How Docamend answers questions about your documents

1. Reading the document

PDFs are read page by page with the position of every line. Word, RTF, ODT and text files are first converted to fixed pages. Scanned pages are read with OCR, usually in your browser. The text is split into passages that keep their headings, so a clause is indexed together with the section it belongs to.

2. Finding the relevant passages

Each passage is added to a search index. When you ask a question, Docamend searches that index for the passages most likely to hold the answer. Quality and Max first rewrite the question and split it into parts. Small documents can be read in full instead.

3. Writing the answer

The selected pages are sent to the AI model with an instruction to answer only from them and to quote the text each point relies on. If the pages do not contain an answer, the model is asked to say so.

4. Checking the quotes

Quotes are matched against the page text and drawn as highlights. In Quality and Max, points whose quotes cannot be found on the page are not presented as supported, and Max searches again for parts of the question left open.

What it costs

Uploading and asking use credits; deeper modes cost more because they read more and make more AI calls. The cost is shown before anything runs, and a run that fails is refunded.

Limits to know

• Search can miss a relevant passage that uses very different wording from the question. Rephrasing, or using Quality or Max, helps.
• The AI model can still misread what a quoted passage means. The quote lets you check; it does not guarantee the interpretation.

Questions