Scanned résumés, photographs and PDFs with no text

A scanned page carries no text layer, so a parser receives nothing at all. Here is how to tell, what recognition can recover, and how to rebuild a real file.

ConxfolioPublished 2026-09-12

— a picture of a page is not a page.

A PDF can contain two very different things: text with positions, or a photograph of a page. They look identical on screen and they behave completely differently the moment software tries to read them.

How to tell which you have

Open the file and try to select a word. If the cursor highlights the word, there is a text layer. If it draws a box around the whole page, the file is an image, and a parser reading it receives nothing at all.

Scans, photographs, files produced by a copier and documents exported by some design tools all end up this way. So do pages printed to PDF from a screenshot, which is a surprisingly common accident.

What it costs you

Everything. An empty extraction produces an empty document: no name, no dates, no sections. The low-score guide covers what a report looks like when almost nothing came back, and this is the extreme case of it.

What recognition can recover

When you check a scan or a photograph with the ATS résumé checker, a page with no text layer is recognised in this browser rather than on a server, so the image bytes do not need to leave the device. The resulting text goes through exactly the same parser as a text PDF, and the report says how the file was read: pages read from the text layer, pages recognised from a scan, or a mixture of both.

That distinction is on the page for a reason. "Recognised from a scan" explains an odd character in a company name; "read from the text" does not.

Recognised text is a starting point, not a document

Recognition misreads names, numbers, unusual spellings and anything set in a decorative face. Use it to recover your content, then check every field with your eye. The import review in the résumé builder lists what was read field by field and leaves anything it could not support empty, which makes the checking straightforward.

Getting back to a real file

Import the scan, correct the fields, choose a design, and export a fresh PDF with a real text layer. That takes about ten minutes and it ends the problem permanently, including for every future application. The export is free in PDF and DOCX, and the format guide covers which to send.

If the scan is all you have

Sometimes an old résumé exists only on paper. Photograph it in even light, flat, with the whole page in frame, and let the recognition do the first pass. The typing you save is worth the photograph being slightly crooked. What you should not do is send the photograph itself, however good it looks.

How to tell what happened after a check

The report distinguishes pages read from a text layer, pages recognised from a scan, and a document that mixed the two. That line explains anything odd underneath it: an unusual spelling of an employer, a date that came back wrong, a heading that was not recognised. It is the difference between a document with a formatting problem and a document whose text is an approximation.

Where a mixed result appears, the usual cause is a file assembled from several sources, such as a text-layer page with a signed or scanned page appended.

Photographs of a page, and when they are enough

A photograph is enough to recover content and never enough to send. The reasons are the same as for a scan, with the addition that a photograph carries shadows, skew and cropping, which reduce recognition accuracy further. Where it is what you have, take it flat in even light with the whole page in frame, check every recovered field by eye, and treat the result as a draft to rebuild from rather than as a document. The low-score guide covers what a report looks like when very little came back, which is what an unrecovered image produces.

Why the recognition runs in your browser

Because a résumé is personal, and a scan of one is a photograph of your address, your history and often your signature. Reading it locally means the image bytes do not need to be uploaded to be understood, and the text that is sent onwards is the text you can see on the review page. The format guide sets out which requests do send something and what each one carries.