a text layer NexKit judged unreliable (for example, decoding to mismatched characters); Force OCR is there specifically for that case.
Frequently Asked Questions
Does this change how my PDF looks?
No. The original page images are kept as-is on every page, whether OCR'd or not. Only an invisible, searchable text layer is added on top of OCR'd pages, so the page still looks like the original.
Which languages are supported?
English, Simplified Chinese, and Traditional Chinese, selectable individually or in combination. Traditional and Simplified Chinese use separate trained models, so pick the one that matches your document (or both, if it genuinely mixes scripts).
Will OCR PDF re-OCR pages that already have text?
Not by default. NexKit analyzes each page first and recommends OCR only for pages that need it, skipping pages with a usable existing text layer to avoid duplicate text. You can choose "All pages" or "Force OCR" to override this and OCR every page regardless.
How does NexKit decide which pages need OCR?
It reads each page's existing text layer directly (no OCR needed to check this) and checks whether it has enough real, decodable text. Pages with little or no usable text, or with a text layer that decodes to garbled characters, are recommended for OCR; pages with good existing text are recommended to skip.
Are my PDF files uploaded?
No. Page analysis and OCR recognition both run locally in your browser. The file, its rendered page images, and the recognized text are not uploaded to NexKit servers by this tool.
Does it support password-protected PDFs?
No. Password-protected and encrypted PDFs are rejected with a clear error. Remove the password first, if you have it, then try again.
How many pages can I OCR at once?
This tool supports PDFs with up to 30 pages and files up to 25MB, since OCR takes more processing per page than other browser-based PDF tools.
Will the recognized text be accurate?
Accuracy depends on scan quality, resolution, font, and language - clean, high-resolution scans of common printed fonts do best. It is never guaranteed to be perfect, especially for Chinese text, so always preview and spot-check the result before relying on it for anything where exact text matters.
What are the Fast, Balanced, and High accuracy presets?
They trade processing time for recognition detail by changing the image resolution OCR runs on and how the page is segmented. Fast is quickest and best for large batches of clean pages; High accuracy renders pages at a higher resolution and is worth the extra time for small or dense text.
Can I get the text without a searchable PDF?
Yes. After OCR finishes, you can preview the recognized text, copy it to your clipboard, or download it as a plain .txt file, in addition to (or instead of) downloading the searchable PDF.
What happens if OCR fails on one page?
Other pages still complete normally. The failed page is flagged and you can retry just that page without re-running OCR on pages that already succeeded.