You have a list of PDF links. Annual reports, product manuals, tender documents, research papers, policy PDFs on a government site. You need the text inside them in a spreadsheet, ideally one row per page so you can see where something was said, and you would rather not install Python, wrestle with a PDF library, or download fifty files by hand.
This post shows the route I use: paste the URLs into a hosted tool, press Start, export a CSV. No code. I also cover the limits, because a PDF text tool that quietly returns empty rows for scanned files is worse than no tool at all.
Disclosure: I built and maintain the tool used here, PDF Text Extractor - Extract Text from PDF URLs on the Apify Store. It is paid per page, and the first 25 pages of every run are free. Everything below works inside that free slice, and I list the alternatives at the end. This article was written with AI assistance and checked against a real run.
What you get
One row per PDF page, with these columns:
| Column | Meaning |
|---|---|
sourceUrl |
The PDF the row came from |
pageNumber |
Page number inside that PDF |
totalPages |
How many pages the PDF has in total (even if you only parsed the first few) |
text |
The selectable text of that page |
textTruncated |
true if the page was longer than your character cap and got cut |
hasText |
false if the page had no text layer (typically a scanned image) |
fetchedAt |
When the page was read |
Page-level rows matter more than they look. If you are checking where a supplier contract mentions "termination", or which page of a 120-page report has the revenue table, "page 47" is the answer you want, not one giant blob of text.
Real output from a real run
Here is the output of an actual run (2026-11-09, run ID ywJjLIRlzLGA4VeGs) on a small public test PDF from the Mozilla PDF.js repository. The input was just:
{
"urls": [
"https://raw.githubusercontent.com/mozilla/pdf.js/master/test/pdfs/basicapi.pdf"
],
"maxResults": 25
}
The run took about 6 seconds and produced three rows, one per page. Exported as CSV (four of the columns selected), it looks like this:
"pageNumber","totalPages","hasText","text"
"1","3","true","Table Of Content
Chapter 1 .......................................................... 2
Paragraph 1.1 ...................................................... 3
page 1 / 3"
"2","3","true","Chapter 1
page 2 / 3"
"3","3","true","Paragraph 1.1
Powered by TCPDF (www.tcpdf.org)
page 3 / 3"
And one row as JSON:
{
"sourceUrl": "https://raw.githubusercontent.com/mozilla/pdf.js/master/test/pdfs/basicapi.pdf",
"pageNumber": 1,
"totalPages": 3,
"text": "Table Of Content \nChapter 1 .......................................................... 2 \nParagraph 1.1 ...................................................... 3 \npage 1 / 3",
"textTruncated": false,
"hasText": true,
"fetchedAt": "2026-11-09T22:44:55.773Z"
}
It is a deliberately tiny document. The point is the shape: line breaks are kept, the dotted table-of-contents leaders come through as they appear in the PDF, and nothing is reflowed or "summarised". You get the text that is in the file.
Step by step (about five minutes)
- Open the tool's Store page and click Try for free. You need a free Apify account; no card is needed to try it.
- In the Input tab, replace the example URL in "Public PDF URLs" with your own links, one per line. Up to 50 URLs per run. They must be public HTTP(S) links that open the PDF directly (no login pages, no "click here to download" pages).
- Set "Maximum pages per PDF" (default 20, up to 100). The tool parses the first N pages of each file.
totalPagesstill tells you the full length, so you can see what you skipped. - Set "Maximum pages" (the total across all PDFs, default 100). If you want to stay inside the free slice, set it to 25.
- Optional: type a word into "Keyword". Only pages that contain it (case-insensitive, plain text, no regex) are saved. Pages that do not match are skipped and not counted.
- Click Start. When the run finishes, open the Output tab, then Export. Apify datasets export to JSON, CSV, Excel, XML, HTML table, RSS or JSONL.
That is the whole workflow. For a quick one-off job, stop here.
Find the pages that mention something
The keyword filter turns the tool into a "which page says X" finder across a pile of PDFs. For example, to find every page in a set of supplier terms that mentions termination:
{
"urls": [
"https://example.com/terms/supplier-a.pdf",
"https://example.com/terms/supplier-b.pdf"
],
"keyword": "termination",
"maxPages": 100,
"maxResults": 200
}
The URLs above are placeholders; use your own. The output is one row per matching page, with sourceUrl and pageNumber, which you can paste straight into a review sheet.
Re-run it on a schedule (for PDFs that get replaced)
Some PDFs live at a fixed URL and get replaced: a monthly price list, a weekly report, a published policy. To re-read them automatically:
- After a successful run, click "Save as a new task" (top right of the Actor page in Console) so the URLs and settings are kept as a reusable task.
- In Apify Console, open Schedules, create a schedule, and add the task. Schedules use cron expressions; for example
0 8 * * 1runs every Monday at 08:00 in the timezone you pick. - New schedules start disabled, so remember to enable it.
Be honest with yourself about what this gives you: a fresh copy of the text on each run, page by page. It does not compute a diff for you. If you need "what changed", export both runs to a spreadsheet and compare, or connect the run to whatever tool you already use for comparisons.
Also note that unnamed run datasets on Apify expire after 7 days. If you want to keep results, export them, or name the dataset.
Limits (read these before you rely on it)
-
No OCR. Scanned PDFs, or pages that are just a photo of text, come back with
hasText: falseand empty text. If your documents are scans, you need an OCR tool instead; several exist on the Store and elsewhere. - No layout reconstruction. Tables come out as lines of text, not as cells. Multi-column layouts can interleave. If you need structured tables, use a table-extraction tool.
- Size and time caps. Downloads are limited to 5 MB by default (you can raise it to 10 MB), and each PDF gets a 15-second parse deadline. Very large files will fail with a recorded error rather than hang.
- Public files only. It respects robots.txt, refuses private network addresses, and does not log in, solve CAPTCHAs or decrypt password-protected PDFs. You are responsible for having the right to process the documents.
-
Character cap per page. Default 20,000 characters, up to 50,000. If a page is longer, the text is cut and
textTruncatedis set totrue, so you can spot it.
What it costs
- The first 25 pages of every run are free.
- After that it is $2 per 1,000 pages ($0.002 per page), charged only for pages actually saved. Failed downloads are not charged, and with a keyword set, skipped pages are not charged.
- A worked example: one run over 200 pages saves 200 rows, of which 175 are charged, so $0.35.
Prices can change; the Pricing tab on the Store page is the authoritative source.
Alternatives, honestly
- A single PDF, once: open it in a browser or a desktop reader, select all, copy. Free and instant. You do not need a tool.
-
You already write code:
pdftotext(Poppler) or Python libraries like pypdf do the same text extraction locally for free. A hosted tool saves you setup, scheduling and hosting, not magic. - Scanned documents: use an OCR service; this tool will not help.
- Many PDFs, repeatedly, no code, results in a spreadsheet: this is the case the tool is built for.
Try it on your own PDFs
There is a ready-made example you can open and run as is: Extract text from public PDF URLs page by page. Swap in your links, keep "Maximum pages" at 25 for a free first test, and check the hasText column first. If most rows say false, your PDFs are scans and you need OCR, not this.
If something breaks on a public PDF that opens fine in your browser, open an issue on the Store page with the URL. I read them.
Top comments (0)