Use AI to write a performance review: 6 prompts and a sample review

The prompt takes the goals, your own evidence and the rating scale, and you keep the rating

One prompt writes a usable first draft of a performance review when it carries three things: the goals set at the start of the cycle, evidence you pulled yourself from the ticket system, the calendar and the numbers, and the rating scale with its definitions. JPMorgan Chase allows its managers exactly that, under a guideline that the tool is "not a substitute for human judgment" and a ban on using AI to assign scores or decide pay and promotions (Bloomberg, 13 November 2025). Here is the prompt we sent to two models on 29 September 2026; the support engineer and her numbers are invented, the shape is the point.

Write the manager's comments for an annual performance review. Use only the evidence below; where you need a fact I have not given, write [NEEDED] instead of inventing it. Structure: one paragraph per goal (outcome first, then the evidence), one paragraph on areas to develop, then a proposed overall rating on our scale with one sentence of justification. Under 250 words. No adjectives about the person's character, only what they did and what it produced.

Role: support engineer, 2 years in role. Rating scale: 1 Below expectations, 2 Partially meets, 3 Meets, 4 Exceeds, 5 Far exceeds. A 3 is the expected rating for a solid year; 4 needs measurable impact beyond the role.

Goal 1: Cut median first-response time on tier 2 tickets from 9 hours to under 4. Evidence: Zendesk report, median first response 3.6 hours for Q2 to Q4; 3,100 tickets handled in the year against a team average of 2,400.

Goal 2: Ship the customer-facing status page. Evidence: shipped 14 March, six weeks after the 1 February target; the delay was a vendor SSO dependency outside her control; 22 incidents posted to it since launch.

Goal 3: Mentor two new hires to unsupervised ticket handling within 8 weeks. Evidence: both new hires handling tickets alone by week 6 (manager's one-on-one notes, April and May).

Development note: two escalations in Q3 where the customer was told a fix date before engineering had confirmed it (Jira SUP-2211, SUP-2287).

Two instructions in that prompt do most of the work. "[NEEDED] instead of inventing it" stops the model decorating a goal with a number you never measured: in our run, Claude Sonnet 4.5 wrote one [NEEDED] where we had given no evidence of a fix after the two escalations, and Gemini 2.5 Pro wrote none and added a sentence of process we never supplied. "No adjectives about the person's character" keeps the draft on what happened, which is also what the employee can argue with and what HR can defend.

The evidence has to come from you, because the model has nothing else. Lattice's 2025 State of People Strategy Report found 49% of managers find it hard to review a year of feedback, and that year is exactly what the prompt needs: the Zendesk or Jira export, the calendar, the commit log, the one-on-one notes, the numbers on the dashboard. Prompt 2 turns those exports into evidence lines before anything gets written.

Below are raw exports covering this person's review period: [paste the ticket list, the calendar export, the commit log or the sales report, one after another]. Turn them into evidence lines, one per line, each naming a date or date range, a thing that was done and a number from the export. Group them under these goals: [paste the goals]. Where an item supports no goal, put it under "Other". Where a number is missing, write [NUMBER NEEDED]. Do not summarize or rate anything, and do not add any item that is not in the exports.

You are not early to this. Resume Builder surveyed 1,342 US managers with direct reports in June 2025: 65% use AI tools at work, and among those, 91% use them to assess performance, with ChatGPT the first choice for 53%.

Use AI to write a self review: same inputs, one honest miss

A self review written with AI starts from the same three inputs, and it argues for one rating, the one your evidence supports, with one thing you would do differently named in plain words. A self assessment with no miss reads as written by a machine or by someone not paying attention, and 26% of employees already suspect they have received an AI-written review (ZeroBounce, survey of 1,000 US professionals, September 2025). Citi's order of rollout says how normal this has become: from June 2025 its employees could draft self-evaluations with the company's AI tools, and managers got the same help four months later.

Write my self assessment for the [period] review. My goals were: [paste]. My evidence, all from my own records: [paste the evidence lines from prompt 2]. Our rating scale: [paste, with the definition of each level]. First person, under 300 words, one paragraph per goal that leads with the result and then the evidence, then one paragraph naming the thing I would do differently: [describe the miss in one line]. Argue for the rating my evidence supports and say which level it is; do not argue for a higher one. Where you need a number I have not given, write [NUMBER NEEDED]. No claims about my attitude, effort or potential, only what I did and what it produced.

