At a glance
Performance reviews are the highest-stakes recurring conversation most professionals have. The 7 prompts below cover the full cycle: self-review prep, peer-review draft, manager-to-IC review, upward feedback to your manager, calibration prep, the difficult-conversation script, and the follow-up. AI structures the evidence and rehearses the harder lines. You deliver the conversation. Nothing about the review gets ghost-written.
Why do most performance reviews go badly?
Two people sit down for the review. Neither has actually walked through the calendar from the past six months. The conversation drifts toward feelings rather than evidence. Recency bias kicks in (the last 6 weeks of work overshadow the prior 6 months). And the difficult parts get softened past the point of clarity, so the IC walks out feeling fine and the manager walks out feeling unheard. The prompts below address all three failure modes. They are designed to be run the night before, not in the room. Live AI use in a performance conversation is rarely a good idea; the structuring happens beforehand, the delivery happens in your voice.
One ground rule: every claim, every metric, every quote in the output must trace back to source material you actually have (1-on-1 notes, peer feedback, project artifacts, calendar history). Do not let AI invent accomplishments or weaknesses. The follow-up question that exposes a fabricated detail is the line that turns a useful review into a credibility hit.
⚠️ What never gets shared with AI
Other people's names, salary numbers, anything that might be in a personnel file, and any confidential HR information. Performance reviews live in the HR layer of trust; uploading a transcript of someone else's 360-feedback to a consumer AI service is a data-handling violation in many enterprise policies. Use anonymized summaries instead. For pre-send editing of anything that will be visible to the reviewee: How to Edit AI Out of Your Writing.
What are the seven review prompts?
The 7-prompt review cycle
- Self-review prep — build your own evidence pack
- Peer feedback draft — write the review of a teammate
- Manager-to-IC review — draft the conversation you owe a direct report
- Upward feedback — deliver the candid read of your manager
- Calibration prep — defend your rating to the calibration room
- The difficult conversation — underperformance, missed promotion, PIP
- Follow-up — what gets sent in writing after
1. Self-review prep
Here are the raw inputs about my work in this period (cut and paste, do not curate):
– Major projects + my role in each: [paste]
– Specific outcomes with numbers where I have them: [paste]
– Times I failed or fell short: [paste, be candid]
– Feedback I have received from manager / peers: [paste]
– Goals I set at the start of the period: [paste]
Build my self-review:
– 3-5 strongest accomplishments with evidence (cite the source line for each)
– 2-3 candid growth areas with what I am doing about each
– The single thing I want my manager to remember (the headline)
– The 1-2 questions I should ask in the review to learn something useful
For each accomplishment, mark whether I have evidence the manager has seen vs evidence I will need to surface.
Why this works: The single biggest mistake in self-reviews is assuming the manager remembers your wins. They have 5-8 ICs and many projects to track; recency bias is real on their side too. The “evidence the manager has seen vs needs to be surfaced” output tells you which wins to anchor on in the conversation. The “headline” output is what you say if the manager only remembers one thing about this review.
2. Peer feedback draft
Our working relationship in this period: [paste 2-3 sentences].
Specific moments I want to reference (anonymized, describing the situation, not the people): [paste].
What I think they should keep doing: [paste your raw take].
What I think they should adjust: [paste your raw take, even if uncomfortable].
Draft the peer feedback in this structure:
– 2-3 sentences naming a strength with a specific example
– 2-3 sentences naming a growth area with a specific example, framed as "here is what I observed and what I'd suggest"
– 1 sentence on the kind of work where I would want them on my team again
Constraints: no "great teammate" or "always positive attitude" cliches. Specific behaviors, specific moments. No comparative language (do not rank them against others). Length: 200-300 words.
Mark anywhere I've been too soft or too sharp. The reviewee will read this; tone matters.
Why this works: Peer feedback is the most-frequently-bot-sounding part of a review. Every “great teammate, always positive attitude” reads as form-filler. The specific-behavior, specific-moment structure is what makes peer feedback land. The “too soft / too sharp” output is the calibration check; AI is good at flagging where the prose has drifted from candid into either flattery or harshness.
3. Manager-to-IC review
The period covered: [date range].
My raw notes from our 1-on-1s and project work: [paste].
Their goals at start of period: [paste].
What I think they did well: [paste your raw take].
What I think they need to grow into: [paste your raw take, including anything you have hesitated to say out loud].
Draft the review:
– The single most important message they should walk away with (1 sentence)
– 3 strengths with specific evidence + impact
– 2 growth areas with specific examples + actionable next steps
– The promotion / rating recommendation with the 1-sentence reasoning
– The 3 questions I want to ask them in the conversation
Hard constraints: no "great work this year" opener. No softened language for growth areas (if they need to write better, say "write better" with the specific example, not "continue developing communication skills"). The reviewee deserves clarity.
