Translation App Privacy: What Google, DeepL and ChatGPT Keep

Translation app privacy depends more on the tier and settings you use than on the app itself. Free consumer translators can keep your words and use them to improve AI models, while paid business tiers and fully offline modes typically delete content right after translating or never send it anywhere.

That split matters more in 2026 than it did a year ago. Google is rolling out a new account setting, Search Services History, that now covers Google Translate. When it is on, your activity can be saved to your account and used to train generative AI models. Apple keeps transcripts of Translate requests for two years by default. DeepL's free version uses submitted texts to train its models for a limited period. None of this is hidden; it sits in privacy policies most people never open.

This guide compares what the most widely used translation tools keep, shows the exact settings worth changing, and explains when an offline translation app is the safer choice. If your question is about live AI translation inside video calls rather than everyday apps, read our companion guide on whether AI translation is secure in calls.

Table of Contents

• What Happens When You Tap Translate
• Translation App Privacy Compared: 2026 Policies
• Six Settings Worth Changing Today
• What Never to Paste Into a Free Translator
• Offline Translation: The Most Private Option
• Translation Apps at Work: What the Rules Say
• When the Conversation Is Live
• Frequently Asked Questions

What Happens When You Tap Translate

Every translation app follows the same basic path. Your input leaves the device, a server runs a model, and the result comes back. The privacy questions sit around that path: what gets logged, who can review it, and whether it trains the next model.

The input type changes what you actually hand over. Most people think about the sentence they want translated. The app often receives much more.

Two details from this table catch people out. First, a photo of a document is not "just text". If you point a camera at a passport or a lab result, the image itself can be saved. Second, conversation mode records both sides. The person you are talking to never agreed to the app's terms.

Translation App Privacy Compared: 2026 Policies

The table below summarizes what each provider says in its own privacy documentation, as checked on October 28, 2026. Policies change, so treat it as a starting point and recheck before you rely on it for sensitive material.

A few points deserve a closer look.

Google Translate and the New Search Services History

Google's support page lists Translate among the services covered by Search Services History, alongside Search, Maps and News. When the setting is on, Google says your history can be used "to provide, develop, and improve its services (such as training generative AI models)". Trained reviewers, including staff at Google's service providers, may read some of it. Data used for training is disconnected from your account and kept for up to four years.

So is Google Translate private? Signed out, less is tied to you. Signed in with the setting on, your Translate activity becomes part of the same history as your searches. Google notes one exception: accounts issued by schools and universities are not used to train generative AI models.

DeepL: Free and Pro Follow Different Rules

DeepL's privacy policy draws a sharp line between tiers. The free translator processes your texts for a limited time "to train and improve our neural networks and algorithms". It also states that its terms of use do not allow texts containing personal data of any kind in the free version. DeepL Pro, DeepL API Pro and DeepL Write Pro work differently: texts are kept only as long as needed to produce the translation, then deleted, and never used for training. So is DeepL safe? The Pro tier is built for confidential work. The free tier is not, by DeepL's own rules.

Apple Translate: Anonymous, but Not Ephemeral

Apple links Translate requests to a random, rotating identifier rather than your Apple Account. That is a meaningful protection. But Apple's privacy statement also says it stores transcripts of audio interactions and text inputs for two years by default and may review a subset. The way around this is simple. When both languages are downloaded, Apple says processing takes place on your device.

ChatGPT: Useful Translator, Different Defaults

Many people now translate by pasting text into an AI chatbot. On personal ChatGPT plans, data sharing for model training is on by default. You can switch it off and keep your history, or use Temporary Chat. OpenAI says temporary chats are not used to improve its models but may be retained for up to 30 days for safety purposes. Voice mode has its own separate training control.

Six Settings Worth Changing Today

You do not need to delete every app to improve your translation app privacy. These six changes take about ten minutes in total.

1. Google Translate: open My Google Activity, select Search Services History and either turn it off or uncheck "Save media". Choose "Turn off and delete activity" if you also want past entries gone.
2. Google auto-delete: if you keep history on, set it to auto-delete after the shortest period offered instead of keeping it indefinitely.
3. Apple Translate: download the languages you use most. Once both languages in a pair are on the device, translation runs locally. Also check that "Improve Siri & Dictation" is off unless you want Apple to store and review audio.
4. ChatGPT: go to Settings, then Data controls, and turn off "Improve the model for everyone". Use Temporary Chat for one-off translations of anything personal.
5. DeepL: if you translate work documents, use a Pro account through your employer. Do not paste personal data into the free version at all.
6. Translation keyboards: remove network access from third-party keyboards you no longer use. On iPhone, that permission is labeled "Allow Full Access".

What Never to Paste Into a Free Translator

Free tools are fine for menus, street signs and public web pages. They are the wrong place for anything you would not post publicly. Keep these out of any translation app that may store or train on input:

• Identity documents: passports, ID cards, visas and residence permits, especially as photos.
• Health information: lab results, prescriptions, discharge letters and insurance claims.
• Legal and HR material: contracts under NDA, dismissal letters, disciplinary notes and performance reviews.
• Customer data: lists, support tickets with names and addresses, and order histories.
• Credentials: passwords, one-time codes, API keys and recovery phrases.
• Children's data: school reports and anything else about minors.