Every number in the self review should be one your manager can check in the system you took it from, because that is where they will look. The three results that carry a self review are the three results that carry a raise request, so the case prompt on use AI to ask for a raise produces the bullets for both, and the self review is the better place to put them on the record first.

The last pass is voice, and it matters more here than in the manager's draft, because a self review that reads like a model tells your manager you did not write it. Prompt 4 uses your own sentences as the sample, and it works on any work email the same way.

Here are three paragraphs I wrote myself, from emails or documents: [paste]. Rewrite the self review below so it matches how I write: my sentence length, my vocabulary, the way I start sentences. Remove every adjective that is not attached to a number. Do not add, remove or reorder any fact, and keep every number and every [NUMBER NEEDED] exactly as it is. [paste the draft]

Which AI is best for performance reviews? We ran one prompt on two

Claude Sonnet 4.5 wrote the more careful review and Gemini 2.5 Pro the warmer one, and both proposed the same rating, a 4, from identical evidence. We sent prompt 1 unchanged to both models through the same API (fal.ai's any-llm endpoint) on 29 September 2026 at temperature 0.3, one run each, with Gemini's reasoning switched on because the endpoint requires it for that model. The whole test cost under a cent.

Claude's output, unedited apart from its bold markup:

Goal 1: First-Response Time The median first-response time target was met, with Zendesk data showing 3.6 hours across Q2-Q4, below the 4-hour goal. This was achieved while handling 3,100 tickets, 29% above the team average of 2,400.

Goal 2: Status Page The customer-facing status page shipped on 14 March, six weeks later than the 1 February target due to a vendor SSO dependency outside her control. Since launch, 22 incidents have been posted to the page.

Goal 3: New Hire Mentoring Both assigned new hires reached unsupervised ticket handling by week 6, two weeks ahead of the 8-week target, as documented in manager one-on-one notes from April and May.

Development Areas Two escalations occurred in Q3 (SUP-2211, SUP-2287) where fix dates were communicated to customers before engineering confirmation. [NEEDED: evidence of improvement or corrective action taken after these incidents.]

Proposed Rating: 4 (Exceeds) The combination of meeting the response-time goal while handling significantly above-average ticket volume, accelerating new hire readiness, and delivering the status page despite external delays demonstrates measurable impact beyond standard role expectations.

Gemini's, unedited:

You exceeded the goal to reduce median first-response time, achieving 3.6 hours against a target of under 4 hours from Q2 to Q4. This was accomplished while handling 3,100 tickets for the year, above the team average of 2,400.

The customer-facing status page was shipped on 14 March. While this was after the 1 February target, the delay was due to a vendor dependency outside your control. The page has been used to post 22 incidents since launch.

You met the goal to mentor two new hires. Both were handling tickets unsupervised by week 6, ahead of the 8-week target, according to one-on-one notes from April and May.

For development, focus on communication during escalations. In Q3, two customers were given a fix date before engineering had confirmed it (SUP-2211, SUP-2287). The process is to wait for an engineering estimate before providing a date to the customer.

Proposed rating: 4 (Exceeds). The significant outperformance on ticket volume and response time created a measurable positive impact on the customer experience beyond the defined expectations for the role.

The rating is where both leaned the same way, and it is the reason the rating stays with you. The scale said a 3 is the expected rating for a solid year and a 4 needs measurable impact beyond the role. Both models took 3,100 tickets against a 2,400 average as that impact, and neither weighed the six-week slip on goal 2 or the two escalations against it in the rating sentence. On this evidence a 4 is arguable; so is a 3 with a line about the escalations, and the model picked the generous one both times. Prompt 5 makes it argue the other side before you decide.

Here is a draft review and our rating scale: [paste both]. The draft proposes a [rating]. Write the strongest one-paragraph case for the rating one level below it and the strongest case for the rating one level above it, using only the evidence in the draft and the scale's own definitions. Then list every piece of evidence the proposed rating ignores. Do not recommend a rating.

