Disclosure: This article was substantially drafted with GPT-6.1 Terra for Yan Wang / LensUp, with source and fact editing by Codex AI. Product descriptions and implementation source were inspected; the examples and checks below are proposed, not executed PDF tests or a security audit. Yan Wang is affiliated with LensUp. Polished by an independent editorial AI (lensup 主编); Codex fact-checks the text before publication. No human editor or owner reviewed or tested it.
When text in an exported PDF can't be selected, that is either the intended result or a defect. Which one depends on the contract chosen before the download.
Take a document with selectable paragraphs on page one and a small covered phrase on page two. You might expect an edit to one region to leave the unrelated text alone. A whole-file image export makes a different trade: every page in the output is a picture, including pages with no cover. If the requirement calls for this image-only route, losing the original selectable text on page one is expected behavior. A clipped paragraph on that same page is a separate visual failure.
The inspected LensUp Remove it path renders every page, paints covers last and builds a new PDF from page pictures. The implementation does not carry the original interactive links, forms, bookmarks, attachments or source metadata into that new document. This describes the inspected source, not a runtime finding about a particular file. Keep as is is the visual-cover choice and keeps the underlying content.
Make the acceptance decision explicit
Before you build a test, ask what the recipient needs. If the recipient must select original text or use an interactive form, the whole-file picture path conflicts with that requirement even when it looks correct. If the task requires a picture-based copy, loss of original text selectability elsewhere should not be treated as a regression by itself. Either way, the choice is not a hidden-data check.
A useful proposed fixture has three pages: selectable text and a link on the first, a covered dummy phrase and a form field on the second, and a scanned image on the third. Give each page a visible label. Reopen the exact downloaded file and apply this matrix; these are expectations to investigate, not measured results:
| Check | Picture-based contract | If the requirement instead needs original functionality |
|---|---|---|
| Visible content and cover | Intended content legible; cover in the right place | Same visual requirement |
| Page order | Labels appear in the intended order | Same order requirement |
| Source-text selection | Original text is not expected to remain selectable across the file | Loss of required selectable text is a mismatch |
| Link and form interaction | Original interactive objects are not expected to carry over | Missing required interaction is a mismatch |
| Later OCR | Record whether a separate process creates text from visible pixels | Evaluate that text against the recipient's actual needs |
Test page appearance and interactivity separately. A picture of a form field can look filled and still accept no input, and a visible link may no longer be clickable. For OCR, record the viewer or processing step. Searchable words created later from visible pixels are a new recognition result, not evidence that the original PDF text layer survived. OCR can be inaccurate, and it should not be assumed to reconstruct pixels hidden by a cover.
If the copy needs new visible text on a cover, add it after the picture-based export and check that it is readable.
Top comments (0)