Grammarly can now fix your Spanish and French grammar

For 16 years, a team of linguists carefully crafted and honed the grammar editing software Grammarly to match natural English language patterns. Now, the company is getting a big assist from AI to expand similar offerings to five more languages: Spanish, French, Portuguese, German, and Italian. This expansion caters to what has been the “number one feature request” since the company’s founding, according to Grammarly’s VP of enterprise product, Luke Behnke. In addition to catching spelling mistakes, the app reworks sentences and paragraphs to match the tone of native speech and improve the overall clarity — now in six languages. Grammarly also translates the six core languages into 19 different languages “without having to leave the tool and go to another translation provider,” Behnke says. The competition for AI-based software for languages other than English is heating up, with Google Search’s AI Mode just broadening the number of languages by five and Apple’s latest AirPods debuting a live language translation feature. Grammarly indicated a broader ambition to become an AI productivity app, starting with the acquisition of the buzzy email application Superhuman in July, followed by the launch of nine AI agents designed for students and educators. The company says that its software has been built on machine learning from the get-go. “We call it an agent. ... I know that’s maybe an overused AI word at this point, but it is kind of the first agent that people ever used,” Behnke says of Grammarly’s proofreading product. Large language models entered the picture in 2023. The software is built on open-source large language models that are fine-tuned by analytical linguists. The number of new linguist teams did not “linearly scale” with the increased number of languages offered by Grammarly, Behnke said. The company used “a smaller group of people” in addition to internal feedback and evaluations of the new languages, Behnke says. Suggestions and paragraph rewrites emerge from these in-house models “hosted on our infrastructure with our own security controls and very tight training rules,” Behnke says. [Image: https://platform.theverge.com/wp-content/uploads/sites/2/2025/09/GEC_Underlines_Spanish_V2.png?quality=90&strip=all] In the beta rollout to “about a million” users, native speakers of the five new languages accepted Grammarly’s suggestions at rates similar to English speakers. The rollout took some users by surprise when the familiar red lines suddenly appeared under text written in their native language, Behnke says. Grammarly also uses third-party LLMs for advanced features. Users can choose to incorporate an external LLM from major companies such as OpenAI, Behnke says. While third-party LLMs cannot train on Grammarly user data, Grammarly does train its own models on user data. Enterprise and education customers automatically have training turned off, while other users have to opt out of training. While the company has not disclosed which languages are in the pipeline, Behnke says that Grammarly customers specializing in customer support services have requested more languages in the global areas where offshore call centers are located.