ChatPDF is a German-built document AI that answers questions about PDFs, Word files and slides with page-level citations, free for two documents a day.
Used this tool? Rate it
Used this tool? Rate it
ChatPDF is a document AI that lets you ask questions of a file instead of reading it. Upload a PDF, type a question, get an answer with a citation pointing to the page it came from. The official framing is direct: it brings ChatGPT-style intelligence and PDF AI technology together for smarter document understanding — summarize, chat, and analyze.
The company behind it is unusually well documented, because German law requires it. The Impressum names ChatPDF GmbH at Bornholmer Str 67, 10439 Berlin, represented by Mathis Lichtenberger, registered at the Amtsgericht Berlin Charlottenburg under commercial register number HRB 280133 B. That is a level of corporate transparency most AI tools never approach — you can look this company up in a public register, which is not something you can do with the average chat-with-your-PDF site.
Positioning is narrower than the name suggests in one direction and broader in another. Narrower, because this is not a general assistant: it answers questions about documents you give it. Broader, because the ability to chat with PDF files long ago stopped being the whole story — the supported formats now include Word (.doc, .docx), PowerPoint (.ppt, .pptx), Markdown (.md) and plain text (.txt), and the product has grown adjacent tools for writing, AI detection, YouTube transcripts, flashcards and slides.
The audience is stated plainly: best-in-class AI tools for students and researchers. The three user personas on the front page are researchers exploring scientific papers and books, students studying for exams and homework, and professionals navigating legal contracts, financial reports and manuals. The vendor reports more than 10 million users and over a million questions answered daily, alongside an a16z Gen AI 2024 Top 50 badge — figures published by the company rather than independently audited.
One thing to flag at the outset, because it matters more than any feature: this product's entire premise is that you hand it your documents, and its privacy policy does not address what happens to them. That finding is documented in detail in the Limitations section, and anyone uploading anything sensitive should read it before anything else on this page.
The single most useful design decision here is page citations. Built-in citations anchor responses to PDF references, eliminating page-by-page searching. In the API this surfaces as inline markers in the form [P2], [P234], accompanied by a structured references array giving the page numbers used.
This matters because it changes the failure mode. An uncited answer from a language model is something you either trust or verify from scratch. A cited answer tells you exactly where to look, which converts verification from a chore into a click. For any use where being wrong has consequences — legal, medical, academic — citation is the difference between a usable tool and an interesting toy.
Create a folder, add your files, and the system will understand relationships across all documents, allowing questions that reference multiple sources, compare content, or find connections between different documents in a single conversation. For a literature review or a set of related contracts, this is materially more useful than uploading files one at a time, because the interesting questions are usually comparative.
The company answers the model question directly rather than hiding behind "advanced AI." Its platform does smart dynamic routing for queries between GPT-4o and GPT-4o-mini, optimizing for a balance of speed, quality and efficiency. Two implications follow. First, this is a wrapper around OpenAI models with document retrieval built on top — competent engineering, but the underlying reasoning is not proprietary. Second, routing means the model answering you varies by query, so response quality is not perfectly consistent by design.
ChatPDF is fully multilingual: you can upload documents in any language and chat in your preferred language, including uploading a document in one language and asking about it in another. For anyone reading research published outside their working language, this collapses a translate-then-read workflow into a single step.
The interface keeps the chat and PDF open together, with answers linked to the original content so verification means glancing right rather than opening another window. Combined with clickable citations that scroll the document to the exact source, this is the specific interaction the company argues makes it better than pasting a PDF into a general chatbot.
There is a real API, not a marketing gesture. Sources can be added three ways — manually on the site, from a publicly accessible URL, or by direct file upload — each returning a source ID. The chat endpoint is at api.chatpdf.com/v1, supports streaming, can return page references, and includes an endpoint for deleting sources. Documentation states the constraints openly, which the Limitations section covers.
Beyond document chat, the site offers an AI Writer, an AI Detector, YouTube Chat, a Research/Scholar surface, AI Flashcards and AI Slides, plus a downloadable Mac application. Whether these are as strong as the core function is not established by the material reviewed here; they are noted as present rather than endorsed.
The best-evidenced use. Upload a paper, ask what the method was, what the sample size was, what the authors concede in the limitations section — and get answers pointing at specific pages. The academic framing is not marketing invention: the widely circulated endorsement that made the product visible was a researcher describing it as ChatGPT for research papers.
Professionals navigating legal contracts, financial reports, manuals and training material is an explicitly targeted scenario. Locating an indemnity clause or a specific figure in a long report is exactly the retrieval-shaped problem this tool handles well — with the important caveat about confidential documents in the Limitations section.
Students revising for exams, working through homework and answering multiple choice questions faster. The flashcard tool extends this. Worth noting that a Play Store reviewer praised the summarisation for exam revision while criticising it for not simply producing assignment answers — arguably the tool behaving appropriately.
