Export

Programs usually store their content in their own, internal format. As long as you stay within the same program, this is not a problem. As soon as you want to use the content elsewhere, you need a translation. That’s exactly what an export does: the program writes its data into a file that other tools can also understand. The counterpart is the import, meaning the reading in of such a file. You can think of an export like translating a text into a language that many people speak — more understandable for everyone, but rarely without some loss of nuance.

Why data shouldn’t stay locked inside a program

Anyone who cannot get their data back out is tied to a provider. Experts call this lock-in, meaning a dependency that makes switching expensive. A functioning export feature is the antidote. It ensures that a move to a different product remains possible.

In Europe, this is even enshrined in law. The General Data Protection Regulation gives every person the right to receive their own data in a common, machine-readable format. That’s why Google, Meta, or Spotify now offer buttons that let you download a complete archive of your account. This is exactly what the right to data portability is about.

For companies, there’s also a practical reason. Data is generated in one place and analyzed in another. Without export, everything would have to be typed in twice. A clean export is therefore the foundation of almost every automation.

From internal format to exchangeable file

During an export, the program reads its internal data and writes it out anew. In doing so, it chooses a defined format, meaning an agreed-upon rule for how the characters are arranged in the file. Well-known examples are CSV for tables, JSON for nested data records, and PDF for documents with a fixed layout. The file extension in the filename usually reveals which format was chosen.

Almost every export loses something. A spreadsheet exported as CSV contains only numbers and text, no formulas and no colors. This is called lossy. The simpler and more widely supported a format is, the more details typically get lost. Anyone who wants to keep everything stays in the program’s proprietary format — and loses compatibility in return.

For AI models, export means something related. A fully trained model consists of many millions of numbers, known as weights. These numbers are written into a file together with a description of the architecture. Common formats for this are ONNX and Safetensors. After that, the model can run on a different system, for example on a phone instead of in a data center.

Export buttons in everyday life and in tech news

In everyday life, export shows up as an inconspicuous menu item. A chat history can be saved as a text file, an account balance as a CSV for a tax return, a presentation as a PDF for printing. Downloading your own music statistics also falls into this category. In all these cases, a file is created that lives on independently of the original service.

In the news, the term often comes up in disputes over market power. When authorities require platforms to provide better export options, it’s about competition. With AI companies, on the other hand, you read about model exports into open formats so that developers can run the models themselves.

A common misconception: an export is not a backup. A backup is meant to restore the original state exactly. An export, by contrast, delivers a simplified version for other programs. Anyone who exports their archive and then deletes their account has often saved less than they thought.