Can ChatGPT Fix ESL Writing? I Did It for Two Years

Mikhail Konkov · Founder, Limato

Non-native English speaker living in Spain. Spent two years pasting every important email into ChatGPT before building an alternative.

Yes — mostly. ChatGPT strips the ESL tells out of your writing better than any grammar checker I used before it, and it costs nothing. I did exactly this for two years: every important email copied into a chat window, rewritten, pasted back. This article is the honest version of that experience, and the part nobody warns you about: it does not make your writing shorter, it writes a different email every time you ask, and once in three runs it will invent a commitment you never made.

Everything below is a real test, run today in chatgpt.com. One email, one prompt, three regenerations, every output pasted verbatim — including the parts that don't flatter my own product.

The email, before anything touched it

This is what a competent non-native professional email looks like. Nothing in it is wrong. That's the point — it passes every grammar checker and still announces, in the first line, that English isn't your first language.

The tells are all structural, not grammatical: the ceremonial opener, "in order to inform you" where a native writes "to let you know", "the reason of this delay is that" instead of "because", and "we remain at your entire disposal" — a phrase translated straight out of Spanish, French, or Russian business correspondence, where it is completely normal. In English it sounds like a hotel concierge from 1958.

What ChatGPT actually fixes

I pasted it in with the prompt almost everyone uses: "Make this email sound like a native English speaker wrote it." First output, verbatim:

Credit where it's due, because most articles on this topic skip this part: the ESL markers are gone. No "I hope this email finds you well." No "in order to inform you." The concierge phrase is gone. The date is fixed — "the 20th of July" is grammatically fine, but almost no American would write it. It even switched to contractions, which is the single fastest way to stop sounding like a textbook.

If you're a non-native speaker who isn't doing this yet: start. A free ChatGPT account will improve your professional English immediately, and nothing below contradicts that.

Now the part I didn't notice for about a year.

Failure 1: it doesn't make you shorter — it makes you smoother

The original was 126 words. That rewrite is 110. Twelve percent.

Non-native business English is not 12% too long. It is roughly twice as long as it needs to be, because every hedge we were taught — "we would like to kindly ask you if it would be possible" — is a native writer's "could we". ChatGPT swapped the vocabulary and left the scaffolding standing: the pleasantry line, the announcement that an email is being written, the standalone apology, the standalone "if you have any questions", the standalone "I look forward to hearing from you". Five sentences that carry no information, all of them still there. It smoothed the surface of a long email. It's still a long email.

And that was the best of the three runs. Which brings us to the next problem.

Failure 2: three runs, three different emails

I hit regenerate. Same chat, same prompt, same text, sixty seconds later:

Two things happened here, and both are worse than they look.

It came back longer than the email I wrote. 126 words in, 131 out. I asked it to make my English sound native, and it handed me back a longer version of a message whose entire problem was length.

It made a promise on my behalf. "This extra time would allow us to finish the remaining work and deliver the milestone to the expected quality." I did not write that. I did not think it. That sentence commits me — to a client who is already annoyed — to both a delivery date and a quality bar, on behalf of a developer who is at home with a fever. It reads beautifully. It is exactly the kind of sentence you paste back without reading closely, because by email number two hundred, that is what you do.

Nobody warns you about this, because it doesn't look like a hallucination. There's no fake citation, no invented statistic. The prose is perfect. It just isn't what you said.

Third run, for completeness:

This one kept "We'd like to ask if it would be possible to extend the deadline" — which is my original ESL hedge, "we would like to kindly ask you if it would be possible", with the "kindly" removed. A native writer asks: "Can we push the deadline to July 27?" The exact construction I went to ChatGPT to get rid of survived the rewrite.

So: three runs, three lengths (110, 131, 117 — average 5% shorter than what I wrote), one invented commitment, one surviving ESL tell. All three are good English. None of them is my English, and they aren't even each other's.

