Ask PDF / Chat with PDF
Open the app: Ask your PDF a question → (products.documentize.ai)
In one sentence
Ask PDF lets you upload one or several PDFs and talk to them: you ask questions in plain language and get answers whose every claim points back to a page, with the quoted passage highlighted.
What makes it unique
Most AI tools give you an answer and ask you to trust it. Ask PDF is built around checking: each sentence of the answer carries a numbered citation, and clicking it opens the source page with the exact passage marked. It also reads all uploaded documents together — it can say where two reports agree or contradict each other — and it keeps the conversation going, so "and in the second report?" works as a follow-up.
When to use it / when not to
- Use it to find a clause, a deadline, a figure or a definition in a long document; to compare several documents; to get a quick summary you can verify.
- Not for a ready-made one-page summary you can keep → Abstracts (an Abstract page inside the PDF) or Cheat Sheet (a printable sheet).
- Not for a permanent, browsable collection that you want to come back to and extend → PDF Knowledge Base.
- Not for scanned PDFs as they are (see below).
Before you start
- The PDFs must have selectable text. The app's own text says: text-based PDFs, up to 10 files and 100 MB in total. DOCX and other formats must be converted to PDF first.
- A scanned PDF has no text to read. If none of your PDFs has text, the app stops with a message that they "look scanned" and asks you to make them searchable first — use Searchable.
- Have your first question ready: the upload and the first question go together.
Step by step
- Drop your PDFs onto the upload area (or click to choose them).
- Type your first question. The page offers three example questions to try, such as "Where do the documents agree or contradict each other?".
- Choose Answer length (Short or Detailed) and decide whether to keep Link answers to source pages ticked.
- Press Start chat. The app reads your documents, then writes the answer.
- Click a numbered citation to see the source page with the passage highlighted. Ask follow-ups in the box at the bottom; press New chat to start over with other documents.
📷 Screenshot placeholder: an answer with citation badges and the source page viewer open
Options explained
| Option | What it does | Default | Recommended |
|---|---|---|---|
| Your first question | What you want to know (long questions are cut at 2,000 characters) | – | One clear question; refine in follow-ups |
| Answer length: Short | About 2–5 sentences, or a short list if you asked for several items | Short | Quick fact-finding |
| Answer length: Detailed | Covers every relevant point in what was found, organised under short headings or lists; also lets the search gather more passages | – | Summaries, comparisons |
| Link answers to source pages | Turns the [1], [2] markers into clickable citations with highlighted quotes. Off: plain answer text, no markers | On | Keep on — it is your safety net |
You can switch answer length at any point in the conversation. There is no output-language selector: answers come in the language of the page you are using, and you may ask in a different language than the documents are written in (the question is turned into search terms in the documents' language). (verify on the live app)
What you get
- The answer on the page, with citations, a document list (pages and how many had text), and a "How this answer was built" panel showing the steps: how your question was understood, what was searched, how many passages were used.
- A text file of the answer and images of the cited pages with highlights, available as results.
- Your PDFs and the knowledge base built from them are kept for about a day (the app's FAQ) so follow-up questions work, then deleted.
Quality & limits
- AI-generated. The writing prompt tells the model to answer from the retrieved passages only, to say plainly when the passages do not answer the question, and to quote numbers exactly. That reduces invented answers; it does not eliminate them. The page itself warns: "AI answers can be wrong — check the cited pages."
- How it finds things. Every page's text is searchable. In addition, the app extracts key facts (definitions, figures, obligations, findings) from the documents; for long documents this extraction samples the text rather than reading every part, within a time budget. Questions about unsampled pages are still answered from the page search.
- Citations are checked mechanically: markers that point to no retrieved passage are removed, and each kept citation shows the passage text it relies on. That proves the passage exists — not that the model read it correctly. Open it.
- Images are not read. Charts, photos and diagrams are invisible to the chat; only text counts.
Works well with
- Use before: Searchable for scans; Deskew before that for crooked scans; Merger if you would rather ask one combined file; Abstracts to get the lay of the land first.
- Use after: Checklist to turn what you learned into an action list; Cheat Sheet for a printable digest.
- Recipes: Understand a long document fast, Turn a spec into an action plan.
Common mistakes & fixes
- "No extractable text was found" → the PDFs are scans; run Searchable, then upload the result.
- "The answer says the documents don't cover it" → believe it first, then rephrase with the document's own vocabulary; the search is word-based as well as meaning-based.
- Asking "summarise everything" over many long documents → ask per document or per topic; use Detailed.
- Trusting a citation without opening it → always look at the highlighted passage for numbers, dates and names.
- Uploading more PDFs mid-chat → follow-ups carry no files; start a new chat (or use PDF Knowledge Base, which can add documents later).
Same thing elsewhere
- CLI: a Documentize.AskPdf plugin exists (put PDFs in an input folder, ask with a question option, get a Markdown answer with quoted sources). Its description is not identical to the web version, so treat the two as separate tools.
- API: available through the Web API as an app endpoint.
- MCP: no dedicated tool for Ask PDF was found; the MCP "ChatWithDocument" tool is the general Chat app, not this one. See channels.
Try it
Upload two short PDFs on the same subject (say two quotes or two reports). Ask "Summarize each document in three bullet points", then "Where do they differ?" Click three citations and confirm each highlighted passage really says what the answer claims. Then switch to Detailed and repeat the second question.