So, which AI is best: for a review the model matters less than the evidence, since two frontier models given the same evidence produced the same rating and differed on voice and one invented sentence. Claude if you want the missing evidence flagged, Gemini if you want a draft already addressed to the employee. Where having both in one place helps is a review season with eight reports. On Whizi you pin the rating scale and the competency definitions to a project once (as of September 2026 a project holds up to 10 pinned files of 10 MB each, .docx and PDF included, and project instructions up to 32,000 characters that ride along with every message), then run each report's evidence through Claude and Gemini without pasting the scale again (how projects work). Where Whizi loses is the other half of the job: it has no HRIS integration and no team accounts, so nothing is pulled from Workday, Lattice or Zendesk for you. Citi's Performance Assist "automatically gathers all of that information directly from our systems", in the words of Patricia Gould, Citi's head of total rewards, and Lattice's AI Agent summarizes past feedback from inside Lattice. On Whizi you export and paste.

Do you have to tell HR, or the employee, that AI helped write the review?

At the two large employers with a published position, the answer differs: Citi tells employees and lets any of them opt out of an AI-drafted review, and fewer than 1% had done so by December 2025, while JPMorgan's guideline, as Bloomberg reported it, limits what the manager may do with the tool rather than requiring a disclosure to the report. JPMorgan forbids AI-assigned scores; Citi keeps the manager responsible for the final evaluation. Everywhere else the answer is your company's AI policy, and Lattice's and Textio's first checklist item is the same as ours: read it before you paste. The named US employer positions, as reported between 13 November and 2 December 2025:

The survey numbers say the question is already live in most teams. In Resume Builder's June 2025 survey, "More than 20% say they allow AI to make decisions without human input either all the time (5%) or often (16%)", and only 32% of the managers using AI to manage people had any formal training in it. ZeroBounce's September 2025 survey put 41% of managers as having drafted or revised a review with AI. Gartner's December 2025 survey of 1,622 respondents found "managers reported an average of four hours saved across different parts of the performance management process when using AI", and its April 2026 advice was to treat AI "as an input to managerial judgment rather than a replacement for it".

The one field experiment on disclosure found a cost to telling people, and a way round it. Tong, Jia, Luo and Fang, in the Strategic Management Journal in 2021, measured AI-generated feedback inside a real firm and wrote that "employees may develop a negative perception of AI feedback once it is disclosed to them, thus harming their productivity", an effect that shrank with the employee's tenure. Their suggested fix is the shape every policy above lands on: the manager delivers the feedback, in the manager's voice, whatever drafted it. Peter Cassat, a partner at CM Law, told HR Dive in December 2025 that using AI to evaluate employees "could subject organizations to discrimination claims as well as substantial compliance obligations", which is the legal reason scoring stays human at the banks. State and city rules differ, and this page is not legal advice; the question for your HR or counsel is whether a drafting tool counts as an automated decision tool where you are. AI for HR lists six US AI employment rules, the ones that reach pay and discipline, and the ones that carve drafting out in their own text.

Can I write a review using AI for free?

Yes. The free tiers of ChatGPT, Claude and Gemini take prompt 1 as written, and the dedicated generators do the same job with a form instead of a prompt. TechnologyAdvice tested eight free generators in March 2026 and kept four: Easy-Peasy.AI (2 reviews a day), Writify.AI (no limit, with ads), Vondy (4 before sign-up) and GravityWrite (1,000 words a month). The forms ask for the employee's name; put a job title there instead, on any tool whose data terms you have not read (what never to paste into an AI chat).

Free or paid, run one check the form tools skip. Textio's January 2023 test of ChatGPT feedback found it presumed a nurse female 9 times out of 10 and wrote feedback for women about 15% longer, with the extra length mostly criticism. Prompt 6 runs that check across your own set before anything is submitted.

Here are [N] draft reviews from my team, with names replaced by letters: [paste]. Compare them and list: any review more than 20% longer or shorter than the median; any adjective about character (reliable, difficult, bubbly, aggressive) and which review it appears in; any pronoun or gendered word; any claim with no number or date behind it. Output a table with one row per finding. Do not rewrite anything.

The rule every policy on this page agrees on fits in a sentence: the evidence comes from you, the draft can come from the model, and the rating and the conversation stay yours.