

If your scanned PDF is not searchable, it may contain page images without a usable text layer. Optical character recognition, or OCR, adds machine-readable text to an image-based document. Search can still fail when the text is incomplete, the wrong language was used or the archive has not indexed the file.
Start by separating PDF search from archive search. A word that cannot be found inside the PDF points to a different problem from a word that works in the PDF but not in the repository.
Open the document in a PDF reader and try selecting a printed sentence. Copy it into a plain text editor. If you cannot select words, the document may be image-only. If the pasted text is garbled, the text layer may be inaccurate or incorrectly encoded.
Test several pages. A document can contain both searchable and image-only pages after files have been combined. Use known words from different sections rather than relying on one successful search.
| Symptom | First check |
|---|---|
| No text can be selected | Whether OCR ran and a text layer was included in the export. |
| Some pages work | Whether all pages were processed with the same workflow. |
| Names and numbers are wrong | Image quality, language settings and characters that look alike. |
| Search works in the PDF only | Repository indexing status, file support and access permissions. |
| Handwriting produces poor results | Whether a suitable handwriting-recognition workflow was evaluated. |
Inspect focus, rotation, faint print, shadows and text lost near the binding. Processing the same poor image repeatedly may reproduce the same errors. Reprocess from a better master or recapture the page where the original can be handled safely.
Check the recognition language and orientation. A multilingual collection may need different settings across batches. Keep the original image available while testing corrections; do not overwrite your only master with an experimental output.
OCR may confuse characters such as O and 0, or I and 1. Those errors matter when staff search for a contract number, student identifier or invoice reference. Build a short test list with known values and compare the results against the original.
Use structured metadata for important identifiers instead of depending entirely on full-text recognition. Our guide to searching thousands of scanned pages explains how searchable text and indexing work together.
If local PDF search succeeds, ask the administrator whether the repository has processed the latest file. Confirm that the file type is supported and that the test account has permission to see the record. Check whether the search tool is looking at document contents, metadata or both.
After replacing a file, follow the system’s supported reindexing procedure. Retest with the same known words and an ordinary user account. Record which version passed so the corrected output can be identified later.
Keep a copy of the problem file and record the viewer, page number and exact search term that failed. Confirm that the expected word is actually visible on the page. A user may be remembering a different version, spelling or document, so begin with a value you can verify directly.
Try the same file in a second approved PDF reader if one is available. This can help distinguish a file problem from a particular viewer’s search behaviour. Avoid uploading confidential records to an unfamiliar online tool simply to test them; use the organisation’s approved environment.
Then work through the layers in order: visible image, selectable text, recognised characters, PDF search and repository search. Record the first stage that fails. This creates a useful support request and reduces the chance of changing several settings without knowing which change addressed the issue.
Select representative pages and prepare a list of words or identifiers that should be present. Include a clear printed heading, a name, a date, a reference containing letters and numbers, and a term near a difficult margin. Record the correct values from the visible original.
Run the test before and after a proposed correction. If the heading becomes searchable but a reference remains wrong, the result is a partial improvement. Keep that distinction visible rather than marking the entire collection “OCR complete” after one successful search.
The test pack should reflect the material. A collection of typed business letters will not reveal the same recognition issues as old newspapers, tables or handwritten registers. Maintain separate sample groups when their characteristics differ substantially.
| Observed result | What to investigate |
|---|---|
| A name has unexpected spaces | Word segmentation, page layout and the quality of the text layer. |
| A reference contains O instead of 0 | Character recognition and whether a verified metadata field is needed. |
| A word split across two lines is missed | How the OCR output and search tool treat hyphenation and line breaks. |
| Only the left side of a spread is searchable | Processing coverage, page regions and capture/export settings. |
| A replaced document returns old results | Whether the repository has indexed the replacement version. |
These symptoms suggest investigation paths; they do not prove a single cause. Inspect the actual text and the relevant software settings before deciding whether to reprocess, recapture or correct an index field.
A text layer can contain words from a table without preserving their relationship to the correct row or column. If the business task is to retrieve a record by a name and associated reference, test that relationship explicitly. Searchability alone does not establish that extracted data is ready for a business system.
For handwriting, distinguish the image, any automated transcription and a human-verified index. Ask what level of review is needed for the intended use. A discovery aid may tolerate uncertainty that would be unacceptable when a value is used for a consequential decision.
Keep uncertain readings identifiable. Do not silently replace an ambiguous character with a guess and present it as a verified transcription. Where correction is authorised, retain the relationship to the source page and the review record.
Suppose a records team cannot find a supplier reference in a scanned agreement. The visible reference is clear, and copying the text reveals that the OCR layer has read a zero as the letter O. Searching the exact visible value therefore produces no result even though other words on the page are searchable.
The team first checks whether a better source capture would address the recognition problem. It then reviews the agreed correction workflow and adds the verified contract reference to the document’s structured index where required. A reviewer checks the corrected value against the visible source before acceptance.
If the corrected PDF works locally but the archive still fails, the administrator checks the indexed version and follows the system’s supported reprocessing procedure. The team retests with an ordinary user account and records the result. This is an example of a diagnostic process, not a claim that every reference problem has the same remedy.
Include the document identifier, authorised storage location, affected page, expected term, observed result and tests already performed. State whether the problem occurs in the PDF reader, repository or both. Mention when the file was replaced or processed, if known.
Provide an approved sample only through the organisation’s permitted channel. A screen image may show a symptom, but the technical team may need the actual file to inspect the text layer. Access to that file should follow the records owner’s instructions.
After the issue is resolved, add a representative test to the acceptance checklist for future batches. That turns one troubleshooting exercise into a repeatable quality check and helps the team detect the same defect before a large collection is handed over.
Consider the purpose of the search. Someone exploring a collection by broad subject may benefit from imperfect full-text discovery. A records officer looking for one exact identifier may need a checked metadata value that points to the correct document.
Use that distinction to scope correction work. Reprocessing an entire collection is a different decision from correcting a small set of high-value index fields. Test the proposed remedy on representative material and compare the improvement against the task users need to complete.
Keep the visible source available for verification and label the status of any transcription according to your procedure. A useful search result helps a person find the evidence; it should not conceal uncertainty in the extracted text. Document the chosen approach so later operators know when to escalate an ambiguous value.
No. Accuracy depends on the source and processing. Verify important details and use manual indexing or review where an error could affect a decision.
For many records workflows, both are useful. The image preserves the visible page, while the text supports discovery. Agree your master and access formats before production.
Need a searchable document workflow? Explore Eemediba’s document scanning services or request an assessment. Describe the record type and the search problem; avoid sending confidential files through an unsecured enquiry.