# Check PDF-to-Word conversions Canonical page: https://agents.paretoid.com/workflows/check-pdf-to-word-conversions/ Category: Operations & Reporting Practice relevance: cross-practice. A general operational method. Review the stated tool and jurisdiction limits before use. Jurisdiction/provider scope: Find changed text, missing clauses, and broken layout before reusing a converted document. Status: Proposed / not tested. The public discussion supports the problem; the method source supports only the documented features. No outcome from this adapted recipe has been verified. Verification: The linked sources support the complaint and documented method components. We have not run this recipe or verified its results. The example below is synthetic, not a customer result. ## The problem Find changed text, missing clauses, and broken layout before reusing a converted document. Why it matters: PDF-to-Word conversion can drop clauses or disrupt pleading layouts. If those errors reach a client or court, they can change the document’s meaning, create rework, or lead to filing problems. Comparing the converted file with the original helps catch them before use. ## Sources and evidence - pain: [Pain evidence: Converting PDF to Word](https://www.reddit.com/r/paralegal/comments/1wsyapw/converting_pdf_to_word/) Public comments describe broken layout and missing-clause risk. This recipe relies on comments, not the inconsistent snapshots of the original body. September 29, 2026 is an indexed absolute date; the open-page date was not independently confirmed. Recency remains provisional. - method: [Method: Adobe Compare Files](https://helpx.adobe.com/au/acrobat/using/compare-documents.html) Adobe Compare Files documents PDF-version comparison; Word legal blackline creates a separate comparison document. Convert the candidate DOCX back to a comparison PDF, keeping the original untouched. - method: [Method: Word legal blackline](https://support.microsoft.com/en-gb/word/compare-document-differences-using-the-legal-blackline-option) Additional official capability reference. It does not validate the complete proposed recipe. Source checked: October 15, 2026. Pain reports are self-reported and identities are unverified. Method documentation supports components, not a tested end-to-end workflow. ## Suggested approach Start when: A converted DOCX is marked ready 1. compare extracted text and rendered pages against the source 2. Keep row or page references and mark missing or ambiguous evidence. Do not apply the suggested changes. 3. The responsible staff member checks the draft against the source records and approves any separate action. ## What you need - Existing Word desktop and Acrobat Pro Compare files access; the actual source PDF, converted DOCX and an explicit approved intended-change list. - Existing firm-approved Acrobat Studio, AI Assistant add-on, or another plan including Acrobat AI; online services/cloud storage, language and region must be permitted. No purchase, login or connection is part of this verification. The manual route works without AI. - Preserve original evidence unchanged in restricted storage. Use synthetic material first. Prepare a separate, explicitly approved minimized AI input; aliases alone do not anonymize narrative or attachments. Do not upload raw originals or an alias key merely because another tab/file is sanitized. - 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 component path; proposed adaptation, not tested - Acrobat Compare accepts PDFs, not a direct source-PDF versus DOCX pair. Word legal blackline requires two Word versions; do not invent an authoritative source DOCX by re-converting the same PDF. - Native comparison and visual page review do the QA; AI only explains a supplied reviewed difference record and cannot certify conversion fidelity. Access and setup: - Actual account access, source versions/completeness, privacy approval and reviewer sign-off were not observed. ### 1. Freeze the actual conversion pair Tool: Word desktop Location: Open the converted DOCX > File > Save As > choose PDF Preserve source.pdf and converted.docx unchanged. Save a separate converted-render.pdf from the exact candidate DOCX and record filenames/versions. Obtain the intended-change list from its owner; no intended changes means an explicitly empty approved list, not guessed consent. Expected output: Source PDF, exact candidate DOCX and candidate-render PDF plus intended-change list. Next step: Open source.pdf and converted-render.pdf together in Acrobat Pro. Check before continuing: Both must be the same claimed conversion version; a later DOCX edit invalidates the render/report. Execution status: Official public documentation checked 2026-10-15; no account, source-data acquisition, prompt run or end-to-end workflow was tested. Step documentation: https://support.microsoft.com/en-us/accessibility/word/use-a-screen-reader-to-save-a-document-in-word ### 