Use AI to ask for a raise: 5 prompts and a sample email

What should a raise request email written with AI say?

A raise request email asks for a meeting, states one number with an effective date, and previews the case in three quantified lines, all in under 150 words. Here is a filled one. The person and figures are illustrative; the structure is what you copy.

Subject: 20 minutes to discuss my compensation

Hi Dana,

Could we find 20 minutes before the October budget review to talk about my pay? I would like to ask for a 5% increase, from $65,000 to $68,250, effective January 1.

The case in three lines: 1. I cut the monthly close from 9 days to 4 by rebuilding the reconciliation workbook, which freed about 20 analyst hours a month. 2. I took over vendor onboarding when Sam left in March and cleared a 40 item backlog by June. 3. I trained both new hires on the team, and each was working without supervision within six weeks.

I can bring the details. Would Thursday or Friday afternoon work?

Thanks, Priya

Every line comes from one of the five prompts below. The three results are the output of the case prompt, the 5% is the output of the number prompt, and the wrapper is the email prompt. What the email leaves out is deliberate: no personal expenses (they invite a "we all have bills" reply), no colleague's salary (it turns the case into a complaint), and no negotiation, which belongs in the meeting.

A 3.2% raise is a pay cut in 2026, because prices rose 3.4%

The average 2026 US merit raise is smaller than inflation. Mercer's US Compensation Planning Survey of more than 1,000 employers, fielded 20 to 31 October 2025, puts the 2026 merit budget at 3.2%, and the Consumer Price Index rose 3.4% in the 12 months to August 2026, per the Bureau of Labor Statistics release of 11 September 2026. On the median full-time salary, $1,251 a week in the second quarter of 2026 (BLS usual weekly earnings) or $65,052 a year, keeping pace with prices takes $2,212. The merit average pays $2,082.

The median worker already beat the merit budget. The BLS release puts it this way: "This was 4.6 percent higher than a year earlier, compared with a gain of 3.9 percent in the Consumer Price Index for All Urban Consumers (CPI-U) over the same period." So a request at the merit average asks for less than the typical full-time paycheck grew, and less than prices rose.

Most people never make the request. In the Federal Reserve's Survey of Household Economics and Decisionmaking for 2025, 17% of workers said they asked for a raise or a promotion, down 3 points from 2022, and 50% said they received one. Asking works: LendingTree's survey of 2,049 US adults in October 2024 found 82% of full-time workers who asked for a raise in the past year got one.

The case prompt turns a messy list of wins into three bullets a manager can repeat

The case prompt produces three quantified results your manager can carry to their own manager or to HR in three sentences, which is how most raises get approved. Paste your wins as they come out of your head, messy and unquantified, and let the model do the sorting.

Here is a rough list of what I did this year, in no order: [paste your list, including anything half remembered]. My role is [title], I have been in it since [month, year], and my current pay is [number]. Turn this into the three strongest bullets for a raise request. Each bullet must name one result and quantify it with a number I supplied or can supply; where I gave no number, write [NUMBER NEEDED] instead of inventing one. Lead each bullet with the outcome, not the activity. Order them by how much they matter to the company, not to me. Then list which items from my list you dropped and why, in one line each.

The placeholder instruction stops the model from decorating your work with a metric you never measured, which is the fastest way an AI written raise request falls apart in the room. The "what you dropped" line is your audit: if the model discarded the project you are proudest of, you probably described an activity ("ran the migration") rather than a result ("cut the monthly close from 9 days to 4").

Fill every [NUMBER NEEDED] yourself before moving on. A bullet with a real number beats two without.

How much is a reasonable raise to ask for in 2026?

For the same job done well, ask for at least 4.6%, the year over year rise in median full-time weekly pay (BLS, second quarter of 2026), which also clears 3.4% inflation. For a bigger job than your title, ask for the promotion average, 8.7% in Mercer's 2026 survey, and use the word promotion, because that money comes from a different budget line than merit. The benchmarks, all US and all for 2026, as of September 2026:

The number prompt applies that rule to your pay and shows the arithmetic, so the figure in the email is one you can defend:

My current base pay is [amount] a year and I have been in my role since [month, year]. These are the 2026 US benchmarks: inflation 3.4% for the 12 months to August 2026 (BLS), average merit budget 3.2% and average promotion increase 8.7% (Mercer), median full-time pay up 4.6% year over year (BLS). Here are my three bullets: [paste]. First, say in one sentence each whether the bullets describe the same job done well or a bigger job than my title. Then give me one ask as a percentage and a dollar figure, show the arithmetic, and name the benchmark it rests on. Do not use any salary figure I did not give you.

