Steal My Prompt Vol. 33: The Self-Performance Review

You open the performance review template. The cursor blinks in the box that says "Key Accomplishments." Three minutes later, you have written "led cross-functional initiatives." You will revise that line twelve times before submitting. None of those revisions will change the fundamental problem.

Most professionals undersell themselves in their performance review by writing about effort instead of outcomes. The fix is not better writing. It is better evidence. You wrote about effort because you cannot remember the outcomes, and you cannot remember the outcomes because they happened four months ago and your inbox has buried them.

I built this prompt to fix both problems in one workflow. The model challenges every weak claim and refuses to accept "led" or "supported" or "drove" without the proof underneath. The output reads like a thoughtful manager wrote it about you, not like you wrote it about yourself.

This week's AI Education section walks through Copilot's Researcher agent. The Power-User Workflow at the bottom of this post shows exactly how to point Researcher at your Outlook, Teams, and document library to surface evidence you forgot you had. The agentic layer is at its best when it is doing work you cannot do by memory.

What You Can Use This For

• The annual or mid-year performance review at work where the document is the only thing leadership reads about you
• Promotion conversations where you need to articulate the case for advancement with evidence, not effort
• A 1:1 with your manager when you need to advocate for yourself without sounding self-promotional
• Clinical or operational leadership roles where multi-month initiatives have impact that fades from memory before it lands on the page
• Job interviews where you need to translate scope and impact into specific outcomes a hiring panel can evaluate
• Year-end self-reflection where you want an honest, evidence-anchored record of your own work

How to Use It

1. Open Microsoft Copilot, Claude, ChatGPT, or Gemini.
2. Turn on the reasoning model. Think Deeper in Copilot. Extended thinking in Claude. The reasoning model in ChatGPT. Deep Think in Gemini. Reasoning catches more weak claims on the first pass.
3. If you use Copilot at a company that has Researcher, jump to the Power-User Workflow below first to compile your evidence base. Otherwise paste the prompt and fill in the inputs directly.
4. Read the model's challenges. Do not soften your inputs in response. The challenges are surfacing weak language, not asking you to retreat.
5. Read the final review out loud. If a sentence does not sound like something a thoughtful manager would write about you, rewrite it.

The Prompt

You are my Self-Performance Review Editor. I will fill in the inputs below. Your job is to challenge weak language, refuse effort framing, demand outcome evidence, and produce a final self-review that reads like a thoughtful manager wrote it about me.

REVIEW PROTOCOL:

- Reject any "led," "drove," "supported," "championed," or "spearheaded" that is not followed by a specific outcome with numbers, dates, or named results

- Reject generic growth areas like "communication," "executive presence," or "stakeholder management" without specific examples

- Force me to distinguish between effort (what I did) and impact (what changed because I did it)

- Ask ONE consolidated follow-up message after reviewing all my inputs

- After I respond, produce the FINAL SELF-PERFORMANCE REVIEW in the exact format below

- Do not pad. Do not flatter. Do not add accomplishments I did not include.

MY INPUTS:

ROLE: [TITLE, LEVEL, FUNCTION, TENURE IN THIS ROLE]

REVIEW PERIOD: [START DATE TO END DATE — TYPICALLY 6 OR 12 MONTHS]

ROLE EXPECTATIONS: [WHAT WAS EXPECTED OF YOU IN THIS PERIOD — GOALS, SCOPE, KEY OUTCOMES]

ACCOMPLISHMENTS: [LIST EVERY SPECIFIC THING YOU DID WITH THE OUTCOME — INCLUDE NUMBERS, DATES, NAMED PROJECTS, AND METRICS WHERE POSSIBLE]

STRETCH CONTRIBUTIONS: [WHERE YOU OPERATED ABOVE YOUR LEVEL OR OUTSIDE YOUR FORMAL SCOPE]

WHAT GOT IN THE WAY: [HONEST BLOCKERS — RESOURCING, PRIORITIES, LEADERSHIP CHANGES, WHATEVER WAS REAL]

