Find one clause in a 300-page PDF using page markers

The tender document is 316 pages. Somewhere in it is the clause about earnest money deposit refund timelines, and your manager wants that clause plus the two pages around it, as a separate PDF, before the 4 pm call.

Your viewer's search box finds "earnest" 41 times. You click Next, read, click Next, read. Twenty minutes gone and you are on hit 19.

There is a faster route, and it turns on one small detail in how PDF Text Extract writes its output.

The marker

The extracted text is not one continuous wall. Each page is written as a labelled block:

--- Page 147 ---
14.3 The Earnest Money Deposit shall be refunded to unsuccessful
bidders within thirty (30) days of the award of contract...

--- Page 148 ---
14.4 No interest shall be payable on the Earnest Money Deposit...

That --- Page N --- line is the whole trick. It means a plain text file, searched in any editor on any machine, hands back real page numbers from the original PDF. Once you have a page number, the rest of the toolkit can act on it.

The walkthrough

1. Confirm the PDF has text in it at all.

Drop the file into File Info first. Two fields decide whether this will work: Contains text fonts and Encrypted. If the first says no, the file is page images and there is nothing to extract — you are in OCR territory. If the second says yes, the file is locked and has to be opened before anything can read it.

Ten seconds, and it saves you running a process that was never going to produce output.

2. Extract.

Load the PDF into PDF Text Extract and press Extract Text. The progress line counts through the document — Extracting page 147 of 316 — so on a large file you can see it moving rather than wondering whether the tab has hung. Parsing happens off the main thread, so the page stays responsive while it works. Nothing is uploaded; the file is read inside your browser, which matters when the document is a commercial tender you are not supposed to be circulating.

3. Read the result panel.

It reports something like 316 pages · 984,213 characters. The character count is a useful sanity check. Under a few thousand characters for a 300-page document means most pages had no text layer, and you have learned that before wasting time on the file.

4. Download the .txt and search it properly.

The download takes the original filename with a .txt extension. Open it in whatever you have — Notepad, Notepad++, VS Code, even a phone text editor.

This is where you gain the twenty minutes. A real editor's "find all" lists every hit at once with its line number, in one pane, instead of walking you through them one at a time. You can scan 41 hits in about fifteen seconds and see immediately that 38 of them are the phrase in a definitions list and three are in an actual clause.

5. Scroll up to the nearest marker.

Above your hit, the closest --- Page N --- line is the page it sits on. That is your answer: page 147.

6. Cut out the pages you need.

Now you have a page number, Extract PDF Pages will pull pages 146 to 148 into a small standalone PDF you can attach to an email. If instead you need the document broken into sections around several such points, Split PDF is the better fit.

Total elapsed: three or four minutes, most of it the extraction.

What the character count is telling you

Divide it by the page count and you have a number you can act on. That 984,213 characters across 316 pages is 3,114 a page, which is what a dense, fully digital tender page looks like. A file reporting 61,000 characters over the same 316 pages averages 193 — not a document with a text layer, but a stack of scans with a running header and a page number on each sheet.

The count in the header covers every page in the PDF, not only the pages that produced text, which is what makes the ratio worth trusting: one number comes from the document, the other from what was recovered.

The gap in the numbering is information

Here is the behaviour that catches people, and it is worth understanding rather than working around.

A page that produces no text is not written to the output at all. No empty block, no marker. So a 316-page PDF might give you markers running --- Page 44 ---, --- Page 45 ---, --- Page 48 --- with 46 and 47 simply absent.

Those two pages are not corrupted. They are pages with nothing selectable on them — almost always a scanned annexure, a signed and stamped page photographed and dropped into an otherwise digital document, or a full-page diagram.

That gap is diagnostic, and it is more useful than it looks. It tells you exactly which pages of a long mixed document are images, without opening a single one. In a tender pack, those are usually the pages that matter most: the signed declarations, the stamped certificates, the annexures somebody scanned at the last minute.

Once you know the numbers, pull those specific pages out with the page extractor and run that short file through Advanced OCR, which is built for exactly this — recognising text in page images, in your browser, without the document going anywhere. You get the readable text from the awkward pages without putting the other 314 through a recognition pass that they never needed.

Why not just search inside the viewer

Viewer search is fine for one lookup. It stops being fine when:

• You need every occurrence at once, not one at a time, to judge which is the real clause.
• You are on a phone. Most mobile PDF viewers search badly or not at all on a file this size.
• You want to send someone the text. A paragraph pasted into a message beats a screenshot of a paragraph, and beats "see page 147" by a wide margin.
• You need a pattern rather than a word. A proper editor will find every date in dd/mm/yyyy form, or every amount over six digits, with a regular expression. No viewer does that.
• You want to compare two versions. Two .txt files diff cleanly. Two PDFs do not.

Practical notes

The extractor takes files up to 1 GB, which is more than any tender document will ever be, and the limit exists because the parsing happens on your own machine rather than a server. A very large file simply takes longer; leave the tab open rather than reloading partway through.

Text comes out with the lines rebuilt from character positions, which means a two-column page arrives with the columns interleaved. For finding a clause that does not matter — the words are all there and the marker still tells you the page. It matters when you intend to read the extracted text as prose.

You can also copy the whole extraction to the clipboard without downloading anything, which is the quicker path when you only wanted one paragraph out of the file.

Why the columns interleave

A PDF does not store lines. It stores fragments of text pinned to coordinates, and the extractor works out which belong together. It does that with a tolerance: anything within eight points of the same vertical position is treated as one line, then sorted left to right. Eight points is about half the height of an 11-point line — tight enough to keep one line off the next, and nowhere near enough to tell one column from another. A line in column one and its neighbour in column two sit at the same height, so they are joined.

The same rule flattens tables. A row comes back as its cells strung together with single spaces, because runs of spacing collapse to one space. Every figure is there and none of the alignment is, so do not read an amount out of an extracted table by counting along the line.

The page-level tools that take over once you know your page numbers are collected on the PDF tools page.