Outside the US, replace the inflation line with your own country's latest consumer price figure; the merit and promotion numbers are US only. Mercer's survey also has employers planning to promote about 9% of staff in 2026, down from 10% in 2025, so the promotion case has to be sharper than last year.

Timing matters as much as the figure. Mercer fielded its survey in the last two weeks of October and WTW collected from September to November, which is when next year's pool gets set. A request made in February negotiates against a number locked months earlier, so ask before the autumn budget closes, or ask explicitly for an off cycle adjustment so your manager knows which pot you mean.

The email prompt writes the 150 word request and a 90 second spoken opening

One prompt produces both wrappers from the same bullets and number: an email that books the meeting and a spoken opening for the meeting itself. Ask for both and use both.

Using these three bullets [paste], write two versions of a raise request. Version A is a spoken opening for a one on one with my manager, 90 seconds when read aloud, under 200 words, in plain sentences I would actually say. Version B is an email to send before that meeting, under 150 words, that asks for 20 minutes to discuss compensation and previews the case in one line per bullet. Both versions state the number I am asking for, which is [amount or percentage], and the date I want it effective. No apologies, no mention of my personal expenses, no comparison to colleagues, and no phrase that sounds like a template.

Read Version A aloud once with a timer. If it runs past 90 seconds, the model padded it; cut the second sentence of each bullet, which is almost always the padding. Then give the email a 20 second edit for your own voice, the same pass AI for email writing recommends for any message that matters, because a manager who has read a hundred AI drafts can tell.

How do you practice the raise conversation with AI?

You practice by opening a fresh chat where the model plays your manager, pushes back with named objections, and grades each of your answers. Use a fresh conversation or a different model, so the manager is not the model that just wrote your case.

You are my manager, [name or role], at a [size] company in [industry]. It is [month]. I am about to ask you for a raise of [amount or percentage]. Be realistic and a little skeptical, not hostile. Push back with the objections a real manager uses: the merit budget is 3.2% this year, the cycle is already set, you need to see it sustained, others on the team are in the same position, and you cannot approve this alone. Interrupt me if I ramble. After each of my replies, say in one line what worked and what did not. Do not agree to anything until I have handled at least three objections. Start by asking me why now.

Paste Version A as your first turn and answer the pushback in your own words. The value is in the third objection, where the prepared script runs out and you find out what you believe about your own case. Three passes is enough. Save the objection that landed hardest and add one sentence to your opening that answers it before it is raised.

Left alone, a chat model playing a manager tends to agree by the second turn, which is why the prompt names the objections and forbids agreement before three. A free ChatGPT account runs this in a second chat; in Whizi you can switch models mid conversation so a different model plays the manager against the case the first one wrote.

The follow-up prompt turns "let me think about it" into a date

Send a follow-up email the same day that restates the number, the one objection you discussed, and the date you agreed to revisit. A maybe with no date drifts into next year's cycle.

I just had a raise conversation with my manager. I asked for [amount or percentage] effective [date]. Their response was [paste what they said, in their words as closely as you remember]. We agreed to [next step, or nothing]. Write a follow-up email under 100 words that thanks them, restates my number and the result they said would matter most, and proposes a specific date to close the decision. If no next step was agreed, propose one. Do not add anything I did not tell you was said.

If there was no agreed date, the email is where you ask for one: "Can we pick a day to close this?" is the sentence that turns a maybe into a decision.

AI invents metrics and pads openings unless the prompt forbids both

The two failures are predictable. A model asked for a strong case will supply a percentage you never measured, which is why the case prompt forces [NUMBER NEEDED]; and it pads, which is why the email prompt caps the spoken opening at 90 seconds. It also does not know your company's cycle, your manager's authority, or whether the team just lost headcount. Those go in the rehearsal prompt, or you rehearse against a manager who does not exist.

The number you finally ask for is a judgment about your own situation, and the benchmarks above set the floor for it rather than the answer. The reply to "no" in the room is yours too.