Everyone who has scanned more than ten pages has produced a crooked one. A sheet feeds slightly askew, a book won’t lie flat, or a stapled corner lifts the paper. The resulting page sits in your PDF at a small, irritating tilt. It looks like a cosmetic problem. It’s actually a functional one, because skew is one of the most common reasons OCR quality drops and scanned documents stay hard to use. This guide covers what skew does, how to prevent it, and how to fix the crooked scans you already have.
Why a couple of degrees matters
Text recognition works by finding lines of text and then decoding character shapes along them. Both steps assume, roughly, that text runs horizontally, and a tilted page undermines that assumption in several ways:
- Line detection degrades. On a tilted page, the end of one line sits lower than the start of the next, and naive line-finding starts merging or splitting lines. Words get read out of order or attached to the wrong line.
- Character shapes distort. Rotated glyphs match their templates less cleanly; an already-marginal scan (fax quality, light toner) loses accuracy it couldn’t afford to lose.
- Text-layer alignment suffers. Even when recognition succeeds, a text layer laid over a tilted image can sit visibly offset from the printed words, so search highlights land beside the text they matched.
Modern engines tolerate mild skew, but tolerance isn’t immunity. Accuracy on a straightened page is consistently better than on a tilted one, and on poor-quality scans the difference is dramatic. Humans mind, too. A crooked page on screen is genuinely harder to read, and it looks careless attached to anything professional.
Prevention is cheaper than repair
If you control the scanner, a few habits eliminate most skew before it exists:
- Square the stack. Tap the pages flush before loading a document feeder, and snug the side guides against the paper. Loose guides are the #1 cause of feed skew.
- Remove staples and paperclips completely. A lifted corner rotates the whole sheet as it feeds.
- Use the flatbed for bound material. Books and thick packets skew badly in feeders; on the glass, align the spine against the edge rule.
- Turn on the scanner’s own deskew, if it has one. Many office machines can straighten during capture, which is the best place to fix it because the correction happens before the image is saved.
- Rescan the worst offenders. If you still have the paper, a ten-second rescan beats any software fix.
Fixing scans you already have
Usually you don’t have the paper. You have a PDF someone else scanned, crooked, years ago. Here are your options.
Manual rotation in Preview is the wrong tool
Preview rotates pages in 90° steps, which is useful for upside-down pages and useless for a 2° tilt. There’s no built-in fine-angle deskew in Preview, so the “free and built-in” route runs out here.
Image editors are possible but absurd at scale
You can export a page as an image, straighten it by eye in an editor, and rebuild the PDF. For one precious page, such as a certificate or an exhibit for a filing, that’s workable. For a 200-page file it’s not a real plan.
OCR suites with cleanup passes
ABBYY FineReader and similar tools include image-preparation steps (deskew among them) you can configure per job. If you already own one for its conversion features, as discussed in the roundup, it’s a solid choice. The usual costs come along, though, in the form of a subscription, an app-centric workflow, and settings to learn.
Deskew as part of OCR is the set-and-forget way
The practical insight is that for most people, deskew doesn’t need to be a separate step or a decision at all. RightClickOCR straightens tilted pages automatically during recognition. It’s not a mode you enable, just part of how every page is processed. Right-click a crooked scan (or a whole folder of them), choose Make Searchable (OCR), and the “Name (Searchable).pdf” that appears beside your original has been recognized with skew handled. That is precisely where skew correction pays off, since the point of the exercise is an accurate, well-aligned text layer for ⌘F and Spotlight.
The same rules apply here as everywhere in this product. Processing is 100% on-device via Apple’s Vision framework, and your original file is never modified, crooked charm and all. If a page is beyond saving (deep shadow from a book gutter, half the text cut off), no deskew pass will invent the missing information; rescanning remains the honest answer for truly damaged captures.
A realistic workflow for a messy archive
- Don’t pre-sort by crookedness. The tool will deal with it. Right-click the folder, run the batch.
- Spot-check a few searchable copies. ⌘F a phrase you can see and confirm highlights land on the words.
- For the handful of pages where recognition clearly struggled, check the source scan and rescan those originals if you have them. Heavy skew plus low contrast is usually the culprit.
- Going forward, fix skew at the scanner with the habits above, and OCR new scans as they arrive.
Crooked scans are entropy; they happen to every archive. The fix doesn’t need to be a project. It can be a property of the way your files become searchable.