2. Run native comparison and page review Tool: Acrobat Pro Location: All tools > Compare files > select old/new PDF > Compare Set source.pdf as old and converted-render.pdf as new. Choose the document type/page scope intentionally; do not select text-only for the visual/layout pass. Inspect Compare Results with Go to First Change and subsequent changes, then manually view every source/candidate page for missing text, numbers, tables, headers, footers and line numbering. Record page pairs explicitly. Expected output: Native Compare Results document and manually checked findings with both locations when they exist. Next step: Copy only sanitized findings/short excerpts into conversion-findings.txt; include the exact intended-change list and coverage statement. Check before continuing: A dropped clause has a source locator and may legitimately have null converted_locator. No detected difference is not a guarantee; unreviewed pages stay unreviewed. Execution status: Official public documentation checked 2026-10-15; no account, source-data acquisition, prompt run or end-to-end workflow was tested. Step documentation: https://helpx.adobe.com/acrobat/using/compare-documents.html ### 3. Load only the approved AI input Tool: Acrobat PDF Spaces Location: Acrobat homepage > PDF Spaces > Create a PDF Space > upload files or Paste copied text > Add to PDF Space Add conversion-findings.txt, sanitized intended changes and coverage notes. The native report or original PDFs may be added only if each is separately approved for cloud processing; they are not automatically visible. Use clear source/version labels and verify the Files panel. Do not add originals that have not separately been cleared for Adobe processing. Expected output: Visible approved source set in the PDF Space; no account integration is implied. Next step: Use the AI Assistant message field in this same space for the next prompt. Check before continuing: Confirm every named source is actually loaded and readable. PDF Spaces supports up to 100 files, 100 MB and 600 pages per file; unsupported/password-protected/handwritten material blocks affected extraction. Execution status: Official public documentation checked 2026-10-15; no account, source-data acquisition, prompt run or end-to-end workflow was tested. Step documentation: https://helpx.adobe.com/acrobat/desktop/explore-pdf-spaces/create-pdf-spaces.html https://helpx.adobe.com/acrobat/desktop/explore-pdf-spaces/supported-formats-limitations.html https://helpx.adobe.com/acrobat/desktop/explore-pdf-spaces/pdf-spaces-faq.html ### 4. Run the bounded prompt and inspect citations Tool: Acrobat AI Assistant Location: The PDF Space AI Assistant message field Paste the prompt below. Open each source-number citation, check it, and use Copy to transfer the answer into a new local Word document or text file. Correct errors using the originals before reliance. Prompt: ```text Use only the supplied native comparison findings, visual-review observations and approved intended-change list. Return document_alias,source_version,candidate_version,source_locator or null,converted_locator or null,difference_type,exact observed evidence,intended_change_id or null,review_action. Separate explained changes from unexplained text/number/layout differences. Retain all findings and page-coverage gaps. If text is absent in the candidate, keep its candidate locator null; do not fabricate one. Do not repair legal wording, infer missing clauses, claim to inspect Word formatting from text or declare filing readiness. ``` Expected output: Explained/unexplained conversion-QA queue with preserved native evidence and coverage gaps. Next step: Save the checked draft as conversion-QA-review.docx; carry source IDs, versions and reviewer corrections into the next review step. Check before continuing: Reconcile AI rows to every native/manual finding, particularly omissions and changed numbers. Execution status: Official public documentation checked 2026-10-15; no account, source-data acquisition, prompt run or end-to-end workflow was tested. Step documentation: https://helpx.adobe.com/acrobat/desktop/use-acrobat-ai/generative-ai-features/ai-get-answers.html ### 5. Human review and final handoff Tool: Responsible staff reviewer and lawyer/process owner Location: Local reviewed report beside the frozen source set Staff checks all pages and each unexplained difference; lawyer approves substantive changes. Any repair is made in a separate DOCX copy, rendered again and rechecked. Expected output: Reviewed conversion report and version-specific disposition. Next step: Release only the human-approved version; keep source and comparison artifacts unchanged. Check before continuing: All unresolved or unreadable evidence stays open. No inferred approval or external action follows from AI output. Execution status: Official public documentation checked 2026-10-15; no account, source-data acquisition, prompt run or end-to-end workflow was tested. Step documentation: https://helpx.adobe.com/acrobat/using/compare-documents.html ## What you will get Approved assistant workspace for difference explanation; Acrobat/Word desktop for native comparisons and rendering. Export qa-findings.csv with document_alias, source_locator, converted_locator, difference_type, expected_change_id, reviewer_status. ## Human review and limits Staff inspect every unexplained text difference and visually check all pages; attorney signs off on substantive differences. A planted dropped clause and changed number must both be caught. 