Folder-based multi-file chat suits comparing several papers on one topic, reconciling versions of a contract, or synthesising a set of reports. Comparative questions are where document AI earns its keep.
Upload in one language, ask in another. For international research, translated technical documentation, or foreign-language contracts, this removes a whole step.
The lowest-effort use: upload something long, ask what it is and whether it contains what you need. At two free documents a day, this costs nothing and answers the question of whether full reading is warranted.
No account is required to start; the service is free to try and works immediately. Drop a file into the upload area on the homepage. Creating a free account later unlocks saved history and multi-document chats.
"Summarize this" produces a generic summary. "What sample size did they use and how did they justify it?" produces something you can actually act on, with a citation. The retrieval architecture rewards specificity — a targeted question gives the system a clearer target to locate in the document.
When an answer includes a page reference, click it. This takes two seconds and is the entire basis for trusting the output. Answers whose citations do not support them are the failure mode worth catching, and citation makes catching it easy.
For anything comparative, create a folder and add the files rather than chatting with each separately. Cross-document questions only work when the documents share a conversation. Note the Plus limit of 50 PDFs per folder.
Files are limited to 2,000 pages or 32 MB. Long theses, scanned books and image-heavy reports can exceed the size limit even when well within the page limit — worth checking before assuming an upload failure is a bug.
The free tier allows analysing two documents every day. If you routinely exceed that, Plus removes the document cap. If you exceed it occasionally, the free tier resets daily and waiting is free.
The platform states you retain full control of your data with the ability to delete your documents at any time, and the API provides a delete endpoint. Given the ambiguity about document handling discussed below, deleting sensitive files after use is a reasonable habit rather than an excess of caution.
The reason to use this over pasting text into a general chatbot is verifiability. If you are not following citations, you are using a worse general chatbot. Build the click into your habit.
The API documentation reveals that the chat endpoint is stateless and caps requests at 6 messages, with older messages dropped once the total exceeds 2,500 OpenAI tokens. Even if the web interface manages this for you, the underlying architecture favours self-contained questions over long conversational build-ups where earlier turns may silently fall out of scope.
Two documents a day is plenty for deciding whether a paper is relevant. A quick PDF summary is often all you need to decide; reserve paid usage for documents you have already established are worth the depth.
The product routes text to language models. Scanned image PDFs and complex tables are inherently harder for this class of tool, and there is no official documentation reviewed here promising OCR quality. If results seem poor on a scanned document, the document is a likely explanation.
Monthly is $13.99; yearly works out to $8.25 per month, advertised as a 40% saving. Since all purchases are non-refundable, the annual commitment is worth making only after the monthly plan has proven itself for your workflow.
This deserves repeating because the product actively invites contract and financial report uploads. The privacy policy does not describe how uploaded documents are handled, retained, or whether they reach model providers. For genuinely confidential material, that absence of documentation should be treated as a reason for caution.
Citation makes this easy, so do it. Retrieval reduces fabrication substantially; it does not eliminate it, and a confidently wrong answer about a contract clause is worse than no answer.
Researchers and academics. The strongest fit: as a research assistant it is the audience that drove the product's initial visibility. Literature review, methods extraction and cross-paper comparison are all well served.
Students. Explicitly targeted, with exam revision and homework support plus flashcards. The free tier's two documents a day suits study patterns better than it suits professional workloads.
Professionals handling routine documents. Contracts, reports and manuals — provided the documents are not confidential. That proviso is doing real work; see Limitations.
Multilingual readers. Anyone regularly reading documents outside their primary working language gets disproportionate value.
Developers. The documented API with page references makes it viable for building document Q&A into other products, within the stated size and token limits.
Who should look elsewhere. Anyone handling genuinely confidential or regulated material — client files, patient records, unreleased financials — should not use a product whose privacy policy is silent on document handling without first getting written answers from the vendor. Anyone needing OCR-grade extraction from scanned documents should test carefully first. Anyone whose documents routinely exceed 2,000 pages or 32 MB needs a different tool. And anyone wanting a general-purpose assistant should use one — this is deliberately narrow.
ChatPDF is primarily a browser-based web application at chatpdf.com, and the company states it is designed to be compatible with all devices, requiring only a web browser. The interface localises across multiple languages, with paths such as /zh alongside the English default.
A Mac desktop application is available via download.chatpdf.com. An Android app is published on Google Play by ChatPDF GmbH, holding a 4.7 rating across 6,440 reviews with more than 100,000 downloads and an Everyone content rating — the largest independent sample of user sentiment available for this product.
The site also advertises a mobile app more generally. An attempt to confirm an iOS listing through Apple's official lookup API returned no result for the expected bundle identifier, so no claim is made here about iOS availability; the absence of a confirmed listing is not proof that none exists.
