Check whether scanned PDFs can be searched

Practice relevance

Cross-practice idea · A general operational method. Review the stated tool and jurisdiction limits before use.

The problem

Find image-only pages, broken OCR, and missed sample strings without overwriting originals.

Why it matters: A PDF may appear searchable while its text layer omits words or misreads numbers. Testing known strings and inspecting suspect pages helps staff understand what search can miss without treating OCR as authoritative text.

Sources & evidence

Proposed / not tested. The public source describes the problem. Documentation supports only the stated method components; no result from this recipe has been verified.

What’s been verified

Source methods and reported problems have been reviewed. We have not run the adapted workflow. The sample below describes expected behavior on synthetic data, not a customer result.

What to measure: In a synthetic pilot, count correctly found exceptions, false alarms, and unresolved inputs. No time or cost saving is assumed.

Pain evidence: Large-document PDF search help ↗Large-document PDF search help, 2022-06-28, older: a paralegal receiving extensive PDF discovery asks how to search it. Adobe Help, 2024-04-01, provides a specific failure: combining a transcript and scans makes OCR insert spaces between letters, defeating search. OCR questions, 2022-04-08, reports number-recognition errors and large-document crashes.
Pain evidence: Adobe Help ↗Additional self-reported public account; no independent identity or outcome verification.
Pain evidence: OCR questions ↗Additional self-reported public account; no independent identity or outcome verification.
Method: Adobe: fix text recognition errors ↗Adobe: fix text recognition errors documents recognized-text review and suspect-word correction.

Source checked: October 15, 2026. Pain accounts are self-reported. Documentation supports components; this adapted recipe and prompt have not been tested. We wrote these prompts. The adapted workflow hasn’t been tested.

Suggested approach

Start when: A scan batch or combined reference PDF is received

inspect text-layer coverage and test known sample phrases/numbers
Keep source references and report unknowns instead of filling gaps.
The responsible reviewer checks the report before any separately approved action.

Limits and review

The assistant drafts or reports only. Check source records and review the output before a human performs any separately approved send or change. No automatic legal decision, money movement, account connection or publication is included.

Who reviews it: Staff compare suspect text with rendered originals, especially dates/numbers. Test mixed digital/scanned pages, rotated scans and deliberately degraded text. Never overwrite evidence originals.

What you need

Excel Copilot, optional; or manual review · prepared inputs

Existing Acrobat desktop with Scan & OCR access (use the firm’s approved licensed installation).
Working copies of synthetic/public PDFs and a reviewer who can read the original images. Never overwrite evidentiary originals.
Optional firm-approved Excel Copilot access for organizing the deidentified test log; native OCR/search tests work without AI.
Use synthetic inputs first. Keep real data in firm-approved accounts and processing, with appropriate access, confidentiality and retention controls. Aliases alone do not anonymize records. Never paste credentials, API keys or unnecessary client details.

How to use

Documentation-verified native OCR and manual QA route; no automatic whole-file AI certification

Before you start

The kit extraction.csv is a manually captured page/test log here, not a documented Acrobat one-click per-page CSV export.
Acrobat AI answers are not proof that every page has a text layer or that OCR is accurate. Whole-file coverage requires checking every relevant page; sampling must be labeled sampling.

Check access and setup

Acrobat edition/version and OCR tool access were not tested.
No native per-page extraction CSV or automatic mixed-page completeness detector was verified.

Create the test baseline
Acrobat desktop and a text editor · Open a separately named working copy; Select text tool
Before OCR, record file_id, source_version and displayed page_count. Visually read a distinctive phrase and, where available, an important number/date on each tested page; enter each as a separate known-strings.csv row. For each page, select and copy all available page text to a plain-text editor and store exactly ONE aggregate extraction.csv row for that file_id/page. Record complete_page_capture, partial_page_capture or unselectable/empty in extraction_status. If only part was copied, record that scope in extraction_status. All known strings for a page will be checked as substrings of that one captured text value.
Expected output
files.csv and known-strings.csv plus initial extraction.csv rows (file_id,page,extracted_text,extraction_status). All are manual QA logs.
Next step
Keep the original and baseline log unchanged; process only the working copy.
Check before continuing
Distinguish a PDF page index from printed/Bates page labels. Copy failure may reflect restrictions, not just image-only content; stop rather than bypass restrictions.
Step documentation 1 ↗
Run OCR on the working copy
Acrobat desktop · All tools > Scan & OCR > In this file
Choose the intended page range and document language, then Recognize Text. For a batch use In multiple files > Add files and select a separate output folder/filenames. Preserve existing digital-text pages rather than rewriting them indiscriminately.
Expected output
Searchable working PDF with an OCR text layer where processing succeeded.
Next step
Reopen the output and rerun the baseline copy/search tests.
Check before continuing
Record OCR errors, skipped pages and the chosen language/page range. A completed OCR job is not a passing QA result.
Step documentation 1 ↗
Review OCR suspects and test known strings
Acrobat desktop · Scan & OCR > Correct recognized text > Review recognized text; Find/search in the PDF
Compare each flagged word to its image. Correct only the OCR working-copy text when unambiguous; use Accept to advance. Independently search each known phrase/number and copy the result into the test log, including spaced-out words or digit mismatches. Keep one aggregate post-OCR extraction row per file_id/page and compare each expected string against that row. Never silently pair several per-page extracts by order.
Expected output
Per-tested-page pass/fail/uncertain evidence and corrected working-copy text where justified.
Next step
Populate extraction.csv from the observed post-OCR text and statuses.
Check before continuing
Review even when no suspect is flagged. Test coverage must list pages checked and unchecked; a failed search does not prove absence of source content.
Step documentation 1 ↗Step documentation 2 ↗
Summarize the actual test results
Excel Copilot, optional; or manual review · Alias-only QA workbook > Copilot icon at lower right > Chat only
Place the deidentified known-string and one-row-per-page extraction logs in an Excel workbook. Open Copilot in Chat only and paste the prompt, or manually search each expected string within the corresponding extracted_text cell.
Excel Copilot, optional; or manual review promptUse only these manually observed PDF searchability logs. Join expected strings to exactly one aggregate extraction row using file_id and page; report duplicate extraction rows as an error. Check each expected phrase/number as a literal substring of that page’s extracted_text, not as equality with the whole text cell. Flag mismatches, spaced-out words, empty extraction and partial/untested page coverage. Quote the expected string and observed evidence. Return file_id,page,expected_string,observed_excerpt,issue,review_action and coverage count. Do not certify the whole PDF, infer untested content, interpret legal substance, edit the workbook or alter PDF files.
Expected output
Exception queue with file_id,page,known string,observed text,issue,review action; coverage summary.
Next step
Human reads each queued original image and verifies corrected working copy.
Check before continuing
Require literal examples and page references; do not let an assistant invent untested page coverage.
Step documentation 1 ↗Step documentation 2 ↗
Release the searchable reference copy
Final human review
Document reviewer · Approved local working-copy folder and QA log
Save the searchable reference PDF under a distinct name with its test log. Resolve or clearly flag failed/uncertain pages; retain original scans unchanged.
Expected output
Searchable working copy with an honest page-coverage and exception record.
Next step
Staff use it for search while consulting original images for substantive accuracy.
Check before continuing
Numbers, names and dates require visual verification before reliance; no claim that OCR is an authoritative transcript.
Step documentation 1 ↗Step documentation 2 ↗