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. What to measure: During the pilot, count findings confirmed by the reviewer, false alarms, and missing-input cases. Do not infer time or money saved. ## Original prepared-input schema Files: source.pdf, converted.docx, converted-render.pdf; intended-changes.csv columns change_id, source_locator, expected_change. Each file uses the same document_alias. Synthetic example input: ```text D1: 30 days becomes 80 days, no approved change D2: a source clause is absent in the converted file D3: a heading change matches the approved change log ``` Expected synthetic output: ```text D1: unexplained number change D2: missing-clause finding with source page and nearest converted location if known; otherwise absent / locator unavailable D3: expected change, keep log reference ``` Failure check: Staff inspect every unexplained text difference and visually check all pages; attorney signs off on substantive differences. A planted dropped clause and changed number must both be caught. ## Starter files Native blank scaffolds and original synthetic examples may use different schemas. Follow NATIVE-INPUT-MAPPING where supplied. These files do not connect accounts or install automation. - [Download starter kit](https://agents.paretoid.com/assets/starter-kits/a01-starter-kit.zip) — All inputs, synthetic examples and review notes in one ZIP. No automation is installed. - [Start here: get the sources and prepare the review](https://agents.paretoid.com/assets/starter-kits/a01/START-HERE.txt) — Documentation-checked acquisition, explicit manual mapping, exact prompts and human review limits - [Setup and review checklist](https://agents.paretoid.com/assets/starter-kits/a01/README.txt) — Required inputs, file use, review steps and limits - [Intended changes blank CSV](https://agents.paretoid.com/assets/starter-kits/a01/intended-changes-template.csv) — Blank input headers for intended-changes.csv - [Intended changes synthetic example](https://agents.paretoid.com/assets/starter-kits/a01/intended-changes-example.csv) — Synthetic test records for intended-changes.csv - [Source fixture](https://agents.paretoid.com/assets/starter-kits/a01/source-fixture.txt) — Required policy or native-input preparation aid; synthetic when marked example - [Converted fixture](https://agents.paretoid.com/assets/starter-kits/a01/converted-fixture.txt) — Required policy or native-input preparation aid; synthetic when marked example - [Synthetic example expected results](https://agents.paretoid.com/assets/starter-kits/a01/expected-results.txt) — Compare a pilot result with the planted exceptions ## Optional original prepared-input prompt Destination: Your firm-approved AI assistant · prepared files This original prompt uses its own prepared-input schema. Do not mix it with native-step tables without the supplied mapping. ```text Using only these synthetic or approved deidentified source and converted documents, prepare a conversion QA report. Identify text additions, deletions, changed numbers and broken headings/line numbering. Cite source and converted locations where available. Missing text may have no counterpart locator; say absent / locator unavailable rather than invent one. Separate intended changes listed in my change log from unexplained differences. Do not repair legal wording or declare the document filing-ready. Preserve both originals. INPUTS Files: source.pdf, converted.docx, converted-render.pdf; intended-changes.csv columns change_id, source_locator, expected_change. Each file uses the same document_alias. OUTPUT Approved assistant workspace for difference explanation; Acrobat/Word desktop for native comparisons and rendering. Export qa-findings.csv with document_alias, source_locator, converted_locator, difference_type, expected_change_id, reviewer_status. Use only the supplied synthetic or firm-approved inputs. Cite source row or page IDs for each finding. Keep unknowns unresolved. Return a draft for the responsible staff reviewer. Do not send messages, connect accounts, change records, move money, or make legal decisions. Keep source files unchanged. If an input or permission is missing, list what is needed and stop that check. ``` Proposed / not tested · editorial prompt Published by Law Firm Agents. Proposed adaptation; not tested. Not legal advice.