Where the text comes from, and what it cannot do
A PDF can hold a page in two very different ways. A PDF made from a document, a web page or a spreadsheet contains real text: every piece has its characters and its position. That text layer is what you copy when you select words in a viewer, and it is what this tool reads. A PDF made by scanning paper holds a picture of each page, with no characters in the file to copy.
ClipLime does not read text from pictures, so a scanned PDF gives no text, and the tool says so plainly: "This PDF looks like scanned pictures, so there is no text to extract. This tool does not read text from pictures (no OCR)." If only some pages are pictures, such as a report with a scanned appendix, the other pages are read and the pages without text are named. You can still save the scanned pages as pictures with PDF to Image and use a program that recognises text.
Two other cases are worth knowing. Some scanned PDFs carry an invisible text layer added by a scanner or an OCR program, and that text is read, mistakes included. In some PDFs the fonts do not say which letters they draw, so the text layer copies as squares and symbols. The tool notices such pages and says they do not read as text.
Reading order, columns and tables
A PDF does not store paragraphs. It stores pieces of text, each with a position, in the order the program drew them, which is not always the order you read them in. The tool rebuilds reading order from the positions: pieces on the same baseline become a line, lines are stacked from top to bottom, and a strip of empty space that runs down the page marks a column. Columns are read one after the other, so a two column article comes out as the left column, then the right, and a title across both stays above them.
A table is not split into columns. When short cells line up in rows, each row is read across and the cells are separated by tabs, so the result pastes into a spreadsheet. A table with long cells that wrap onto several lines can be read in an odd order, and a row where the gap between cells is narrow may lose its tab.
There are limits. Overlapping text boxes, text wrapped around pictures and unusual designs can come out in a different order from the one you would read. Text turned another way, such as a vertical label in a margin, is added after the main text. Headers, footers and page numbers are text on the page, so they appear with each page: a running head above the columns comes first and a footer or page number under them comes last. Where two columns almost touch, pdf.js can report a line of each column as one piece of text, and those two lines then come out together. Look at one page before you rely on a long document.
Line breaks, paragraphs and hyphens
As in the PDF gives one line of text for every line of the page, with a blank line where reading jumps to another column or block. It is the safest choice for tables, addresses, code and poems.
Paragraphs joins the lines of a paragraph into one line and puts a blank line between paragraphs. A line continues into the next when it runs up to the right margin and the first word of the next line would not have fitted after it. A gap between lines, a change of type size, a bullet or number, or an indented first line starts a new paragraph, and short lines such as a list or an address stay lines. A paragraph that runs from the bottom of one column to the top of the next is joined too, when the column ends flush with its margin and the next one does not start a new sentence. It is a careful guess, so check the result if the layout matters. A paragraph that runs onto the next page is not joined across the page break.
A word split at the end of a line is joined again only when the evidence is clear: the hyphen is a soft hyphen, the invisible kind typesetters insert, or the word appears whole elsewhere in the text, or in another form such as a plural. Otherwise the hyphen stays, so a real compound such as long-term is never damaged, and a split word may keep its hyphen. Page markers put a line like --- Page 3 --- before each page.
Other scripts, form fields and word counts
Right to left text, such as Arabic and Hebrew, comes out in logical order, the order you would type it, with numbers and Latin words kept intact inside it, and brackets the right way round. Chinese and Japanese text is joined without spaces, as it is written, and Korean keeps its spaces. Arabic presentation forms and ligatures such as fi are turned back into ordinary letters, so the text can be searched.
With Form fields and comments on, text typed into form fields (with the name of the field) and the text of comments and notes is added at the end of each page. Switch it off for the page text alone.
Words on each page lists a count for every page and marks pages without text. Words are runs of text separated by spaces, and Chinese, Japanese and Korean characters are counted one by one, so those counts are rough. To count or tidy the result further, use Word Counter or Text Cleaner.