Compare PDF versions: text differences are only half the check

Use text and visual checks together

A PDF can change without changing its extracted text. A chart, signature image, font, table position or annotation may differ while the words remain identical. Conversely, two visually similar pages may have different underlying text or reading order. A reliable comparison checks both content and appearance, particularly when numbers, contractual terms or document structure matter.

When a comparison is worth doing

Anyone who has been sent “the revised version” by email knows the problem: the changes are rarely marked, and the reviewer is trusted to spot them by eye. A comparison is worth the few minutes when a number, date, name or clause matters, when a document has been through several hands, or when you have to confirm that only the agreed changes were made.

Typical cases include contract and policy revisions, price lists and quotations, financial statements issued as PDF, tender documents with addenda, and a final print file compared with its last approved proof. In each of these a single changed digit can matter more than a rewritten paragraph, and a human reading two long documents side by side is the least reliable way to find it.

Prepare comparable documents

Choose the original and revised files, confirm their dates and check the page counts. If pages have been inserted or moved, page-by-page comparison can flag many later pages because corresponding content is no longer at the same index. Align the documents or inspect the affected range manually. A scanned file needs visual review or OCR before a text comparison can find its words.

Choosing which file is the original

The report describes changes from one document to the other, so the direction matters. Treat the earlier approved version as the baseline and the new one as the revision, and keep that order consistent. Mixing them up turns every addition into a removal and makes a report that is correct but confusing to read.

Before you start, write down the filenames, the dates and the page counts. If the page counts differ, find out why first. A new page inserted early in the document shifts everything after it, and a page-by-page comparison will then flag nearly every later page even though the content is the same. In that case compare the unchanged sections separately, or remove the inserted page from a working copy using the Delete Pages tool and compare again.

Read the difference report carefully

Additions and removals describe extracted selectable text, not semantic meaning. Normalizing spaces can reduce noisy changes but can also hide a spacing distinction important to code or tabular data. Begin with changed pages and then review pages with no extractable text. The JSON report contains document text and can be sensitive; keep it under the same handling rules as the PDFs.

A worked example: a revised quotation

Suppose a supplier sends quote_v1.pdf and quote_v2.pdf and says only the delivery date has changed. Load v1 as the original and v2 as the revision. The report shows page 1 with text differences, where the delivery date reads differently, and also a second difference in the payment terms, from 30 days to 14 days. That unannounced change is the reason for running the comparison.

Open the overlay for page 1 and drag the opacity of the revised layer. The logo and address line up exactly, and the changed lines appear doubled. On page 3 the report says the text matches, but the overlay reveals that a signature image has been replaced by a typed name. Text comparison alone would have missed that. The result is a short, specific list to send back to the supplier rather than a vague request to “please confirm nothing else changed”.

Use the overlay for page appearance

Select a page and adjust the revised-layer opacity. Matching content appears aligned; moved or changed shapes can appear doubled. Different page sizes are aligned at their origins rather than automatically warped into a match. This helps reveal changes but is not an automated visual-equality score. Zoom in for small punctuation, totals and footnotes, and inspect both individual page previews.

Why a clean report can still be wrong

A report that says “text matches” means that the selectable text extracted from the two pages is the same. It says nothing about everything that is not text. Images, charts drawn as vector shapes, form fields, stamps, signatures and annotations may all differ. Pages with no extractable text, such as scans, will appear to match because there is nothing to compare, so check separately which pages have no extractable text.

Reading order is another source of surprise. Two documents with identical words can store them in a different order inside the file, and some PDFs store a table cell by cell while others store it row by row. A tool can flag a change that a human sees as unchanged, and the reverse is possible when text is split in different places. Treat the report as a list of places to look, not as a verdict.

Finish with a human decision

A comparison result is a review aid, not a legal certification, signature check or proof that nothing changed. Record which files you reviewed and any limitations such as scanned pages or unavailable text. For high-stakes documents, follow the organization’s approval process and have a second reviewer confirm important differences. Do not treat a zero-change text report as permission to skip the visual pass.

Keeping a record of the review

A review is only useful later if you can show what was done. Keep both files unchanged, save the report next to them, and note the date, who reviewed it and any limits, for example “pages 4 to 6 are scans and were checked visually”. Name files so the pairing is obvious, such as contract_v3_vs_v4_report.json.

Remember that the report contains the document’s text. If the contract is confidential, so is the report. Store it where the contract is stored, not in a shared folder with broader access.

Questions people ask

Can it compare a scan with a digital PDF? Not by text, because a scan has none. Use the visual overlay, or run OCR on the scan first.

Why are so many pages flagged after one insertion? A page inserted early shifts all later pages, so page-by-page comparison pairs the wrong pages. Align the documents first.

Does the report prove no other changes were made? No. It lists differences in extracted text and supports a visual check. It is a review aid.

Common patterns in the report

After a few comparisons you start to recognize typical patterns. A burst of changes across every page after one insertion suggests misaligned pages rather than real edits. A small number of changed numbers on a few pages suggests a genuine revision. Changes in only whitespace or line breaks suggest the document was re-exported from a different program, which is common when someone “just saved it again”.

Pay the most attention to small, quiet changes in numbers, dates, names and negations (“shall” and “shall not”), because those are the changes that alter meaning while looking minor. Reading the changed lines aloud, or having a second person check them, catches more than reading silently.