Diagnose an unsearchable PDF before running OCR

If you can see a word but search cannot find it, the reader is not necessarily malfunctioning. The page may be an image of text rather than stored characters. It may also contain a partial or damaged text layer that works on some pages but not others.

Do not begin by recognizing the whole document. First establish what kind of failure you have, page by page. That keeps good native text intact and gives you a smaller result to verify.

Test what the PDF actually stores

Use a sentence that is clearly visible and run three checks:

1. Selection: Can you highlight individual words, or does the pointer select one large image area?
2. Copy and paste: Does the pasted text match the page, or is it empty, garbled, or in an unexpected order?
3. Search: Can the reader find a distinctive word from that same sentence?

Repeat the test on the first page, a middle page, and any appendix or scanned attachment. Mixed PDFs are common: a digital cover sheet can contain usable text while the supporting pages are photographs.

A successful selection test is only the beginning. Multi-column layouts, unusual character mappings, and page drawing order can make copied text difficult to use even when search works. That is an extraction-quality problem, not proof that every page needs recognition.

Extract native text before choosing OCR

The PDF text extraction tool reads the native selectable layer already stored in the file. It keeps page boundaries visible and reports pages with no native text, then lets you copy the result or download a TXT file. It does not claim to reconstruct columns, tables, or positioned labels exactly.

Use its page-separated output as a diagnostic:

• If a page produces accurate text, keep that native layer.
• If a page produces nothing and the page is visibly a scan, it is a strong OCR candidate.
• If the output exists but important characters are wrong or scrambled, compare the source carefully before deciding whether targeted OCR would be more useful.
• If only a few pages fail, recognize those pages rather than processing the entire document by default.

The PDF and extracted text stay within the browser workflow, and the tool has no document upload endpoint.

Use OCR for image-only or unusable pages

The OCR and searchable PDF tool is designed for printed English text. You can select all pages or a validated range, review page progress, confidence, word boxes, line grouping, and recognized text, then create a PDF that preserves the visible page image while adding an aligned invisible text layer.

This is not a handwriting solution. Faint print, blur, rotation, decorative lettering, and complex layouts can lower confidence. A clearer source near the tool's 300 DPI working resolution is preferable when you can obtain one.

Document pages, recognized words, confidence data, and generated files remain in the browser tab. The first use may still request same-origin recognition resources, so “browser-local processing” should not be confused with a promise of zero network requests.

Verify searchability instead of trusting the success message

The OCR workspace reopens the generated PDF and checks its page count, dimensions, page images, selected-page text, and text-position bounds before enabling the download. Your own content check is still necessary because structural validity does not prove that every recognized character is correct.

Use this verification routine:

• Close the source and open only the generated PDF.
• Search for several distinctive words from different recognized pages.
• Try selecting a full sentence and paste it into plain text.
• Compare names, dates, totals, reference numbers, and punctuation with the page image.
• Review low-confidence areas and small print rather than sampling only the cleanest page.
• Run the extracted-text view again if you want to inspect the invisible layer without the page image influencing what your eyes read.

If search works on some pages but not others, return to the page range rather than repeating recognition across the whole file. If extraction already gives accurate native text, OCR is unnecessary. The useful question is not simply “Is this PDF searchable?” but “Which pages have a dependable text layer, and how did I verify it?”

Sources and further viewing

• Primary video: My CG Tutor