ChatGPT prompts for a job search that don’t invent your experience

An AI chat assistant writes from what you give it. Give it too little and it fills the gaps with things that sound right and aren’t yours. So start every job search chat with the same block of facts.

• Your resume, as text: copy it out of the PDF.
• A few lines on each project: what you built, the hard part, the link, and any number you can prove.
• What you want: the kind of role, the cities or remote, your batch year or years of experience.
• Start a new chat for each job, so one company’s post doesn’t leak into another’s email.

These prompts work the same in ChatGPT, Claude, Gemini or any other chat assistant. Brackets are for your own text; replace every one.

A hiring post on LinkedIn mixes the role, the person, the address and the must-haves into one paragraph. This prompt pulls them apart, and copies the address exactly, so you write to the right person.

Here is a hiring post. Read only what it says; don't guess.

[PASTE THE WHOLE POST]

Tell me, in short lines:
1. The role, the company, and the place (or remote).
2. Who posted it, and how they want to hear from people: an email address, a form or a message. Copy any address exactly as written.
3. The must-haves and the good-to-haves, as two lists.
4. Years of experience or batch years, and any deadline.
5. Anything that asks me to pay, or looks unusual.
If the post doesn't say something, write "not stated".

A post that asks you to pay for training, a test or a joining kit is a warning sign; don’t pay. If it gives no address, see how to find the right email address for a job, and never guess one.

This is where invented experience creeps in. Asked to “make my resume match”, a model may add the skills the post wants. Asked to pick from your own lines and quote them, it can only use what is there.

Below are my facts: my resume and my notes on my projects. After them is a hiring post.

<my facts>
[PASTE YOUR RESUME AND PROJECT NOTES, WITH LINKS]
</my facts>

<post>
[PASTE THE POST]
</post>

Pick the 3 to 5 lines from my facts that best answer what the post asks for. Quote each line word for word, then say which requirement it answers. Use only my facts. If the post asks for something my facts don't show, list it under "Not in my facts" and don't fill the gap.

The “Not in my facts” list is useful too. A must-have on it is a reason to skip the job, or to say honestly in the email that you are learning it. A good-to-have on it is fine to leave out. To fit the resume itself to the post, see how to tailor your resume to a job description.

A job email that gets read is short and specific: which role, one piece of proof, and why this company. Give the model that shape and the lines you picked, and it has nothing left to make up.

Write a cold email to [NAME, THEIR ROLE] at [COMPANY] about the [ROLE] opening in this post. Use only the lines I picked. Add no skill, tool, number or result that isn't in them.

<post>
[PASTE THE POST]
</post>

<lines>
[PASTE THE LINES FROM THE LAST STEP]
</lines>

Three short paragraphs, under 150 words in all:
1. Who I am, and which role I'm writing about.
2. The one piece of my work that best answers the post, with its link.
3. Why this company, from this one thing I noticed: [WHAT YOU NOTICED ABOUT THEM].
Plain words. No flattery, no "I hope this email finds you well", no exclamation marks. End with my name and phone number: [NAME, PHONE].

The third paragraph needs one thing you noticed yourself: their product, a feature they shipped, a line in the post. A model asked to say why you like a company it knows little about will write flattery. More on the shape in how to cold email for a job, or fill in the free cold email builder, which uses templates, not AI.

Give me 5 subject lines for this email. Put the role first, then my name and one fact a recruiter would look for: my batch year, my years of experience or my main skill. Under 60 characters each, written like "Frontend Developer: Neha Rao, B.Tech 2026". No questions and no "Application for".

[PASTE THE EMAIL]

Pick the one a busy reader would understand without opening the email. If you wrote to a post, its role title, word for word, is the best start.

A week ago I sent the email below and got no reply. Write a follow-up I can send as a reply in the same thread.

<first email>
[PASTE IT]
</first email>

<new>
[ONE NEW THING SINCE THEN, OR "nothing new"]
</new>

At most three sentences. Name the role once. Add the new thing if there is one; if there isn't, keep it to two sentences. Don't apologise, don't blame, and don't write "just checking in" or "bumping this".

Send it about a week after the first email, as a reply in the same thread, and only once. The reasons are in following up after no reply.

Run this on every email, whether a model wrote it, you did, or both. It catches the two things that cost you most: a claim you can’t back up in an interview, and a phrase that tells the reader nobody wrote this for them.

Check this email before I send it. Don't rewrite it yet.

<email>
[PASTE YOUR DRAFT]
</email>

<my facts>
[PASTE YOUR RESUME]
</my facts>

1. List every claim in the email (a skill, a tool, a number, a result) and quote the line in my facts it comes from. Mark any claim you can't find as NOT FOUND.
2. List the phrases that sound machine-written or like a template, such as "I am writing to express my keen interest", "passionate", "leverage" or "I hope this finds you well".
3. Tell me if the email is too long to read in one look on a phone.
Then give me a corrected version without the NOT FOUND claims and the phrases from point 2.

Then read the corrected version aloud. If a sentence isn’t something you would say, change it.

AI makers say this themselves. Anthropic’s guide to reducing these mistakes says even the most advanced models “can sometimes generate text that is factually incorrect”, and suggests habits that the prompts above already use:

• Tell it to use only the facts you gave it, not what it knows in general.
• Let it say it doesn’t know: the “not stated” and “Not in my facts” answers.
• Ask it to quote the line each claim comes from, and drop any claim it can’t support.

Three more of our own:

• Never ask it to “make my resume stronger” or “add relevant skills”. Both invite invention.
• Numbers only from your facts. A model will happily write “improved performance by 40%” for work you never measured.
• Read every line before it goes. The interviewer will ask about it, and you are the one who answers.

A chat assistant can read a post you paste, but the posts themselves are found on LinkedIn, X and company sites. LinkedIn’s post search has its own rules (quoted phrases, capital OR, and nothing past five operators): see how to find hiring posts on LinkedIn.

In Jobreach, the page called Prompts holds Search Prompts: ready-made searches for LinkedIn and X that find hiring posts asking for your resume by email, made from your kinds of role, your level and the places you pick. You run them on LinkedIn or X yourself; Jobreach never signs in to either or searches them for you.