Short answer: To use AI to correct your writing in a new language, write a short text on your own first, then paste it into ChatGPT, Claude or another AI assistant with a prompt that asks it to explain every mistake, give the rule behind it and match your CEFR level. Better still, ask it to only mark the errors so you try to fix them yourself before it reveals the answer. Keep a running error log and review it weekly. The copy-paste prompts below turn each correction into a small lesson instead of a silent rewrite.
Why a Correction Only Helps When You Understand It
If you paste a paragraph and ask an AI to "fix this", you get a polished text and learn almost nothing. The next time you write, the same errors come back. The point of correction is to notice the gap between what you wrote and what a proficient writer would write, and to understand why.
Research supports the value of written correction, with a caveat. A meta-analysis of 21 studies by Kang and Han (2015), published in The Modern Language Journal found that written corrective feedback can lead to greater grammatical accuracy in second language writing, but that its effect depends on factors such as the learner's proficiency, the setting and the type of writing task. In practice, that means corrections should fit your level and you should actively work with them, not just read them.
Your level matters because the Council of Europe's CEFR global scale describes very different writing goals at each stage: a B1 writer "can produce simple connected text on topics which are familiar or of personal interest", while a C1 writer produces "clear, well-structured, detailed text on complex subjects". Telling the AI your level stops it from correcting a B1 text as if it should read like C1.
Set Up Every Correction Session the Same Way
Before any of the prompts below, give the AI four pieces of context. It takes ten seconds and makes the feedback far more useful:
- Your level: your CEFR level (A2, B1, B2...) or a description such as "I can write simple emails but make many tense mistakes".
- Your native language: many errors come from transfer, and the AI can explain them better if it knows where they come from.
- The language for explanations: your native language at A1 to B1, the target language from B2.
- The purpose of the text: a casual message, a work email or an exam essay need different corrections.
If you are new to chatting with AI for study, our guide on how to use ChatGPT for language learning covers the basics. Here we focus only on writing correction that teaches.
Five Copy-Paste Prompts That Make the AI Teach
The examples below assume an English speaker learning Spanish. Replace the language, level and text with your own. The sample text contains typical learner errors: "Ayer yo fui a la playa con mis amigos y estaba muy divertido. Estoy emocionado para el viaje de la próxima semana porque la problema es que nunca he visto la nieve."
1. Explain each mistake
This is your default prompt. It forces the AI to treat each error as a mini lesson.
I am learning Spanish at B1 level. My native language is English. Correct the text below. For each mistake, show: (1) what I wrote, (2) the correction, (3) a one-sentence explanation of the rule in English, (4) one new example sentence using the rule. Only correct real errors, not style. Then give me the full corrected text. Text: [paste your text]
With the sample, a good answer flags estaba muy divertido (should be fue muy divertido, a completed event in the past), emocionado para (should be emocionado por) and la problema (problema is masculine: el problema).
2. Self-correction first, then reveal
Finding and fixing your own errors makes you think harder about each one. Ask the AI to hold back the answers.
I am a B1 Spanish learner. Do not correct my text yet. Instead, rewrite it with each error marked like this: [ERROR: type], where type is tense, gender, preposition, word choice, spelling or word order. I will try to fix them myself. Then check my attempt, tell me which fixes are right, and explain the ones I got wrong. Text: [paste your text]
3. Recast or explicit correction
A recast is a quiet reformulation: the AI rewrites your sentence correctly without naming the error, the way a friend might repeat what you said in better words. Explicit correction names the error and the rule. Recasts keep the flow and suit fluency writing such as journaling or chat messages. Explicit correction suits exam preparation and errors you keep repeating. You can ask for both at once:
Give me two versions of feedback on my Spanish text. Version A: a recast only, my text rewritten naturally with the changes in bold and no explanations. Version B: explicit correction, a numbered list of each error with the rule. I will first compare my text with Version A and guess why each bold part changed, then read Version B to check. Text: [paste your text]
4. Rewrite at my CEFR level
AI rewrites often jump to a level you cannot yet produce. This prompt keeps the model text within reach, plus one small step up.
Rewrite my Spanish text so it is correct and natural for a B1 writer. Use only grammar and vocabulary a B1 learner can reasonably use. Then show me one sentence rewritten at B2 level and explain which structure makes it B2, so I have a next step. Keep my ideas and my meaning. Text: [paste your text]
5. Build an error log
Single corrections fade. An error log shows you which mistakes keep coming back, so you can target them.
