How to compare two PDFs — find every change, even when the layout shifted
Comparing PDFs sounds straightforward — show me what changed. The reality is messier: did a paragraph move, or did it get rewritten? Did a table get reformatted, or did its data change? Did the date update, or did the whole calendar quarter shift? Good PDF diffing distinguishes structural changes from content changes from formatting changes — and lets you act on each separately.
How PDF diffing works
We extract the text from both PDFs along with structural information (paragraphs, tables, headings, page boundaries). A two-pass diff first aligns common content — paragraphs that are identical or nearly so — and then identifies what changed in the gaps. The output is a redline document showing additions, deletions, and substitutions, plus a structured summary of changes by section.
AI-powered comparison adds a layer on top: classify each change by severity (typo vs. material change vs. clause restructuring), and explain in plain English what's different. That's the difference between 'paragraph 3 shows N edits' and 'paragraph 3's notice period changed from 30 days to 60 days, increasing the buyer's exposure'.
When PDF compare is the right tool
Specific cases:
- Contract redlines. Counterparty sent back V2 — what did they actually change? Compare V1 to V2 and read the diff.
- Policy revisions. Internal handbook updated; what's new from last year's version?
- Specification changes. Engineering spec revised; flag what's actually different.
- Regulatory filings. Annual filings with last year's text amended — the difference is what's new this year.
- Translation review. English vs. translated version — does each paragraph correspond properly?
Common mistakes
Patterns that produce diff results that aren't useful:
- Comparing PDFs with different formatting. Reformatting (font change, page size change, column adjustment) produces hundreds of false-positive 'changes' that aren't really content changes. Run normalize-formatting first.
- Comparing without OCR on scans. If one or both inputs is a scan with no text layer, the diff has nothing to compare. OCR both first.
- Ignoring the severity classification. AI compare classifies each change as cosmetic, material, or critical. Read the critical ones first; skip the cosmetic ones.
- Comparing the wrong versions. If you compare V3 to V5, you miss the V4 → V5 changes mixed with the V3 → V4 changes. Compare consecutive versions for clarity.
- Trusting the diff for legal certainty. Diffs are aids, not authority. For high-stakes contracts, have counsel review the diff and the full document.
Tips for the best diff
Habits that produce more useful comparisons:
- Normalize first. Run both inputs through Normalize to standardize fonts, page sizes, and spacing. The diff is sharper.
- Use the severity filter. Filter to 'material changes only' for a 30-second scan; turn it off for a full review.
- Export the diff as a redline DOCX. Easier to share with non-technical reviewers than the PDF redline.
- Combine with chat. After running compare, ask 'summarize the most material changes' to get a paragraph you can paste into your status update.
- Save the comparison. If you'll need this comparison again, save the diff alongside the source documents in your DMS.
Limits and pricing
Free preview: simple text diff for up to 10 pages. AI compare with severity classification: 15 credits per diff up to 100 pages. Output: redline PDF, redline DOCX, structured JSON of changes, and a plain-English summary. Compatible with every reader; the redline PDF works in Acrobat, Preview, and every browser.