How to use ChatGPT for work when your written English is shaky
You will end this with a one page brief you can dictate in under two minutes and reuse every month. The shape of the request decides what comes back. Grammar has nothing to do with it. Fragments work. Hindi words work. Dictation works. What changes the quality is whether you said the five things before you asked.
How do I use ChatGPT for work if my English is not strong?
You give ChatGPT five pieces of information in any order and in any English you have. Role, context, task, format, constraints. Grammar does not enter the calculation. Missing context does.
Take one real job you already do by hand. A spare parts distributor with two branches and 40 people pulls a stock and sales statement out of the accounting software every month. The owner then spends a Saturday deciding what to reorder and what has gone dead. That statement and that decision are the example for the rest of this.
Most guides put a paragraph of polished English in front of you and tell you to copy the style. That is the wrong thing to copy. Copy the five slots. Fill them in broken sentences, in Hinglish, in a voice note, in a list of half lines. The output does not improve because your sentence was correct.
Say the five parts out loud before you type a word
Name the five parts before you open the chat, because the thinking is the work and the typing is not. Keep each one to a line. You are not writing, you are listing.
For the spare parts statement, the five lines go something like this. Role: you are working with a distributor of automotive spare parts in a tier two city. Context: two branches, 40 people, the attached file is last month's stock and sales, the slow movers tie up money before the festive order. Task: find the items that have not moved for three months and sit above a normal stock level. Format: a table of item, branch, months without movement, stock value, and a one line note on each. Constraints: use only the attached file, flag anything where the data looks incomplete, no recommendations yet.
It is done when a person who has never seen your business could read those five lines and know what you want. Check one thing before you send it. Did you say what the answer is for. The constraint line is where most of these briefs are thin, and a thin constraint line is how you get a confident table built on numbers that were never in the file.
Dictate the brief, then set your standing instruction once
Dictate the brief instead of typing it, then correct the transcript on screen. Speak your five lines the way you speak to your accountant. For the spare parts statement, say the two branches, the three month cutoff and the columns you want. Then attach last month's statement before you send.
Fix the transcript in two places only. Make sure the numbers came through correctly, and make sure the task line says one thing and not three. Item names in Indian businesses transcribe badly, so an internal part code often survives better than a spoken product name.
Then stop repeating yourself. ChatGPT has a custom instructions setting that applies to every new chat. Put your standing rules there once. Short sentences. Plain English, no jargon. Say which numbers came from my file and which did not. Ask me for the missing piece instead of assuming. Written once, it stops being something you have to remember in English every morning.
Check the output before you use it, and know what to do when it is wrong
Run three checks on every table before you act on it, and do them on the numbers you already know. Pick two rows you can verify from memory or from the file. Add the stock value column and see whether the total is close to what your own statement says. Then look for any row you cannot trace back to a line in the file you attached.
For the slow moving stock table, the failure looks ordinary. A branch name appears that you closed last year. A part shows three months of no movement when you sold it in bulk last week. A total does not tie to the statement. None of that means the exercise failed, and none of it is a language problem.
What you do then is narrow the brief, not rewrite it in better English. Send back one line: this row is wrong, here is the actual figure, redo the table using only rows that appear in the attached file. Then make the same correction part of the constraint line next month. What you build over a few months is the habit of judging an output before you act on it. That habit gets built in your own language, in the numbers you already hold.
How do I learn AI when I am not comfortable writing in English?
Can I write the prompt in Hindi or Hinglish?
Yes, and you can ask for the answer in English if that is what your team needs to read. Mixing languages inside one request is normal. Keep the part numbers, item names and figures in the form they appear in your own records, because those are the things a transcript or a translation gets wrong first.
Should I learn prompt writing or learn my own work better?
Your own work, every time. The five slots take an afternoon. Knowing that a three month no movement flag is meaningless for a seasonal item comes from running the business, and that is the part no course supplies. The brief is only a way of handing over what you already know.
Can I hand this to my assistant to run every month?
Hand over the typing, keep the judgment. Your assistant can dictate the same five lines, attach the statement and bring you the table. You still decide which flagged items are genuinely dead and which are seasonal. If the brief leaves that decision to whoever runs it, the brief is not finished.