How to Write Your Self-Review With AI (Prompts Included)

Ellie at Willa, Editorial lead, Willa · Published September 14, 2026 · Updated September 14, 2026

Performance review season is here, and the self-review is the part most of us dread. You sit down to write about your own year and suddenly every project feels smaller than it was. You hedge. You write "helped with" when you mean "led." Then you submit something that undersells you to the exact people who decide your rating, your raise, and your next promotion.

AI can fix most of this, and not in the "paste your job title into ChatGPT and pray" way. Used well, Claude or ChatGPT becomes the colleague who interviews you about your year, catches you underselling, and turns a messy pile of wins into clear, confident paragraphs that still sound like you. Here is the full workflow, with prompts you can copy today.

Why is writing a self-review so hard?

Because evaluating your own work and communicating it are two different skills, and the second one is where most of us fall down. Research backs this up. A study by economists Christine Exley and Judd Kessler found that women describe their ability and performance less favorably than equally performing men, even when there is no strategic reason to hold back. The gap showed up as early as sixth grade.

The feedback you receive is often skewed too. Textio's research on performance feedback found that women receive 22% more feedback about their personality and 30% more exaggerated, non-actionable feedback than men. If your manager's notes say "great team player" while your male colleague's say "shipped the migration two weeks early," your self-review is your chance to put the specifics on the record yourself.

This is why AI helps so much here. The model has no anxiety about your accomplishments. It will state plainly what you did, and you can edit from there. Editing a draft that is too confident is far easier than writing one from a blank page.

How do I use AI to write my self-review?

Work in five steps. Budget about 90 minutes total, ideally split over two days so you can edit with fresh eyes.

1. Gather your raw material first. Before any prompting, collect evidence: sent emails announcing things you shipped, project docs you owned, praise messages, closed tickets, dashboards you built, meetings you ran. Search your own name in Slack or email for the year. Ten messy bullet points beat a perfect memory.

2. Have the AI interview you. Do not ask it to write anything yet. Ask it to ask you questions, one at a time. Talking through your year surfaces wins you forgot, and answering questions is easier than staring at a form.

3. Map wins to the review criteria. Paste in your company's review questions or competencies (reworded if they are confidential) and ask the AI to sort your evidence under each one, flagging gaps.

4. Draft, then de-hedge. Ask for a first draft in first person, then run a dedicated pass where the AI hunts for hedging language: "helped," "assisted," "was lucky to," "we" where "I" is accurate.

5. Rewrite it in your voice. AI drafts sound generic by default. Give the model two or three samples of your own writing and ask it to match your style. We wrote a whole guide on this: How to Make AI Write Like You, Not a Robot.

What prompts should I use for a self-evaluation?

Copy these and adjust the details. They work in Claude, ChatGPT, or Gemini.

The interview prompt:

You are helping me prepare my self-review for my performance evaluation. I am a [role] and my review covers [period]. Interview me about my year, one question at a time. Ask about projects I led, problems I solved, people I helped, skills I grew, and anything I did that was outside my job description. Push for specifics: numbers, timelines, and outcomes. Keep going until we have at least 10 concrete accomplishments.

The criteria-mapping prompt:

Here are my review criteria: [paste criteria]. Here are my accomplishments from this year: [paste your list]. Map each accomplishment to the criteria it best supports. Then tell me which criteria have weak or missing evidence so I know where to add more before I draft.

The drafting prompt:

Draft my self-review using the mapped accomplishments above. Write in first person, plain language, and a confident but factual tone. Lead each section with the outcome, then explain what I specifically did. No buzzwords, no filler. Keep each section under 150 words.

The de-hedging prompt:

Review this draft and flag every place I undersell my work: hedging words, passive voice, credit spread to "we" when the work was mine, and accomplishments described without their impact. Suggest a stronger, still accurate rewrite for each flagged line.

The tough-critic prompt:

Act as a skeptical review committee member reading this self-review. What claims feel vague or unsupported? What questions would you ask me? Where would you want a number or example? List the weak points so I can fix them.

How do I quantify accomplishments without obvious numbers?

Not every role produces clean revenue metrics, but almost every accomplishment has a shape you can measure. Ask the AI: "Here is what I did: [describe it]. Help me find the measurable angle." Then look in three places: scale (how many people, projects, dollars, or customers were touched), time (what got faster, what deadline was hit, what you shipped early), and risk (what fire you prevented, what error rate dropped, what stopped breaking).

"Improved onboarding documentation" becomes "Rewrote onboarding docs used by 14 new hires; new team members now ship their first project in week 2 instead of week 4." Same work, different weight. If you truly cannot get a number, name the stakeholder impact instead: who depended on this, and what did it let them do?

Keep a running brag document after this cycle so next season takes 20 minutes instead of a weekend. A note titled "Wins" plus a five-minute Friday habit is enough, and it doubles as raw material when you negotiate salary or a promotion later.

What should I avoid when using ChatGPT or Claude for a review?

Three mistakes come up constantly.

First, pasting confidential information into a personal AI account. Customer names, revenue figures, unreleased plans: leave them out or use your company's approved AI tool. You can anonymize ("a major enterprise client") without losing the story.

Second, submitting the AI's first draft. Reviewers are reading a stack of these, and generic AI phrasing is easy to spot and easy to discount. Your examples and voice are the credibility. The draft is scaffolding, not the finished building.

Third, letting the AI inflate you into fiction. Confident and false are different things, and one fabricated metric can taint every true sentence around it. Check every number against your evidence pile before you hit submit. The tough-critic prompt above exists exactly for this.

Key takeaways

• Collect evidence before you prompt: emails, praise, tickets, and docs from the whole review period, not only the recent quarter your memory favors.
• Use AI as an interviewer and editor, not a ghostwriter. The interview surfaces forgotten wins; the de-hedging pass fixes underselling that research shows is common and costly.
• Lead every accomplishment with the outcome, and find the measurable angle through scale, time, or risk.
• Never paste confidential data into a personal AI account, and verify every number in the final draft yourself.
• Start a running brag document now so your next review, and your next negotiation, starts from a full page instead of a blank one.

Writing about your own work gets easier when you are around people doing the same thing out loud. Willa members share review-season prompts, trade drafts for feedback, and practice talking about their wins at our live events. If that sounds like your kind of room, come join us.

• performance-review
• self-review
• prompting
• ai-at-work
• career-development

Join Willa, the community where women learn AI together.

All posts