What you’ll get

searchability-review.csv: file_id, source_version, page, test_string_id, finding, source_locator, visual_check_needed

Inputs and a sample test Synthetic example · not a measured result

Input files

files.csv: file_id, source_version, page_count; known-strings.csv: file_id, page, expected_public_or_synthetic_string
extraction.csv: file_id, page, extracted_text, extraction_status; rendered synthetic/public PDFs for visual comparison

Synthetic example

P1: image-only page with empty extraction
P2: known word extracted with spaces between letters
P3: expected number differs from OCR

Expected output

P1: no searchable text
P2: phrase-search failure
P3: number mismatch for visual review

Test a failure

Staff compare suspect text with rendered originals, especially dates/numbers. Test mixed digital/scanned pages, rotated scans and deliberately degraded text. Never overwrite evidence originals.

Try with the sample files

Follow the tool steps above and use the inputs named in each step. Where provided, native-* files are blank scaffolds and NATIVE-INPUT-MAPPING explains their fields. Original examples and the optional full review prompt use a separate prepared-input schema. Keep them separate unless the mapping explicitly connects them. Downloading a kit does not connect accounts or prove the workflow works.

Download inputs and checklist 10 files

Instructions, blank templates and separate synthetic examples; these files do not install an automation. Blank templates still need verified source data.

Download starter kit All inputs, synthetic examples and review notes in one ZIP. No automation is installed.
Start here: get the sources and prepare the review Documentation-checked acquisition, explicit manual mapping, exact prompts and human review limits
files.csv Blank input template with exact recipe headers.
files.example.csv Separate synthetic fixture input. Do not mix these rows into real exports.
known-strings.csv Blank input template with exact recipe headers.
known-strings.example.csv Separate synthetic fixture input. Do not mix these rows into real exports.
extraction.csv Blank input template with exact recipe headers.
extraction.example.csv Separate synthetic fixture input. Do not mix these rows into real exports.
Expected fixture findings Read-only expected output and failure conditions for the included synthetic input files.
Setup and review checklist Recipe-specific prerequisites, file mapping, prompt, setup steps, checklist and limits.

Optional full review prompt for prepared inputs

Prepared-input review prompt with Excel Copilot, optional; or manual review · prepared inputs

For these public or synthetic PDFs, prepare a searchability QA checklist. Identify image-only pages, empty text extraction, spaced-out words and mismatches against the supplied known strings. Cite page numbers. Preserve the original scans and existing digital text. Do not interpret documents, correct substantive wording or treat OCR as an authoritative transcription. Queue uncertain pages for visual review.

INPUTS
files.csv: file_id, source_version, page_count; known-strings.csv: file_id, page, expected_public_or_synthetic_string
extraction.csv: file_id, page, extracted_text, extraction_status; rendered synthetic/public PDFs for visual comparison

OUTPUT
searchability-review.csv: file_id, source_version, page, test_string_id, finding, source_locator, visual_check_needed

Use only these synthetic or firm-approved inputs and preserve source row/page IDs. Keep missing or conflicting evidence unresolved. Return a draft for the named reviewer. Do not send messages, connect accounts, change records or permissions, move money, or make legal decisions. Keep source files unchanged. List missing prerequisites before continuing.

Proposed / not tested · editorial prompt. This prompt uses the original prepared-input schema shown in the sample test. Step prompts may use different tables. Follow NATIVE-INPUT-MAPPING where supplied; do not mix the original examples with native-step inputs without mapping and checking every field.

Read original method Read pain evidence