Flag any growth area where I have not actually given them this feedback in the last 6 months. Surprise feedback in a review is a failure of management, not a feature of the review.
Why this works: The “flag any feedback I have not actually given before” output is the single most useful manager-side prompt output. Reviews that introduce new criticism are how managers blow up trust with strong performers. The “single most important message” output is what survives 6 months later; the rest of the review fades, but that one sentence stays.
4. Upward feedback (your manager review)
Our working relationship: [paste, including how often we meet, kinds of work they support me on, kinds of work they do not].
What I appreciate about how they work with me: [paste your raw take].
What I wish they would do differently: [paste your raw take, including the things you have been quietly frustrated about].
The context of the feedback channel: [skip-level review / anonymous 360 / direct conversation / written form].
Draft the feedback:
– Open with a specific thing I value (not flattery, but something they could not be sure I had noticed)
– Name one specific behavior I would want to see change, with the specific situation and the impact it had
– Suggest the smallest possible change rather than restructuring how they work
– Close with a question that opens dialogue, not one that closes it
Constraints: no "I'd love to see more support". No "maybe we could explore better communication." Specific request, specific situation, specific impact. 4-6 sentences.
Mark how risky this version is for me to send through [the chosen channel]. If high-risk, suggest a softer version that still preserves the core message.
Why this works: The “how risky to send through this channel” output is the political-judgment calibration most ICs cannot do alone. Anonymous 360 surveys allow sharper feedback; named feedback in front of HR demands more care. The “softer version that preserves the core message” output is the path forward when the candid version would cost more than it earns.
5. Calibration prep
My proposed rating: [paste].
The peer ratings on the same level: [list of how others on the same level got rated, anonymized].
The evidence I have: [paste from prompt 3].
Prep the defense:
– The 30-second elevator pitch for why this rating is correct
– The 2-3 most likely counter-arguments from the calibration room, and how I would respond to each
– The piece of evidence that is most likely to be challenged, and what I will say if it is
– The compromise position if the room pushes back (one rating tier higher or lower) and what the cost of accepting that compromise is
– The line I will not cross (if X happens, I push back hard)
Mark where my evidence is thinner than I want it to be. If I lose the calibration, what should I do differently next period to make this case stronger?
Why this works: Calibration meetings reward managers who can articulate the rating in 30 seconds with evidence. ICs lose ratings because their manager could not defend them, not because the work was bad. The “line I will not cross” output is the political backbone most calibration discussions lack; knowing it ahead of time is what separates managers who advocate from managers who fold.
6. The difficult conversation
– Missed promotion despite expectations
– Underperformance review
– Pre-PIP warning conversation
– Behavior issue (peer complaints, missed commitments, conduct)
The specific facts: [paste].
What the reviewee thinks they did this period: [paste, based on your 1-on-1 conversations or their self-review].
The gap between their read and yours: [paste].
Draft the conversation script (not a monologue):
– Opening sentence that names the situation directly (no "I want to talk about your year" opener; name what we are here to discuss)
– The specific evidence I will share, in order
– The 3 likely emotional responses (defensiveness / shut-down / agreement) and what I will say to each
– The 1-2 things I am explicitly NOT asking them to defend in this conversation (so they do not have to fight on all fronts)
– The concrete next step with a date
– The line that closes the conversation when I have said what I came to say
Mark the parts where my evidence is too thin to lead with vs where it is solid enough to anchor on.
Why this works: Difficult conversations fail when the manager rehearses the monologue rather than the dialogue. The “3 likely emotional responses + what I will say” output is the muscle memory that lets the manager hold the conversation when it gets hard. The “things I am NOT asking them to defend” output is the gracious move that distinguishes a hard conversation from a hostile one; you do not need to indict everything to deliver the one specific message.
7. The follow-up
The key things we agreed on: [paste].
The things still open: [paste].
The emotional tone of the conversation (calm, charged, neutral): [paste].
Draft the follow-up note:
– Open by referencing the specific moment that mattered in the conversation (not "thanks for the chat")
– Restate the things we agreed on (so they are on record)
– Name the open items with a date for each
– Close with an invitation to continue the conversation if anything I said landed wrong
– 4-6 sentences, professional warm, no exclamation marks
Hard constraint: do NOT re-litigate any disagreement in writing. If we did not align on something, name it as "here is something we should keep discussing" rather than restating the case in the note.
Why this works: The “do not re-litigate in writing” rule is the most-violated rule in post-review follow-up. Writing your case again, in writing, after the conversation ended, escalates rather than closes. The “name it as something to keep discussing” framing is the path that preserves the relationship while keeping the disagreement visible.