For programmatic use, api.chatpdf.com serves the documented v1 API.
The free tier allows analysing two documents every day. In the company's own words, it offers a free plan that lets you analyze 2 documents every day, with no account required to begin. This is a genuine free tier rather than a trial — it resets daily and does not expire.
ChatPDF Plus costs $13.99 per month, or $8.25 per month billed annually, advertised as a 40% saving. Plus benefits are listed as unlimited PDFs, unlimited questions, 2,000 pages per PDF, 50 PDFs per folder, and a high quality model.
Two details deserve attention. First, "unlimited" is bounded: the 2,000-page-per-PDF and 50-PDFs-per-folder limits apply on the paid tier too, and these appear only in the upgrade dialog — not in the public FAQ. Second, the "high quality model" phrasing, read alongside the disclosed GPT-4o and GPT-4o-mini routing, suggests paid users are weighted toward the stronger model, though the documentation reviewed here does not state that explicitly.
All purchases are non-refundable per the terms of service. You can cancel at any time from your account, and cancellation takes effect at the end of the current paid term. Combined with the annual plan's savings, this makes monthly-first a sensible approach.
There is no separately published API pricing page in the material reviewed here.
ChatGPT with file upload is the obvious comparison, and the company addresses it directly: its argument is that a purpose-built side-by-side interface with clickable citations that scroll to the source makes document work faster than a general tool. That is a real advantage for document-heavy work, while ChatGPT remains far more capable at everything else.
ChatDOC and AskYourPDF occupy the same category with similar core propositions. Choosing between them typically comes down to citation quality, file size limits and pricing for your specific volume — details worth checking directly, since this class of product changes quickly.
NotebookLM from Google takes a different shape: source-grounded research notebooks rather than per-document chat, with a strong free tier. For synthesising many sources into an ongoing project it is arguably the better metaphor.
Adobe Acrobat AI Assistant integrates into the PDF tool many professionals already use, which for enterprises with existing Adobe agreements can settle the procurement question by itself.
Self-hosted RAG — an open-source retrieval pipeline over your own infrastructure — is the answer for confidential material. It requires real technical work and offers nothing like this polish, but documents never leave your control, which for some material is the only acceptable design.
The honest summary: ChatPDF's differentiators are citation quality, cross-language handling and a genuinely usable free tier, backed by unusual corporate transparency. Its disadvantage is that its privacy documentation has not kept pace with what the product does.
This is the most significant finding on this page, and it is stated precisely because precision matters here.
ChatPDF's privacy policy, last updated 5 January 2023, is a general website privacy notice. A word-level review of the full text found no occurrence of "upload", none of "train", none of "OpenAI", and none of "processor" or "sub-processor". The three appearances of "document" all refer to the policy text itself, in phrases like "the contact details at the bottom of this document". The policy's sections cover what information is collected, how long it is kept, minors, and user rights — the standard template topics — applied to account data, cookies and social logins. It states that the company does not process sensitive personal information and does not receive any information from third parties, and its retention rule is the generic formulation that information is kept only for as long as it is necessary for the purposes set out in the notice, with no specific number of days given.
What it does not do is describe what happens to the files you upload. For a product whose entire function is receiving your documents and processing them through third-party language models, that is a substantial gap. Note carefully what this does and does not mean: it does not mean documents are used for training, and it does not mean they are not. It means the vendor has not committed either way in its published policy, and users cannot resolve the question from public documentation.
Contributing context: the policy predates significant product evolution — the file-format expansion, the adjacent tools, the API — and the terms of service, updated more recently in February 2024, contain no training language either. The relevant published commitments are narrower ones: the platform states documents are SSL-encrypted in transit and encrypted at rest with a SOC2 Type II certified storage provider, that users can delete documents at any time, and that documents are strictly private by default. Those are meaningful security statements. They are not statements about training use or about which sub-processors see document contents.
Anyone uploading confidential material should seek written confirmation from the vendor rather than inferring from silence.
Google Play's data safety section for the Android app declares no data shared with third parties, while listing collection of location, personal info and four other data types with encryption in transit and deletion available. Set beside the disclosed routing of queries to GPT-4o and GPT-4o-mini, this warrants noting. The likely reconciliation is that Google's framework treats model providers as service providers rather than third parties, which is a common and defensible reading. Both facts are recorded here as observed; readers processing sensitive documents may want a clearer answer than an inference.
The Impressum lists ChatPDF GmbH at Bornholmer Str 67, Berlin, with VAT ID DE362088018. The terms of service describe the company as registered at Lammertzweg 19, Schleswig-Holstein, 24235 Laboe, with VAT number DE347986415. Google Play's developer information matches the Berlin address. The most plausible explanation is a relocation with the terms not updated, but that is inference — the documented fact is that two current official documents disagree. For most users this is trivia; for anyone needing to serve notice or verify the counterparty, it is not.
