"The result looks wrong"
Summary
The app finished, you downloaded the file, and something is off: text cannot be selected, bookmarks are gone, a form changed, pages are sideways. Nearly every case comes from one of three causes: the app rebuilt the pages (so text became pictures or the reverse), an earlier step in your chain removed structure, or the input was not what you thought (a scan instead of a text PDF). Work down this page by symptom.
Details
Before anything else: compare with your original
Documentize never changes your original (Tasks, options and results). Open both side by side. If the original already has the problem, the fix is upstream. If only the result does, find the symptom below. Also check the order of your steps against Golden rules of ordering.
Text problems
| Symptom | Likely cause | Fix |
|---|---|---|
| Text is no longer selectable after Deskew | Deskew rebuilds each page as an image. Do not count on a text layer surviving | Run Searchable afterwards |
| Text not selectable after tiling in Resize | Tile mode converts pages to images | Tick OCR in Resize, or avoid tile mode |
The .txt from Extract Text is empty |
The PDF is a scan | Searchable first |
| Lines broken or columns mixed up | Reading order is approximate | Try Raw or Flatten in Extract Text, or Parser |
| Strange characters, wrong fonts | Fonts referenced instead of embedded, or a damaged encoding | Run Optimize or Recovery. For gibberish after repair, check the report's font repairs and run Searchable on the page images |
| Text empty or "gibberish" from OCR Image | Wrong language, sideways, or too small an image | See ocr-ai-quality.md |
| An ordinary text PDF returns almost nothing from Unwatermark | It reads text only and returns plain text, not your layout | Keep the original for the formatted version |
Bookmarks, contents and structure
- Bookmarks disappeared after merging or splitting. The structure-changing apps (Merger, Splitter, Delete Page) come first; build the outline last with Add TOC or TOC AI.
- A table of contents still lists a removed chapter. Delete Page does not edit the old contents. Rebuild it.
- TOC entries land on the wrong page. You typed printed page numbers, or you added pages afterwards. Count physical pages in the viewer. The book also flags that it is not certain whether Add TOC numbers refer to pages before or after the contents page is inserted; test on a short file.
- TOC AI gave bookmarks but no contents page. By design. Use Add TOC for a page.
- TOC AI missed headings. They look like body text; build entries by hand with Add TOC.
- Page numbers of Splitter do not match the printed numbers. Splitter uses physical page position.
Forms and annotations
- Fields are still editable after Flatten. The names did not match; copy them exactly (case counts) or leave both boxes empty to flatten all.
- Values vanished after Flatten. The field was empty when frozen. Fill first, flatten after.
- I cannot correct a typo after Flatten. Flatten is irreversible on that copy. Go back to your original.
- Extract Form Data gives headers but no values. The form was not filled in or was flattened. Extract before flattening.
- Annotations or comments gone. If your goal was to keep them, do not flatten.
Pages: orientation, size, order
- Sideways or upside-down scan pages. Rotate with Rotation. For the angle, look at where the top of the text sits: 90 degrees twice is 180.
- Slightly crooked page. That is Deskew territory, but it only fixes small angles. Below about 0.3 degrees, or with too little text, it leaves the page alone on purpose.
- Rotation changed nothing on "Auto Skew". Wrong language, or too distorted; try Deskew.
- Booklet pages upside-down on the back. Switch duplex mode or flip on the other edge in the print dialog (Booklet).
- A blank page in the middle of a booklet. Expected when the page count is not a multiple of 4.
- Merged in the wrong order. Merger orders by file, not page. Split first for page-level control.
- Extra blank piece after Splitter. A range exceeded the page count.
Pictures and file size
- Pictures blurry after Compressor. It is lossy for images. Go back to the original and use a lighter level. Never compress a compressed file again.
- File got bigger. Searchable adds a text layer, Deskew re-encodes pages as images, and Splitter repeats shared fonts per piece. Finish with Compressor.
- Compressor barely shrank it. The file was mostly text or already small. The panel reports the real saving, even if tiny.
- Pictures from Extract Image look poor. They are stored that way in the PDF.
Convert and create
- Word result has broken columns or an empty document. PDF to Word is reconstruction; for scans switch OCR on (Converter).
- Excel result looks like chopped text. Use "tables only".
- Office to PDF changes the page breaks. The file is rendered by a different engine than Office; review the result.
- GIF became a still image. PDF has no animation.
- PDF from Chat Create is one page. Set Pages above 1.
PDF/A
"Not valid" is common. The site's own list of causes: fonts referenced rather than embedded, colours without an output intent or ICC profile, transparency, encryption, JavaScript and external links. Fix and re-run; new errors appearing as old ones clear is normal. See PDF/A Validator and Converter (PDF/A target).
When the app says it could not repair
If Recovery reports that no stage produced an acceptable result, it lists the damage notes. Loosen the expected-page or text-coverage settings, or accept that too little is left. Recovery's default policy "Full" strips metadata and attachments; choose None or Keep attachments if you need them.