The task
It is two weeks before promo packets are due. You have a Drive folder of design docs, launch posts, screenshotted dashboards, and three Slack threads where someone said “great work.” You also have the rubric for the next level, which you’ve read four times and can mostly recite. What you do not have is the 1-2 page narrative that turns “here is everything I did” into “here is the case that this person is operating at the next level.” The hard part is not finding evidence. It is choosing the 5 things that build a coherent scope-and-impact arc, and dropping the 12 things that quietly weaken it.
Where AI helps — and where it does not
AI is strong at restructuring evidence into impact narratives, mapping work to leveling rubrics, compressing activity verbs into outcome sentences, and tightening prose without losing your numbers. It can also catch the “task log” failure mode and reorder paragraphs so scope leads instead of trailing.
What AI cannot do: invent metrics, judge whether your actual scope is “next level” or one cycle short, or know your company’s unwritten promo bar (some companies promote on consistent meets-expectations; others demand visible stretch behavior). It also cannot tell you which of your projects calibration committees will see as a level-up signal vs. a competent-execution signal; that is a conversation with your manager.
A specific failure mode: AI tends to give equal real estate to every bullet you paste, including the small ones. Tell it explicitly which 3 items carry the case, and that the remaining bullets should be one-line supporting evidence at most.
What to feed the AI
- Your current level and target level, each with a one-line rubric summary
- The 3 anchor items that carry your case, each with scope (people / dollars / surface), one outcome metric, and one “what was hard about this”
- 3-5 supporting items in bullet form, one metric each, no narrative
- Cross-functional partner quotes (paraphrased is fine, attribute by role not name)
- Last cycle’s review + the one piece of feedback your manager called out for this cycle
- One example of someone at the target level whose work yours most resembles (this anchors “what next-level looks like” without you having to define it)
- The single sentence you want the calibration committee to remember about you
- Any gap you are already addressing, with what you have done so far (do not pretend gaps don’t exist; calibration spots them)
Copy-ready prompt
You are helping me draft a promotion self-review from {current_level} to {target_level}.
Target-level rubric (one-line per dimension): {paste}
My 3 anchor items: {scope + outcome + what was hard, for each}
Supporting items: {3-5 bullets with metrics}
XFN partner observations: {paraphrased, by role}
Last cycle's review + the feedback I was told to act on this cycle: {paste}
A person at target level whose work mine most resembles: {name + what about their work}
The one sentence calibration should remember: {sentence}
A gap I am closing + what I have done so far: {paste, or "none worth surfacing"}
Write a 600-word self-review with sections:
1) Scope — open with the surface, headcount, and dollars or users I influenced. One sentence on how this is different from last cycle.
2) Impact — the 3 anchor items. Lead each with outcome, not activity. Drop the verb if the outcome already implies it.
3) Leadership — observable behaviors: who I unblocked, what decisions I owned, one thing I changed about how the team works.
4) Growth — name the gap, name what I have already done about it, name the next step. Do not hide it.
Voice: past tense, verb-first, third-person evidence ("the team," "users"), first-person only for ownership claims. No filler adjectives — every "great" or "significant" gets replaced with the number or the artifact.
End with one closing sentence the committee can quote back.Shorter variant — calibration paragraph only
Write the 100-word calibration paragraph for my promo case from {current_level} to {target_level}.
Strongest evidence: {one anchor item, scope + outcome}.
Strongest behavior signal at target level: {one observed behavior}.
One gap closing in flight: {paste}.
Format: 1 sentence on level + scope. 2 sentences on the strongest impact and what made it next-level. 1 sentence forward-looking on the gap. No filler.Sample output
A strong scope opener: “I owned the checkout funnel through this cycle — 3 PMs, 6 engineers, ~$4M ARR contribution, up from one workstream and 2 engineers last cycle. The expansion was not handed to me; I scoped the next two surfaces (refunds, B2B billing) and pulled them in after demonstrating the funnel cut would land.”
A strong impact bullet: “Reduced step-2 drop-off from 18% to 11%, recovering $610K annualized. The hard part was that the obvious fix (form simplification) had been tried twice; the actual signal was that mobile users were tapping out on a Stripe redirect we had grown blind to. Caught it by running my own session replays for a week.”
A strong growth paragraph: “I came into this cycle weak on pricing strategy — last cycle’s feedback was explicit. I paired with finance for the Q2 SKU rework, owned the willingness-to-pay survey, and shipped the new pricing page with measured uplift of 7% on average revenue per signup. The next layer is owning the pricing committee, not just the executions; I am shadowing for Q3.”
How to refine
- Compress activity into outcomes: “Drop the verb if the outcome already implies it. ‘Cut drop-off from 18% to 11%’ is stronger than ‘I worked with the team to cut drop-off from 18% to 11%.’”
- Lead each section with the rubric criterion it answers: “Open each section with one sentence that names the target-level rubric dimension this is evidence for. The committee should not have to map paragraphs to rubric in their head.”
- Pre-empt the calibration question: “After Impact, add one sentence answering: ‘why is this next-level work and not just well-executed current-level work?’ That sentence is the one the committee quotes.”
- Force third-person evidence: “Use ‘the team,’ ‘users,’ ‘finance partner’ for impact claims. Reserve first-person (‘I scoped,’ ‘I caught’) for ownership decisions only. This is the line between confidence and bragging.”
- Cut to 600: “Cut to 600 words. Whatever survives is the case. If a section disappears, the missing dimension is your real gap and you should know that going into calibration.”
Common mistakes
- Listing every project with equal weight: committees discount unanchored bullets; pick 3 anchors and let the rest be supporting
- No metrics, or vanity metrics with no business tie-in: “users engaged” without revenue or retention link reads as decoration
- Ignoring the rubric entirely: every paragraph should map to a rubric dimension; if none does, the work is at the current level
- Hiding the known gap: calibration spots it anyway and trust drops; name it, show your fix
- Writing the same length per anchor as per supporting item: anchors get specificity, supporting items get one line each
- Letting AI invent partner quotes: paraphrase real observations from real partners; fabricated quotes get caught and the packet dies
- First-person verbs everywhere: “I led,” “I drove,” “I owned” stacks read like bragging; use third-person evidence for impact
- Closing on activity: the closing sentence is the one calibration repeats; make it about scope at the target level, not what you did
FAQ
- Will my manager notice it is AI-written?: Only if you skip the personalization step. Feed the model your real metrics and paraphrased partner observations, then rewrite the opening line and the closing sentence in your own voice. Those two sentences are what calibration quotes back.
- How long should it be?: Match your company’s template. If none: 600-900 words for IC promotions, 1000-1500 for senior IC, manager, or staff. Length past that signal-degrades; anything beyond the template gets skimmed.
- The draft sounds like bragging. What changes?: Add: “Replace every first-person ownership verb with third-person evidence, unless the sentence is naming a decision I owned. Bragging is first-person for impact; ownership is first-person for choices.”
- What if I do not have an anchor item that is unambiguously next-level?: Then the packet should not be filed this cycle. Self-reviews can move a borderline case; they cannot move a missing case. Use the draft to spot the gap and target the next cycle.
- Performance review vs promotion packet, same doc?: No. A review answers “did they meet level.” A promo packet argues “are they operating at the next level.” Different evidence weight, different verbs, often different anchor items.