Ask for a raise with AI

Who is this for?

This page is for someone who already has the job and wants to ask for more pay. You might be an hourly worker, a salaried employee, or a contractor renewing a rate. You want a short case built from work you did, a range you checked yourself, and a way to practice a manager who pushes back. You do not want a chat that invents a "market rate" and then emails your boss.

The chat was not in your year. It does not know your city, your industry, your wins, or what your company pays. Let the AI interview you first is the habit: up to five questions, one at a time, then a summary, then your OK, then a draft. The limits on what it must not invent are Set the limits before you ask. A pay figure is a number you paste, on Give it the numbers. A blank is a blank.

A year-end writeup of the same work is a different page: Write your performance self-review with AI. Practice for a job you do not have yet is Prep for a job interview with AI. The week of applications is Run a job search with AI. An offer you have not accepted, including a first job, is Negotiate a job offer with AI. The note after the interview is Write an interview thank-you with AI. A public profile uses the same honesty rule on Rewrite your LinkedIn profile without inventing jobs. This page is the conversation about pay for the role you already hold.

What do you need before you start?

Write this down before you open a chat. The note is the brief. If a line is missing, leave it missing.

• Your role, in the words your company uses, and how long you have been in it
• The city, or the remote setup, and the industry. You supply these. The chat does not guess a cost of living
• Three to five wins you can point to. A project, a shift you covered, a customer you kept, a problem you fixed. A number only if you can find it in a record you already have
• Your current pay, if you know it and you are allowed to type it. If you do not know the number, write UNKNOWN
• What you want from the conversation, in one sentence. A higher rate, a review date, or a title change with the pay that goes with it. One ask
• The date of the meeting, if you have one, and who will be in the room

You do not need a paid plan. You do need your own facts. A chat that "knows salaries" and has not seen a page you opened is guessing.

When should you not use AI for this?

Close the chat if any of these are true.

• The meeting is a disciplinary record, a legal complaint, or a medical leave. Write that yourself, or use the tool your company named. This page will not tell you what to put in a grievance.
• You were about to paste a coworker's review, a salary spreadsheet of other people, or a message marked confidential. Keep your data private with AI is the line.
• You want the chat to "find the average raise" and treat that percent as yours. There is no percent on this page. If you do not have a figure you copied, the figure stays blank.
• A teacher assigned the negotiation as work you turn in, and the rules say you write it. You still write it.

How do you let it interview you before it writes a case?

A raise case fails when the first message is "write a script that gets me a 10% raise." The chat fills the blanks with a typical employee and a percent it invented. You answer the questions. It does not answer them for you.

Five questions is the cap. One at a time means you can stop a bad question before it becomes a paragraph. If you do not know an answer, say UNKNOWN. Do not ask it to remember your year.

Do not draft a raise script. Do not invent wins, dates, or pay numbers. Do not name a percent.

I want to ask for a raise. I will answer. You will ask.
My role, in my words: [title, or "ask me"].
City or remote setup: [what you will say].
Industry: [what you will say].
Wins I already know I can point to: [short list, or "ask me"].
What I will not claim: results I cannot point to, other people's pay, praise I did not earn.

Ask me up to 5 questions, one at a time. Wait for my answer before the next question.
Ask about the role, the wins, who benefited, and what I want from the meeting.
If I say I do not know, write UNKNOWN. Do not guess a salary, a city, or a metric.
After my answers, summarize them in bullets. Stop. Do not draft yet.

Read the summary. Fix any win that is stronger than what you said. If it added a team size, a dollar saved, or a percent, delete that bullet and send the prompt again. The summary is the only source for the next step.

The same interview, for a self-review you will file, is the self-review guide. Do not mix the two jobs in one chat. A review asks what you did. This page asks what you will say about pay.

How do you build the case from wins you named?

The case is a short list a manager can follow. Each bullet is one win from your summary. If the win needs a number you do not have yet, the bullet says so. It does not borrow a number from a typical employee.

