Chat (create a PDF by chat)
Open the app: Create a PDF by chat → (products.documentize.ai)
In one sentence
Chat lets you describe a document in plain words and get it back as a formatted PDF — then keep chatting to change only the parts you want changed.
What makes it unique
- Every answer is a PDF. You do not copy text from a chat window; the reply is the document you download.
- It plans before it writes. Ask for a number of pages and the AI first makes an outline, then writes each section, and adds a Table of Contents page and a "generated on" timestamp.
- Page-by-page revision. On a follow-up ("change the ending", "add a chapter about pricing") only the pages that need to change are rewritten; the rest keeps its content.
- It edits your own PDF. Upload a PDF and say "fix the typos", "translate to German" or "rename Bob to Alice": the app decides this is an edit, and returns an
_editedversion in which unchanged pages are copied as they were and changed pages are set as plain text. - It reads links. Paste a web address (or a shared DeepSeek chat) alone in the message and you get that page's text, with menus and ads stripped, as a PDF.
Do not confuse it with Ask PDF: Ask PDF is for understanding one document (questions and answers). Chat is for producing documents. It does overlap — you can upload a PDF and ask a question, and the answer comes back as a PDF — but its strength is creating and revising.
When to use it / when not to
- Use it for first drafts: reports, guides, explanations, outlines, meeting notes, story drafts, a clean PDF of a long web article.
- Use it to make small wording changes to a PDF you already have.
- Not for factual or legal final copy without checking. AI text can be wrong; see Trust and verification.
- Not for layout design. The output is clean text pages with headings; use Add Table, Illustrator and Merger to enrich it.
- Not for Q&A across many files — that is PDF Knowledge Base.
Before you start
- Have a clear request: purpose, audience, tone, length.
- Optional: a PDF for context or for editing. Only PDFs are accepted. Scanned PDFs without a text layer give the AI nothing to read — run Searchable first.
- Decide on the Pages number if length matters; leave it on Auto otherwise (the field accepts 1 to 50).
- Replies come in your interface language.
Step by step
- Open Chat. The left panel is the conversation; the right panel shows your PDF.
- (Optional) Set Pages to the length you want and attach a PDF with the paperclip.
- Type your request and press Send. Progress messages tell you what is happening ("Planning the outline…", "Writing page 3 of 6…").
- Read the PDF in the output panel and download it.
- Send a follow-up to revise: the next version updates the document. A new chat starts from scratch.
📷 Screenshot placeholder: chat on the left, generated PDF with Table of Contents on the right.
Options explained
| Option | What it does | Default | Recommended |
|---|---|---|---|
| Message | Your request, or a single link to turn into PDF | – | State topic, audience, tone, structure |
| Pages (1–50) | Target page count; more than 1 triggers outline + section-by-section writing with a table of contents | Auto | Set it when you need a specific length |
| Attach PDF | Gives the AI your document as context, or as the thing to edit | none | Edit requests: say exactly what to change |
| Follow-up message | Revises the previous document page by page | – | One change at a time |
What you get
- A PDF with headings and paragraphs in plain text style (no markdown symbols), plus — for multi-page documents — a Table of Contents page and a timestamp. Right-to-left languages are laid out right to left.
- For an edited upload:
<yourfile>_edited.pdf. Your original is never changed. - Results are kept about two days.
Quality & limits
- AI-generated: content may contain errors and invented details. Treat as a draft.
- Text-only output: no images, colours or complex tables are produced.
- Uploaded PDFs: very long documents are summarised in parts before answering, so detail can be lost; edit mode applies to PDFs of up to 200 pages, longer ones are answered rather than edited.
- Links: up to 3 web links are read per message, only public pages. A link that cannot be read is reported rather than guessed.
- Time: one generation has a working budget of about 12 minutes; if it runs out during a long revision, pages not yet reached are left unchanged. Free-plan time limits may also apply.
- The site's FAQ states uploaded files and generated documents are removed automatically after processing.
Works well with
- Use before: Searchable (for scanned inputs), Extract Text.
- Use after: TOC AI or Add TOC for clickable bookmarks, AI Illustrator, Converter (PDF → Word to polish it), Compressor.
- Recipes: Manuscript to illustrated e-book, Lab: from idea to illustrated PDF.
Common mistakes & fixes
- "The PDF is only one page" → set Pages above 1; Auto lets the content decide.
- "My follow-up rewrote everything" → name the page or section you want changed.
- "Nothing happened after I uploaded a scan" → no text layer; make it searchable first.
- "I pasted a link plus a question and got a different result" → a message that is only a link becomes a clean PDF of the page; add words around it and the page is used as context for your question instead.
- "There is no bold or bullet formatting" → the output is deliberately plain; style it afterwards in Word via Converter.
Same thing elsewhere
- MCP: the
ChatWithDocumenttool (query, optional file URL, system message) returns a task you poll. Whether it returns a PDF in the same layout as the web app: verify. - CLI: the plugin
Documentize.Chatanswers a prompt about an uploaded PDF. See Beyond the browser.
Try it
Set Pages to 4 and send: "A one-page-per-topic guide to caring for a small vegetable garden for beginners." Download the PDF and check the Table of Contents. Then send "Make the page on watering shorter and add a tip about mulch." Confirm the other pages are unchanged. Finally upload that PDF and send "Translate to Spanish" to see an _edited version.