Files are limited to 2,000 pages or 32 MB each, disclosed in the API documentation rather than the public FAQ. The 50-PDFs-per-folder cap appears only in the upgrade dialog. Users with long or image-heavy documents can hit these without warning from the main marketing surfaces.
The chat API is stateless, accepts at most 6 messages per request, and silently drops older messages once the total exceeds 2,500 OpenAI tokens. That is a modest conversational window, and the dropping is automatic rather than flagged — relevant if you are building on the API and expecting long conversational memory.
Searches for coverage in established technology press — TechCrunch, VentureBeat, ZDNet — returned no independent reporting on the product or its founder. What exists is company database entries, founder posts on LinkedIn, and affiliate-oriented review sites, none of which constitute independent verification. This is stated as a fact about the evidence base: the product's popularity is real and measurable through app store data, but it has not been the subject of scrutiny by technology journalists.
Against a strong 4.7 average from 6,440 Play Store reviews, the critical reviews are specific. One reports the app closing automatically during study sessions and, notably, that after purchasing the premium plan there was no help or complaint email address to report the issue. Another paying user found summarisation good for exam revision but felt the product fell short at producing assignment answers — arguably reasonable behaviour, but a mismatch between what was expected and what was delivered.
Because the product routes to GPT-4o and GPT-4o-mini, its reasoning is bounded by those models. Anything they get wrong about a document, this will get wrong. The value added is retrieval, citation and interface — genuinely valuable, but not a different class of reasoning.
Two further contractual points are worth knowing. The privacy policy states the company does not knowingly collect data from or market to children under 18 years of age, and that by using the Services you represent that you are at least 18 or the parent or guardian of a minor — a higher stated floor than many consumer AI tools set. Separately, and reassuringly, the terms state that the company does not assert any ownership over your Contributions and that you retain full ownership of them along with any associated intellectual property rights.
All purchases are non-refundable. Cancellation stops future billing at the end of the current term but does not recover money already paid. Start monthly.
The 10 million users and one million daily questions on the front page are company-published figures. They are plausible given app store download numbers, and they are not independently audited. Treat them as marketing claims rather than verified metrics.
Yes, with a real limit. The free plan allows analysing two documents every day, and no account is required to start. Beyond that, ChatPDF Plus provides unlimited document analysis at $13.99 monthly or $8.25 per month billed yearly. The free tier resets daily rather than expiring, which makes it genuinely usable for light work.
Beyond PDF, it supports Word (.doc, .docx), PowerPoint (.ppt, .pptx), Markdown (.md) and text (.txt). The size ceiling is 2,000 pages or 32 MB per file, stated in the API documentation. Paid folders hold up to 50 PDFs each.
The company answers this directly: it performs smart dynamic routing between GPT-4o and GPT-4o-mini, balancing speed, quality and efficiency. Answer quality is therefore bounded by those OpenAI models; the product's own contribution is document retrieval, page-level citation and interface.
The published documentation does not say. This is the most important caveat about this product. Its privacy policy, last updated January 2023, contains no mention of uploads, training, or sub-processors, and the terms of service are also silent on training. The company does commit to SSL encryption in transit, encryption at rest with a SOC2 Type II certified storage provider, documents being strictly private by default, and deletion on request. But on the specific question of training use, there is no published commitment either way — so for confidential material, ask the vendor directly rather than assuming.
ChatPDF GmbH, a German company. The Impressum names Bornholmer Str 67, 10439 Berlin, represented by Mathis Lichtenberger, registered at Amtsgericht Berlin Charlottenburg under HRB 280133 B. Note that the terms of service list a different address in Laboe, Schleswig-Holstein, and a different VAT number — the two official documents disagree, most likely because of an unupdated relocation.
Yes, and this is the product's main advantage over pasting text into a general chatbot. Built-in citations anchor responses to specific PDF pages, and in the API they appear as inline markers such as [P2] alongside a references array of page numbers. Clicking a citation scrolls the side-by-side view to the source.
Yes. Create a folder, add your files, and the system reasons across all of them in one conversation — comparing content, referencing multiple sources and finding connections. Paid folders are capped at 50 PDFs.
Yes, and it handles mixed-language use well. You can upload a document in any language and ask questions in your preferred language, including asking about a foreign-language document in your own — explicitly designed for international research.
Yes, documented at chatpdf.com/docs/api/backend with endpoints under api.chatpdf.com/v1. Sources can be added by URL or file upload, chats support streaming and page references, and sources can be deleted. Two constraints matter: the chat endpoint is stateless, and it accepts at most 6 messages per request with older ones dropped past 2,500 OpenAI tokens.
No. The terms of service state that all purchases are non-refundable. You can cancel any time from your account, effective at the end of the current paid term. Given that, and given the free tier and monthly option, testing before committing annually is the sensible sequence.
