AI Data Extraction

Not by itself. Only characters stored as text in the PDF are read, so a scan or a photo with no text layer ends in an error that suggests OCR PDF. Make the file searchable with OCR PDF, then run the extraction on that output.

The ZIP holds five files. model-output.json is the model's JSON reply as it was returned. extraction.json holds the same data. With a field schema, it lists every field you asked for, in your order and under your names, and a key the model respelled is mapped back to your name. A requested field the model did not return at all is null in extraction.json too, and manifest.json lists it under extraction.missingKeys. With a prompt, it is the reply as given. extraction.md lays the data out as a Field and Value table, or as a numbered list if the reply is a list. In extraction.md a null reads "not found, or several different values". It marks a requested field the model never returned as not returned by the model, and lists any keys you did not ask for in a separate table. manifest.json records the provider, the model, your prompt or schema, the page and character counts, any cut, and which fields were missing, respelled or extra. README.txt describes the package. There is no CSV or spreadsheet in it. For data laid out in rows and columns, Extract Tables writes one CSV per table plus an Excel workbook.

The provider receives the text our server read from your PDF plus your field names or instruction, not the file. On training, pdf.xyz can speak only to its contracts, and the agreements it uses its AI providers under prohibit training on data sent through the API. Anything a provider does outside those terms is out of pdf.xyz's hands, so keep documents that cannot bear that risk away from the AI tools.

Every AI tool on pdf.xyz spends from one pool per connection, 30,774 free characters of AI text a day. One extraction never costs more than 10,258 characters of that pool, whether the PDF is one page or hundreds, so at least 3 extractions fit in a day if no other AI job has used the pool. The charge is taken when the job is submitted, after the upload, and a job our server then refuses or that fails still uses it. A PDF this page can already tell has no selectable text is stopped before upload and costs nothing. And when pdf.xyz's own daily spending cap for AI is reached, every AI tool pauses until the next day, however much of the free allowance is left.

Text past the first 500,000 characters is dropped before the model sees it. The warning on the page, the manifest and extraction.md name the last page read in full whenever that can be worked out. A null was judged only on the text that was read, so the value may still appear later in the document. Extract Pages can cut the file down to the pages that matter before you start.

When the model's reply is not valid JSON, our server can put the job back in its queue for another attempt by itself. Once the attempts run out, the job fails with nothing to download and a message saying a retry may succeed. A wrong value comes from the model's reply or from the text it was given: an OCR layer with misread characters, or side-by-side columns read across as one line, can mislead it. Compare the value with the page, and when a page holds several similar figures, name the exact one you want in a prompt or in a descriptive field name. The model is told to answer null rather than guess. A null means the model gave no single value: it may not have found one, or the text may hold several different values for the field. A value that is not null may still be one of several different values, so to get every value, ask for a list: the array type in a field schema, or a list in your prompt. A single-value field asked of a column of figures is where this happens most, since every row holds a candidate.

Up to 100 MB per file.

Yes. 30,774 free characters of AI text a day for your connection. One run never counts more than 10,258 of them, however long the document is, and the allowance resets daily. However long the document, at least 3 extractions fit in a day if nothing else draws on it. The other AI tools draw on the same allowance.

Up to 500,000 characters of selectable text — roughly 81 pages of dense text, or about 291 of sparsely-set pages. Longer documents are read up to that point only, and the tool tells you which page it reached rather than treating the rest as absent.

What you ask for, in one of two ways: list the fields you want — up to 50, each a string, number, boolean, date or array — or describe it in a prompt. You download a ZIP with the extracted data as JSON, a readable Markdown version of it and a manifest recording how it was made.

Yes. The result is written by an AI model from the PDF's selectable text, and a model can miss or misread details. Check anything you rely on against the document itself.