A photo of a menu in a foreign language, a screenshot from Italian software, a delivery note in Polish: ChatGPT can recognise the text in images like these and translate it straight away. This article shows you how to do it step by step, where the limits are and what you can do when translating individual images turns into a recurring task.

How to translate an image with ChatGPT

1. Open a new chat in ChatGPT, in your browser or in the app.
2. Upload the image using the paperclip or plus icon. In the app, you can also take a photo directly.
3. Add what you need, for example: "Read the text in this image and translate it into English. Keep the order and the paragraphs."
4. Check the result. For numbers, names and technical terms, it's worth comparing with the original.

This also works on the free plan, but OpenAI limits the number of file and image uploads there. If you process a lot of images every day, you'll soon hit that limit.

Prompts that give better results

The more precisely you describe what you want, the more useful the answer. Three examples:

• "Translate the text into English and output it as a table with the columns Original and Translation."
• "This is a delivery note. Extract the supplier, date, item numbers and quantities, and translate the item descriptions into English."
• "Only translate the text on the sign and briefly explain what it means in context."

The second prompt shows what goes beyond simple translation: ChatGPT understands what kind of document it is looking at and can pull out specific fields. This is exactly the capability that matters later for automation.

‍

‍

What it works well for and where the limits are

Translation works reliably for clearly legible printed text: screenshots, photos of signs or menus, product labels, simple forms and presentation slides. Mixed languages in one image are usually not a problem either.

It gets harder with handwriting, blurred or skewed photos and complex tables with many columns. Rows can shift, or individual values are misread. The layout of the original is also not preserved. You get the text, but not a translated image. If you need a finished, designed document in another language, you'll have to lay it out again afterwards.

For legally relevant documents such as contracts or official letters, an AI translation is no substitute for a professionally checked translation.

Data protection: which images you shouldn't upload

OpenAI may use what you upload to ChatGPT to improve its models unless you switch this off under Data Controls in the settings. For a private photo of a menu, that's not an issue. For business documents, it's a different matter.

Don't upload images containing personal data, customer data, health information or confidential business figures to a private ChatGPT account. This includes invoices with customer addresses, job applications or medical certificates. Under the Swiss Federal Act on Data Protection (FADP), your company remains responsible for this data, even when an AI tool processes it. Business use therefore needs a proper setup with clear settings on how data is used.

When one-off tasks become a process

Translating an image in the chat takes a minute. But if 50 foreign-language delivery notes, invoices or forms arrive every week, it becomes a task someone does by hand: upload, write a prompt, copy the result, transfer it to the right system. This is the point where automation pays off.

Example: capturing foreign-language delivery notes automatically

A trading company receives delivery notes by email from suppliers in Italy, Poland and the Czech Republic. An automated workflow with Make could look like this:

1. Make monitors a specific mailbox and picks up every new attachment.
2. The image or PDF is sent to an AI model, for example from OpenAI, Anthropic or Google. The model extracts the supplier, date, item numbers and quantities and translates the item descriptions into English.
3. The result comes back as a structured data record, not as free text.
4. Make writes the data to the target system, such as a monday.com board or the ERP.
5. Uncertain cases, such as an illegible scan, go to a person for review instead of being processed automatically.

That last step matters. A good automation doesn't blindly process everything. It recognises when a person needs to take a look.

Costs can be controlled too. If you connect the AI model in Make using your own account with the provider, each call counts as one credit in Make, and you pay the token costs directly to the AI provider. We explain how Make works and how its costs add up here:

When AI is the right choice and when it isn't

Not every document needs an AI model. If the delivery notes always have the same structure and come in one language, classic text recognition (OCR) with fixed rules is often enough. That's cheaper and more predictable. AI is worth it when documents look different, arrive in several languages or the content needs to be understood, for example to assign an item description to the right internal product category.

How Blinno supports you with AI integration

Blinno is a Make Gold Partner and a monday.com Platinum Partner. We build workflows like these for SMEs, from document capture and automatic reply suggestions to AI agents that take over tasks in the CRM or ERP. We work with the AI tool that suits you, whether that's ChatGPT, Claude, Gemini or Copilot.

We start with a workshop in which we look at your processes and clarify where AI actually helps and where a simple automation is enough. We then implement the use cases and train your team so the solution is used in day-to-day work.

Have a process in mind with lots of documents, images or translations? Book a free first call with us.