What is the worst thing you can do with AI on a review?
- Upload personnel data to a consumer AI service. The HR-data trust layer is real. Anonymize ruthlessly. Strip names, salary numbers, and anything that could identify the person across other sources.
- Let AI invent accomplishments or weaknesses. Every claim must trace to source material. The follow-up question that exposes a fabrication is the line that costs you credibility for years.
- Read AI-generated review prose verbatim in the conversation. The reviewee hears the scripted cadence. Use AI for structure; deliver in your own words.
- Surprise the reviewee with growth-area feedback they have never heard. If it is in the review and you have not said it in a 1-on-1, you have failed as a manager. Prompt 3 flags this; respect the flag.
- Use AI for sympathy reviews after a hard period. Bereavement, illness, family crisis, layoff aftermath. AI involvement here loses everything; write the review yourself, in your voice.
What if you do this every cycle?
If you manage a team and run reviews twice a year for 5-8 ICs, the 7 prompts are a candidate for the ladder. Save each as a Claude skill. Bundle as a plugin called review-cycle that runs the prep for each direct report from your accumulated 1-on-1 notes and project history. The setup is one quarter of investment; the time saved across two cycles is a full week of manager time per year, and the quality lift on the actual reviews is what compounds: ICs feel heard, calibration goes faster, the difficult conversations land cleaner.
📊 The Prompt-to-Workflow Ladder
Tier 1: the prompts (this post). Tier 2: the skill (one per review stage). Tier 3: the plugin (review-cycle bundle). Tier 4: the workflow (prep packs auto-generate from 1-on-1 notes on a schedule). When to climb →
What are common questions about AI in performance reviews?
Will my HR team know if I used AI?
If you read AI-generated prose verbatim in the review form, yes. The cadence is recognizable to anyone who reads many reviews. HR teams that review for promotion calibration are now spot-checking for AI patterns. The fix is the editing pass: structure from AI, words from you.
Should I disclose I used AI to prepare?
Unprompted, no. If asked, yes (framed as “AI helped me organize my thinking; the conversation and the assessments are mine”). Most managers do not care that you used AI to prepare. They care that the review is accurate, specific, and delivered in good faith.
What about peer feedback in 360 surveys?
Same rules. The anonymization is more important here; do not upload identifying information. The editing pass is more important too; 360 feedback in templated AI prose is the most common signal of low-effort participation, and managers reading the survey results notice.
What if my manager is using AI for my review?
Many managers are. The review you receive may be partly AI-drafted. The signal is the same as any other AI prose: cadence, generality, the absence of specific moments. If your review reads as templated, ask in the conversation: “Can you share the specific moment from the past 6 months that anchors this feedback?” The good manager has it ready; the inattentive one struggles to produce it. Either way, you learn something.
Where do these prompts come from?
They are the management section of the larger AI Prompt Library. The Library has over 500 prompts across 33+ categories, including the full management stack: 1-on-1s, hiring screens, onboarding plans, performance reviews, calibration, PIP design, and the harder conversations like delivering a layoff or telling someone they did not get the promotion they expected.
Sources to read next?
- Harvard Business Review on performance management
- SHRM performance management research
- McKinsey on people and organizational performance
- Radical Candor (Kim Scott) feedback framework
- Performance appraisal (Grokipedia)
✏️ Before the review form ships
A review written in AI cadence reads as low-effort to the reviewee. Strip the tells before submitting: How to Edit AI Out of Your Writing → The full 29-pattern catalog from Wikipedia's “Signs of AI writing” guide is documented there for any written review submission the reviewee will see.
The AI Prompt Library · $39
over 500 tested prompts including the full management stack
The seven review prompts above are a free preview. The full Library has the management stack: 1-on-1 templates, hiring screen rubrics, onboarding plans, calibration prep, PIP design, layoff scripts, and the harder cases like telling someone they did not get the promotion they expected.
1-on-1 Deep Work Session with James · $175
Build your review-cycle workflow in 2 hours
A focused 2-hour session. We run prompts 1, 3, and 5 on a real upcoming review (anonymized), save them as skills, and you walk out with a reusable system your future-self will thank you for. Worth especially if you manage 4+ reports.
Get Smarter About AI Every Morning
Free daily newsletter. Built for people who want to use AI well, not chase every model.
Free forever. Unsubscribe anytime.
You may also like
- Best Claude Prompts for Sales — cold outreach + objection scripts
- Best AI Prompts for Email Writing — the everyday inbox workhorse
- Best AI Prompts for Customer Support — the queue without burning out the team
- How to Edit AI Out of Your Writing — strip AI patterns before you publish
- Prompt to Workflow: The AI Ladder — when to climb from prompt to workflow