Using only the summary I approved, draft a raise case I can say out loud.
Do not add a win, a date, or a number that is not in the summary.
Do not mention pay yet. Pay stays in the next step.

For each win:
- One sentence I would say
- Who benefited, if I already said so
- If a number would make it stronger and I marked it UNKNOWN, write NEEDS A NUMBER YOU LOOK UP
- If I have no number and the win is still true without one, leave the number out

Keep it to 5 bullets or fewer. Then list the questions I still need to answer from my own records.
Do not write an email. Then stop.

Open your records for every line marked NEEDS A NUMBER YOU LOOK UP. A ticket count, a sales total, an error rate, a headcount. If you cannot find it, the bullet stays without that number. A made-up percent is the line a manager will question.

Catch facts the AI made up applies to your own story. A result you did not live is a claim you would have to defend.

How do you leave pay blank until you paste figures you opened?

This step is where people get hurt. A chat will happily name a salary. That number is not your market until you have seen it on a page, for a job that matches, in a place that matches.

Open these yourself. Copy what you see. Write down the date you looked.

• BLS Occupational Employment and Wage Statistics and the Occupational Outlook Handbook. Search your occupation. Read the area. A national figure is not your city's figure unless the page says you are looking at your area.
• Levels.fyi, if people in your kind of work use it. Read the title and the location on the page. A crowdsourced range is a lead. It is not your company's band.
• Glassdoor and Payscale. Same rule. Copy the range that matches the title and the place. If the page does not match, do not use it.
• A company band, a posted range on a req, or a figure in an HR tool you are allowed to see. That is the number your manager may actually be holding. A public website does not replace it.

If you do not have a figure yet, leave the brackets empty. Do not ask the chat to fill them.

Do not invent pay. Do not estimate a "typical" raise. Do not convert a national figure into my city's pay.

My current pay, if I am allowed to type it: [amount, or UNKNOWN]
Figures I copied myself. Each line includes the source, the job title on that page, the place, the number, and the date I opened it:
[paste only rows you copied, or write NONE YET]

If a row is missing, write BLANK. Do not fill BLANK.
Put the copied figures in a small table: source, title, place, number, date.
Then say, in one sentence, which of those numbers I am willing to say out loud.
If I pasted NONE YET, say I am not ready to name a number. Then stop.
Do not tell me what I "deserve." Do not write the email.

Check the table against the pages. If a row does not match what you saw, delete the row. Give it the numbers is this habit in one move: the chat organizes figures you pasted. It does not go find them.

Your ask can be a range you can explain, or a single number you can explain. If you cannot explain where it came from, you are not ready to say it.

How do you rehearse a skeptical manager?

You practice out loud. The chat plays the manager. It should push back. A rehearsal that only praises you is a yes-man, and a yes-man will not prepare you for the room.

The move is Rehearse the hard conversation. The failure mode is Stop the yes-man. Ask for the hard version. You answer in your words. You do not memorize its sample answers.

Play my manager. Be skeptical. Do not coach me in the same message. Do not praise me.

You have only the case and the pay table I already approved.
If a number was BLANK, you do not know it, and you must not invent it.

Start with one objection. Wait for my spoken answer. I will type what I said.
Then one follow-up that a skeptical manager would actually ask.
After 4 of my answers, stop the roleplay.
Then tell me:
- what landed, in one sentence
- what I dodged
- one fact I stated that was not in my approved case

Do not rewrite my answers into a better speech. Do not email anyone.

Say the answer out loud before you type it. If you can only say it while reading the chat, it is still a script. Prep for a job interview with AI uses the same rule for a job you are trying to get. Here the interviewer is the manager you already have.

If the feedback includes a metric you never said, ignore that sentence. Your case is the ceiling.

How do you map the three answers you might hear?