From the corrections above, create an error log table with these columns: Error, Correction, Rule (short), Category (tense, gender, preposition, word choice, other). Then tell me which category appears most and give me five short fill-in-the-blank sentences in Spanish to practice it. Do not show the answers until I reply.
Copy the table into your own notes or spreadsheet. Do not rely on the chatbot to remember it: memory features differ between tools and settings. Recurring items make good flashcards, as explained in our article on how spaced repetition works in AI apps.
Which Prompt to Use When
| Prompt type | Best for | What you learn | Effort |
|---|---|---|---|
| Explain each mistake | Your default for any text | The rule behind every error | Low |
| Self-correction first | Errors you probably know but miss | Spotting and fixing your own mistakes | High |
| Recast | Journals, chat messages, fluency writing | Natural phrasing without interruption | Low |
| Explicit correction | Exam practice, repeated errors | Named rules and error types | Medium |
| Rewrite at my CEFR level | Seeing a model text you can copy from | Reachable structures plus one next step | Low |
| Error log | Weekly review | Your personal error patterns | Medium |
A Weekly Writing Correction Routine
About 20 minutes a session is enough. The key is reusing corrections, not writing more. Set targets that match your level, as described in our guide to realistic CEFR goals.
| Day | Task | Prompt |
|---|---|---|
| Monday | Write 100 to 200 words on your week, without a dictionary if possible | Explain each mistake |
| Tuesday | Write a new short text on a different topic | Self-correction first, then reveal |
| Wednesday | Rewrite Monday's text from memory, then compare | Rewrite at my CEFR level |
| Thursday | Write a quick message or journal entry in ten minutes | Recast, then explicit correction |
| Friday | Review the week's errors and practice the top category | Error log |
| Weekend | Say the corrected sentences out loud or use them in a conversation | None, or a speaking session |
Pitfalls to Avoid
Over-correction
Without limits, AI tends to rewrite your style, swap simple words for rare ones and produce a text that no longer sounds like you. Add "only correct real errors, and list at most three style suggestions separately" to your prompt. If the corrected version is far above your level, use the CEFR rewrite prompt.
The AI can be wrong
Language models sometimes invent rules, flag correct sentences as errors or mix regional varieties, such as Spain and Latin American Spanish, or British and American English. When a correction surprises you, ask "Is this a real error or a style preference? How confident are you?" and check the rule in a grammar reference, a good dictionary or with a teacher.
Letting the AI write for you
If the AI drafts and you only edit, you practice reading, not writing. Always write the first version yourself, even if it is short and full of mistakes.
Privacy
Do not paste personal details, other people's names or confidential work emails. Replace them with placeholders. Check the data settings of the tool you use: OpenAI's Data controls in ChatGPT page says you can turn off "Improve the model for everyone", and that Temporary Chats are not used to improve its models but may be retained for up to 30 days for safety purposes (checked November 12, 2026). Other assistants have their own settings.
Where Speaking Practice Fits
Writing correction builds accuracy because you have time to think. Speaking forces you to use the same structures in real time, which is where many learners stall. Once a structure is fixed in your error log, practice it out loud: with a human tutor, a language exchange partner or an AI speaking tutor such as Enverson AI. Our Find Your Fit matcher helps you choose a speaking tool that suits your level and goals.
Frequently Asked Questions
Can ChatGPT or Claude correct my writing in another language?
Yes. General AI assistants can correct short texts in many languages and explain each error if you ask them to, but they can also get rules wrong, so check anything surprising in a grammar reference or with a teacher.
Should the AI explain my mistakes in my native language or in the target language?
At A1 to B1, explanations in your native language are usually easier to understand and remember. From B2 upward, asking for explanations in the target language turns each correction into extra reading practice.
Is it better to get a corrected text or an explanation of each mistake?
Use both. The explanation teaches you the rule and the corrected text shows natural phrasing. A meta-analysis by Kang and Han (2015) found that written corrective feedback can improve grammatical accuracy, with effects that depend on factors such as proficiency, so try to correct yourself before you look at the answer.
How often should I write for AI correction?
A short text of 100 to 200 words, three to five times a week, is easier to review properly than one long essay. What matters most is that you reread the corrections and reuse the fixed structures in your next text.
Is it safe to paste my writing into an AI chatbot?
Leave out names, addresses and confidential work details, and check the tool's data settings. In ChatGPT, for example, you can turn off Improve the model for everyone or use Temporary Chat, according to OpenAI's help center (checked November 12, 2026).