What Are Developers Building with PyMuPDF?

How Thousands of Inbound Inquiries Reveal the Changing Landscape of Document AI

Most software companies rely on surveys or market research to understand what their customers want. We have access to a slightly different kind of data.

Over the past four years, developers have submitted thousands of inbound inquiries after discovering PyMuPDF and we’ve taken a look into what this data can tell us.

These were not leads collected through a specific marketing campaign. They were developers who found PyMuPDF and voluntarily described what they were building. As a result, this dataset is less like a traditional marketing funnel and more like a market signal showing what developers are currently trying to build with PDFs.

We analyzed these inquiries in anonymized, aggregated form and asked one question:

What are developers actually building with PyMuPDF?

The answer was clearer than we expected.

Most Projects Begin with Document Extraction

Among the inquiries where the intended use case could be identified, 74.6% mentioned extracting text, tables, or document structure.

Other tasks, such as conversion, rendering, redaction, and compression, also appeared consistently, but at substantially lower rates. For developers, PyMuPDF was not simply a tool for displaying PDFs or converting them into other formats.

It was a tool for turning the information contained in PDFs into reusable data.

A few years ago, this finding might not have been particularly surprising. Document extraction has long been one of PyMuPDF’s core use cases.

What has changed is why developers are extracting documents.

Extraction Is Now the First Step in an AI Pipeline

Mentions of AI, RAG, and LLMs began increasing rapidly in CRM inquiries.

Among the inquiries where the intended use case could be identified, 39% involved systems related to AI, RAG, or LLMs. The specific applications varied, including document question-answering systems, internal knowledge search, contract analysis, research retrieval, and document-based agents.

But they all had one thing in common:

They first needed to extract the information contained in PDFs.

Before an LLM can answer questions about a contract, report, or manual, the document must be converted into a form the system can read. If the reading order is incorrect, table structures disappear, or headings cannot be distinguished from body text, the quality of both retrieval and generated answers declines.

In other words, the quality of a document AI system is determined before the document ever reaches the model.

From this perspective, document extraction and AI are not separate use cases. They are two ways of describing the same workflow from different points of view.

From the developer’s perspective, the task is extracting information from a PDF. From the perspective of the overall system, the task is creating input data that an AI model can use.

Demand for AI Use Cases Is Growing Rapidly

This does not appear to be a temporary trend.

The share of inquiries mentioning AI-related use cases increased from 12.5% in 2023 to 38.5% in 2024 and 42.5% in 2025.

Year Share of AI-Related Use Cases

• 2023 12.5%
• 2024 38.5%
• 2025 42.5%

It was not only the percentage that increased. The overall volume of inbound inquiries also grew significantly during the same period.

This is difficult to explain solely as existing document-tool users adopting AI-related terminology. It is more likely a signal that a growing number of developers and companies are actively looking to connect PDFs to AI systems.

PDFs are no longer limited to being final documents that people read and archive.

They are increasingly becoming a major input format that AI systems must read, search, and analyze.

How Developers Discover PyMuPDF Is Also Changing

What developers are building is not the only thing that has changed. The way they discover PyMuPDF is changing as well.

Among the inquiries that included information about how the respondent first learned about PyMuPDF, GitHub was the most common source. ChatGPT ranked second, followed by the Artifex website and search engines.

How Developers Discovered PyMuPDF Share

The fact that ChatGPT ranked above search engines is particularly noteworthy.

In the past, a developer might have searched for a phrase such as “Python PDF extraction” and compared several libraries. Increasingly, developers are describing their problem to ChatGPT and evaluating the tools it recommends.

AI is not only part of the systems being built with PyMuPDF. It is also becoming a channel through which developers discover PyMuPDF.

The use cases are changing at the same time as the product discovery process.

Document AI Is Moving Beyond the Experimental Stage

AI-related demand was not limited to inquiries about free, open-source usage.

AI-related use cases were also common among PyMuPDF Pro inquiries from organizations evaluating commercial licensing and technical support.

This suggests that document AI is no longer limited to individual experiments or weekend projects. Companies are beginning to integrate document-processing capabilities into production services and business systems. As they do so, they are evaluating practical concerns such as performance, reliability, licensing, maintenance, and support.

The market is moving from experimenting with models to building document AI systems that can operate reliably in production.

Good Document AI Starts with Good Extraction

PyMuPDF was not originally developed with the goal of becoming AI infrastructure.

For years, the focus has been on solving a more fundamental problem: reading PDFs quickly and accurately, and making their text, images, coordinates, and structure accessible to developers.

However, as the AI industry has increasingly adopted document retrieval and RAG to produce more reliable answers, the importance of this foundational work has grown.

A retrieval system cannot retrieve information that is better than the text it receives. Likewise, an LLM cannot fully reconstruct the meaning of the original document if the document structure has already been damaged during extraction.

The Quality of AI Output Cannot Exceed the Quality of Document Extraction

PyMuPDF4LLM was designed to convert PDFs into Markdown that can be used in RAG and LLM pipelines. And PyMuPDF can extend these capabilities to Office documents, complex tables and layouts, and the support requirements of commercial environments.

However, the most important finding from this CRM analysis was not about the features of any particular product.

It was that the purpose for which developers use PDFs is changing.

PDFs are quietly becoming one of the primary input formats for AI. This is not a market forecast based on what we expected to happen. It is a shift that is already visible in the inquiries developers have submitted over the past four years.