How a PDF table becomes a workbook
How a digital PDF is recognized
The file is checked with pdf-parse before any paid step. If the extracted text is at least 40 characters and averages at least 15 characters per page, the PDF is treated as native. That is the path that builds a workbook without a paid plan. It is the text you could already select in a viewer, gathered into rows. A cover with a short title and a stack of photographed tables can fall under those numbers, and the job is then treated as a scan even if a few words are selectable. A normal export from a spreadsheet or a report with a real text layer is the case this path is built for. The button on the page says Convert with AI for every file, including a digital PDF that never calls a vision model. The useful expectation is simple: enough real text, a workbook you can open, and a minute to compare it with the original.
What the native workbook contains
On the native path the table reader looks for tables and writes them with ExcelJS. When merged tables are present, those rows become the sheets. Otherwise each page of tables becomes its own sheet, named Sheet1, Sheet2, and so on. The download is an .xlsx file, the spreadsheet format Excel and Sheets both open. If the reader finds no tables, the text is split into cells on tabs or on two or more spaces, one row per non-empty line. That fallback helps a simple column layout. It is not a promise about a designed invoice, a footer total, or a note sitting beside the grid. Open the workbook next to the PDF and fix the cells that matter. If the table reader throws, the job stops with a conversion failure. It does not quietly switch to the vision model. A failure on a file you know is digital is a reason to export a cleaner PDF, not a reason to assume a paid plan will be used automatically.
What a paid plan adds for a scan
A scanned or image-like PDF skips the table reader. The account needs an active paid plan. Monthly is shown as Pro, six-month as Pro+, and yearly as Ultra, and any of those active plans counts as premium. Without one, the response is premium required, HTTP 402, and the message says AI OCR for scanned documents requires an Ultra/Pro plan. With a paid plan, each page is rendered as a PNG at a scale of 2.5 and sent to OpenAI. The model is gpt-4o-mini unless the server is configured with a different vision model. The prompt asks for JSON only: sheets of rows, every row the same length, empty cells padded with blank strings, multi-page tables gathered into logical sheets, and cell values as plain strings. Temperature is zero so the grid stays literal. Those strings are written into the .xlsx workbook. The prompt does not create Excel merged-cell objects, so a visual merge on the page may come back as repeated text or as a blank neighbor. Proof the sheet. A faint scan, a stamp over a number, or a sideways page gives the reader less to work with. There is no published page cap. A very long scan can still run out of time, because the route is allowed up to 60 seconds.
Passwords, names, and how long the upload stays
A password-protected PDF should be unlocked first, with the password you already know, on Unlock PDF. The unlocked file is the one to convert. The PDF is uploaded so the tables can be read. The privacy policy says uploads and outputs are deleted automatically, typically within one hour and no later than 24 hours. This route publishes no file-size constant, so there is no megabyte figure to promise here. The limit that does apply to a free account is the daily gate: eight runs, unless the account is unlimited. That gate is checked before the call, for a digital table and for a scan alike. A paid plan skips the gate. Free downloads add -convertify4u before the extension. A paid plan keeps the plain stem taken from the original file name, with .xlsx on the end. The workbook bytes are the conversion. The extra word in the name is only how a free download is labeled in the browser.
Excel, plain text, or Word
Choose this page when the next step is a spreadsheet: a filter, a sum, a chart, or a paste into a model you already keep. Choose PDF to Text when you want the words in a .txt file and you do not need columns. A digital PDF can take that text path without a paid plan, and a scan there uses a different reader. Choose PDF to Word when you want a document you can edit, including a scan that the Word path can read on its own server. Choose Excel to PDF when the spreadsheet already exists and someone needs a file they can open without Excel. A photo of a printed grid is not what this chooser accepts. The control accepts PDF files. If all you have is a picture, Image to Text returns the words as plain text, and a document scanner can straighten the page before you save a PDF. Keep the original PDF until you have checked the numbers you will send on. The converter gives you a workbook. The last look is still yours, and it is quick on a clean export and worth more time on a scan.