A meeting ends in a few recognizable ways. Prepare for three. You still choose what to say when you hear them. The chat does not negotiate, and it does not send a follow-up.

Using only my approved case and my approved pay table, draft a one-page map for three outcomes.
Do not invent new wins or new pay numbers.
Do not write an email that is ready to send. Write words I might say, labeled as talk, not as a message.

1. No budget. What I can ask instead: a date to revisit, or a written note of the decision. I choose whether to ask. Leave the date blank unless I already know it.
2. A low number. Questions I can ask about how they got there. Do not tell me to refuse. Do not name a counteroffer unless it is a number I already pasted.
3. Yes. What I repeat back so we share the same number, the same start date, and whether it is hourly or salary. I confirm it in writing myself after the meeting.

Under each outcome, give me 4 sentences or fewer.
Mark anything I must not say unless I looked it up: LOOK UP FIRST.
Then stop. Do not send anything.

Read the map once more before the meeting. Cross out any sentence you would not say. If "yes" includes a start date you did not hear in the room, you do not write that date into a recap later. You ask for it.

After the meeting, if you want a short note, write the facts you heard, then ask for a draft. You send it. A note that adds a number nobody said is how a yes becomes a fight.

Here is what was actually said, in my words: [paste your notes].
Draft a recap of 80 words or fewer that I might send.
Use only those notes. If a number or a date was not in my notes, write LOOK UP FIRST.
Do not add gratitude I did not ask for, and do not threaten.
I will send it myself, or I will not send it.

What mistakes should you avoid?

These are the misses that waste the result. Skip them before you send anything.

Common mistakes

• Starting with "get me a raise" and pasting no wins.
• Letting the chat turn UNKNOWN into a round percent.
• Treating a national BLS wage, a Levels.fyi row, or a Glassdoor guess as your company's band.
• Pasting a coworker's pay "so it knows the market."
• Memorizing the roleplay and freezing when the manager asks something else.
• Asking the chat to email your manager, or to negotiate in your place.
• Sending the recap before you have checked every number against what was said.

What should you keep private?

Your own results can be fine in a personal chat if your company allows that tool. A compensation file, a coworker's pay, and a medical reason you might mention are not context. Follow the company rule. If the only chat you have is a work account, assume an admin can read it. Do not paste a number you were told to keep quiet.

Privacy: what not to paste

Leave out a Social Security number, a home address, a password, and a photo of a pay stub that shows an account number. The rate and the wins are enough.

What should you check before you use it?

Answer each line before you walk into the room.

Director check: look before you use it

• Did it interview you, or did it guess your year? If the first reply was a full script, delete it and send Step 1 again.
• Is every win one you can point to? If you would not say it to your manager, cut it.
• Is every pay number one you pasted from a page you opened? If not, it is still BLANK.
• Can you name the source, the title, and the place for the number you will say? If you cannot, do not say the number.
• Did you practice out loud? A typed answer is not the meeting.
• Are you the one who will speak, and the one who will send any note? The chat does not email. The chat does not negotiate.

What is one extra tip?

Keep the approved summary with the notes from the meeting. Next year you will not have to rebuild the case from memory. The written version of the same year, for the form you file, is Write your performance self-review with AI. If the conversation goes well and you want the public profile to match, use only facts you can say out loud on Rewrite your LinkedIn profile without inventing jobs.

One extra tip

The pushback habit has a name: Stop the yes-man. The rehearsal habit has a name: Rehearse the hard conversation. More job guides are on Job search and Work. Grok can run these prompts on grok.com, in the Grok apps, or on X. The check is the same in ChatGPT, Claude, Gemini, Copilot, or Grok.

Which sources back this page?

Open the page that matches the number you might say. A line on this site is not a wage.

• BLS Occupational Employment and Wage Statistics
• BLS Occupational Outlook Handbook
• Levels.fyi
• Glassdoor
• Payscale

Copy the figure, the title, the place, and the date. If the page does not match your job, leave the cell blank.