The habit is common. In a TELUS Digital survey of 1,000 employees at large US companies, reported by Dataconomy in February 2025, 57% of those using generative AI at work admitted entering confidential information into public AI assistants. 31% had entered personal details such as names and email addresses. 68% reached these tools through personal accounts rather than company-approved ones.

Removing names is not always enough. The Hamburg data protection authority's checklist for LLM-based chatbots warns that context alone can identify people. A "satisfactory" job reference for a salesperson at a named car dealership points to one person even without a name. The same checklist advises opting out of AI training and switching off chat history wherever possible.

Offline Translation: The Most Private Option

An offline translation app keeps the whole process on your phone. Nothing leaves the device, so there is nothing to log, review or train on. For most private one-to-one use, this is the strongest privacy setup available.

The major platforms support it in different ways:

• Apple Translate processes requests on the device once both languages are downloaded.
• Google Translate offers downloadable language packs for offline use. Check which features still need a connection, because camera and conversation modes can behave differently.
• Samsung's Galaxy AI phones include a setting that limits AI features to on-device processing. Features that need the cloud stop working while it is on.

Offline translation has trade-offs. Quality is usually lower than the best cloud models, especially for idioms and long sentences. Fewer language pairs are available. And your phone becomes the only copy, so a lost, unlocked phone exposes your saved translations. Use a passcode and turn off saved history if the material is sensitive.

Translation Apps at Work: What the Rules Say

At work, the question is not only "is this translation app private?" but also "are we allowed to use it?" Three rules come up most often.

GDPR. If staff paste customer or employee data into a translator, the provider processes personal data on your behalf. That requires a data processing agreement under Article 28. Free consumer tiers rarely offer one. Business tiers such as DeepL Pro, Google Cloud Translation and Azure Translator do.

US healthcare. The federal rule implementing Section 1557 of the Affordable Care Act sets a specific standard for machine translation. Under 45 CFR 92.201(c)(3), when a covered entity uses machine translation for text that is critical to a patient's rights or benefits, or where accuracy is essential, a qualified human translator must review it. Privacy and accuracy point the same way here: free consumer apps are not a substitute.

Terms that change. Providers sometimes update their policies to allow AI training on data they already hold. In February 2024 the US Federal Trade Commission warned that quietly changing terms of service to permit AI training through a retroactive amendment could be unfair or deceptive. For you, the practical lesson is to recheck the policy of any translation app your team relies on at least once a year.

For a structured way to evaluate tools for a team, see our overview of real-time translation apps for business.

When the Conversation Is Live

Translation apps are built for short exchanges: a sign, a menu, a quick question at a pharmacy. Business calls with international clients, patients or colleagues are different. Passing a phone back and forth through conversation mode records both voices into a consumer app, under consumer terms, and splits the conversation into awkward turns.

A meeting platform with translation built in keeps the whole conversation under one agreement. Meeyra translates speech live inside the video call, in the browser, across 42+ languages. Audio and video are encrypted in transit. Speech is processed in short segments, and the temporary audio files are deleted once each segment is translated. Meeyra does not store recordings of your voice. When translation is on, a text transcript of the original sentences and their translations is saved to the meeting record so you can review it afterwards. Meeyra does not claim strict end-to-end encryption for translated meetings, because the server has to process audio to translate it. Our guide to end-to-end encryption in video calls explains that trade-off.

Meeyra is not a document or photo translator. For contracts and scanned forms, use a business-tier text translator with a data processing agreement. For live conversations, see how AI translation and voice translation work in Meeyra, and review the protections on our security page.

Frequently Asked Questions

Is Google Translate private?

Google Translate is not fully private when you are signed in and Search Services History is on. Your activity can be saved to your Google Account, used to train generative AI models and reviewed by people. Turning the setting off, or using offline language packs, reduces what Google keeps.

Does Google Translate store your data?

Yes, it can. Google's support page lists Translate among the services whose activity is saved as Search Services History when the setting is on. You can turn it off, uncheck "Save media" or delete past activity in My Google Activity.

Is DeepL safe for confidential documents?

DeepL Pro is designed for confidential work: texts are deleted after translation and not used for training. The free version is not suitable, because DeepL uses free texts to train its models and its terms prohibit personal data there.

Is it safe to translate with ChatGPT?

It is safe for public or non-sensitive text. On personal plans, chats can be used for model training by default, so turn off "Improve the model for everyone" or use Temporary Chat. Do not paste client, patient or employee data into a personal account.

What is the most private translation app?

The most private setup is any translation app running fully offline, with both languages downloaded and history turned off. Nothing leaves your phone, so nothing can be logged or used for training. For work, a business tier with a data processing agreement is the safer cloud option.

Is it safe to translate a passport photo with a translator app?

It is risky with cloud-based camera translation, because the whole image can be uploaded and saved, including your ID number and face. Translate the few words you need by typing them, or use offline mode.

Can the other person's voice be recorded in conversation mode?

Yes. Conversation mode sends the audio of everyone who speaks to the app's servers unless processing happens on the device. Tell the other person you are using a translation app, and prefer offline mode for private conversations.

The safest translation habit is simple: match the tool to the sensitivity of the words. Use free apps for public text, offline mode for personal matters and business tiers or built-in meeting translation for work. If your team needs live translation in calls, you can create a free Meeyra account and test it in your next meeting.