ChatGPT vs Google Translate

What the studies found

ChatGPT came out ahead in some comparisons, but not in every language. If you've switched from Google Translate, the research gives you reasons to keep using ChatGPT and reasons to check that choice.

Kong et al., in BMJ Quality & Safety 2025, compared ChatGPT-4 with Google Translate on emergency department discharge instructions. Here's the share of sentences judged accurate:

• Spanish: 97% for ChatGPT-4 and 96% for Google Translate, with no significant difference.
• Chinese: 95% versus 90%.
• Russian: 89% versus 80%.

The study also counted instruction sets with at least one error. For Chinese, that was 24% with ChatGPT-4 versus 56% with Google Translate. For Russian, it was 56% versus 66%.

That's a different measurement. A percentage of accurate sentences isn't the same as a percentage of instruction sets without errors.

Brewster et al., in Pediatrics 2024, found clinically significant errors in 33.3% of Haitian Creole translations from ChatGPT, compared with 23.3% from Google Translate and 8.3% from professional translators. For Spanish, the figures were 3.3% for ChatGPT and 6.7% for Google.

Where ChatGPT wins

The clearest results in ChatGPT's favor here are Chinese and Russian in the Kong study. ChatGPT-4 had a higher share of accurate sentences in both languages, and fewer instruction sets contained an error.

Spanish is less clear-cut. ChatGPT had fewer clinically significant errors in the Brewster study, though the gap wasn't significant, and the Kong study found no significant difference in sentence accuracy between ChatGPT-4 and Google Translate.

Does that make ChatGPT the better choice for the text you're pasting today? These studies don't settle that. They tested medical instructions, not your particular text, and the results changed with the language.

Still, if your reason for switching was accuracy, there's evidence behind that choice for Chinese and Russian in the discharge instructions tested. It isn't a win across every comparison.

Where Google Translate is safer

For Haitian Creole in the Brewster study, Google Translate had fewer clinically significant errors than ChatGPT. Professional translators had fewer than either tool.

That result matters if you've started treating ChatGPT as an automatic upgrade. In that comparison, it wasn't.

Google Translate also covers almost 250 languages. That's a separate consideration from accuracy: language coverage tells you what's available, while the studies above tell you how the tools performed on specific medical texts.

And neither study gives you an error-free option. Even the higher sentence-accuracy results came with errors elsewhere in the instructions.

Cost: chat plans vs per-character APIs

As of November 2026, Google Translate is free in its app and on its website. ChatGPT's free plan allows unlimited everyday text chats, and Plus costs $20 a month.

For pasting text by hand, you don't have to pay for either. But API pricing is a separate comparison from what you pay to use a chat window.

Google Cloud Translation charges $20 per million characters after 500,000 free a month. OpenAI's gpt-6.1-sol costs $2 per million input tokens and $10 per million output tokens; gpt-6-luna costs $0.10 and $0.50, respectively.

What does that look like in characters? One token is about 4 characters of English, so translating 1 million characters works out to roughly 250,000 input tokens and 250,000 output tokens. That's about $3 with gpt-6.1-sol or $0.15 with gpt-6-luna, before prompt overhead.

Those are estimates. The OpenAI prices count input and output tokens, while Google's price counts characters.

Using either for a website or app

For a website or app, there's also the question of what happens to tags and placeholders, not just how the translated sentences read.

Translinga, made by Bitgate, is a translation API for Shopify stores, WordPress sites and apps. It removes Liquid tags, HTML and placeholders before translating, checks each reply tag by tag, and retries any string that comes back with a tag missing. You can add a $context note to a JSON file to explain what the strings are for; key names count as context too.

Google Translate v2 requests work too, with the same language codes as the DeepL and Google Translate APIs.

Translinga's Fast tier costs $2.50 per million characters, and Quality costs $5.00 per million. Every character counts, including spaces, once per target language. Credit is prepaid from $5 and never expires.

It's in early access: request a key through the Get early access button on translinga.com. No card is needed, and the first $1 is free (about 400,000 characters on Fast).