Over two years, this is the part that quietly costs you. Your writing never develops a voice, because there isn't one — there's a model rolling dice inside a plausible range, and the range is "generic competent professional." Colleagues who read a lot of your email feel it before they can name it: everything is fine, nothing is you.

Failure 3: the round trip is the whole cost

Here's the workflow, honestly counted:

1. Write the email in Gmail.
2. Select, copy, switch tab.
3. Type the prompt again (or scroll up to find the chat where you typed it last week).
4. Paste, wait, read.
5. Copy the output — and lately it doesn't even come back as a message. All three runs above landed in a canvas document, complete with a title ChatGPT invented for it ("Request to Extend the Second Milestone Deadline"). Now you're extracting an email out of a document.
6. Paste into Gmail. Fix the Markdown. Delete the title.
7. Switch back. Reread. Notice the invented commitment — if you're still paying attention.

Forty seconds, maybe sixty. Which is nothing for a contract proposal, and absurd for a two-line Slack reply to a colleague. So what actually happens — and I watched myself do this for two years — is that you use it for the emails that feel important and skip it for the fifty small messages a day where your English is being judged just as much. The tool only reaches the writing you already thought was worth protecting.

The prompt that gets you most of the way

If you're going to keep using ChatGPT for this — and for long documents, you should — stop asking for a vibe and start setting constraints. This is the prompt I converged on after a lot of bad emails:

The two lines doing the real work are "do not add any information, offer, or commitment" (that's the fix for the invented promise) and the explicit compression target. Left alone, ChatGPT does not compress at all — my three runs came back at 110, 131 and 117 words against an original of 126. You have to name a number, or it will happily hand you back something longer than what you wrote.

What the prompt doesn't fix: you still carry it around, it still fades on long documents, and it still lives in another tab.

What a dedicated tool does differently

This is the gap I eventually got tired of and built Limato to close. It is not a smarter model — it's the same class of model with the rules baked in, running inside the field you're already typing in. Select the text, press a shortcut, the rewrite replaces it in place. Gmail, LinkedIn, Slack, Notion, a CMS, a form on some airline's website.

The same email, through the Native tone, unedited output:

126 words to 56 — 55% shorter, where ChatGPT's three attempts averaged 5%. The hedge is gone: "we would like to kindly ask you if it would be possible to postpone" became "Can we push the deadline to July 27?". Nothing was invented — no quality bar, no delivery promise. Every fact in the output was in the input, because that is the rule the tone is built around, and it isn't a line in a prompt you have to remember to type.

You can try that on your own text without installing anything — the free ChatGPT humanizer runs one rewrite in the browser, no signup.

Be honest about the trade-off, though: it went in harder on register. "Dear Mr. Peterson" became "Hi Mr. Peterson", and if you're writing to a Japanese enterprise client, that's wrong and you should use the Formal tone instead. Aggressive compression is a choice, not a free lunch — it's the right one for the 90% of messages where non-native English is too long and too stiff, and the wrong one for the ceremonial 10%.

So should you stop using ChatGPT?

No — and I'd distrust anyone selling you a rewriting tool who says otherwise.

ChatGPT is better than a dedicated rewriter whenever the writing is a thinking problem: structuring a long document, arguing with an outline, working out what you actually want to say to a difficult client. You're in the chat window anyway. There's nothing to switch to.

The round trip only becomes absurd at the point of writing — the Slack reply, the LinkedIn comment, the support ticket, the fifty small messages where your English is being read and quietly assessed by people who will never mention it. That's the writing ChatGPT never reaches, because forty seconds of copy-paste is more than the message is worth. It's also, over a year, most of the English anyone ever sees from you.

Two rules I'd keep from the two years:

• Never paste a rewrite back unread. Whatever tool produced it. The invented sentence always sounds better than the one you wrote — that's precisely why it's dangerous.
• Fix the small writing, not just the important writing. Nobody forms an impression of your English from the one proposal you agonised over.

Related reading: 12 ESL patterns native speakers rarely use — the specific habits the rewrites above are removing. And why I built Limato, which is the long version of the two years this article compresses.