GROWTH AREAS: [WHERE YOU KNOW YOU NEED TO IMPROVE — BE SPECIFIC]

CAREER DIRECTION: [WHERE YOU ARE TRYING TO GO IN THE NEXT 12-24 MONTHS]

AUDIENCE: [WHO READS THIS REVIEW — DIRECT MANAGER, SKIP-LEVEL, HR, PROMOTION COMMITTEE]

---

AFTER MY RESPONSE TO YOUR FOLLOW-UP, OUTPUT THE FINAL REVIEW:

# SELF-PERFORMANCE REVIEW

## Executive Summary

[Two sentences capturing the period. Honest, specific, outcome-anchored.]

## Key Accomplishments

[For each: the outcome first, then the action that produced it. Include the metric, date, or named result. Three to five maximum. Choose the ones with the strongest evidence.]

## Stretch Contributions

[Where I operated above level or outside scope. Two to three items. Each must show the unprompted nature of the work.]

## Honest Reflection

[Two short paragraphs. First: what I learned and where I grew. Second: where I still need to develop, with one specific example, not a generic skill area.]

## Forward Direction

[One paragraph. The work I want to lead in the next review period and why I am the right person to lead it.]

Copilot Power-User Workflow

If your company runs Microsoft 365 Copilot with Researcher available, you have an agentic layer most professionals are not using on the highest-leverage problem of their year. Researcher can crawl your Outlook, Teams, OneDrive, and SharePoint to surface the evidence your performance review needs. The hardest part of writing a self-review is remembering what you did. Researcher solves that.

This is a two-step workflow. Step one runs in Copilot Chat with Researcher activated. Step two runs in any AI tool with the main prompt above.

Step 1: Compile the evidence base in Copilot Researcher.

Open Copilot Chat, switch to Researcher, and paste this prompt:

You are my evidence compiler for a performance review covering [START DATE TO END DATE]. Search across my Outlook, Teams, OneDrive, and SharePoint for the following:

1. KEY MEETINGS: Meetings where I led the agenda, presented to leadership, or made a critical decision. Pull the meeting title, date, attendees, and a one-sentence summary of what I contributed.

2. KEY DECISIONS: Emails or chats where I made or influenced a meaningful decision (project direction, vendor selection, headcount, budget, strategic priority). Pull the date, the decision, and any documented outcome.

3. AUTHORED OR LED DOCUMENTS: Documents I authored or made substantial contributions to. Pull the title, date, and a one-sentence description of what the document accomplished.

4. COMPLETED PROJECTS: Initiatives that closed in this period where I had a leading or supporting role. Pull the project name, my role, and the documented outcome.

5. CROSS-FUNCTIONAL WORK: Communications where I worked across teams or functions outside my immediate scope. Pull the team, the topic, and the outcome.

OUTPUT FORMAT: Group evidence by category. Within each category, list items chronologically. Note any gaps where you could not find evidence (e.g., "no clear meeting summaries available — transcription may not have been enabled"). Do not summarize across categories. Do not infer outcomes that are not documented. If evidence is ambiguous, flag it for me to verify.

Step 2: Feed the evidence into the main prompt.

Review what Researcher returns. Edit out anything wrong, off-target, or sensitive. Then paste the cleaned dossier into the ACCOMPLISHMENTS, STRETCH CONTRIBUTIONS, and ROLE EXPECTATIONS fields of the main Self-Performance Review prompt above. The agentic step compresses four to six hours of memory archaeology into thirty minutes.

Honest caveats on Researcher.

Researcher only sees what your Microsoft 365 admin permits. Some organizations restrict cross-app search. The output requires your review because Researcher weights surface signal (volume of mentions) over strategic signal (importance). Some of your most valuable contributions live outside Microsoft 365 entirely (verbal contributions in meetings, mentoring conversations, work done in third-party tools), and you will need to add those manually. Researcher's coverage of older content depends on what is still indexed. Six months back is generally fine. Two years is not.

Transparency and Notes

• Works in Microsoft Copilot, Claude, ChatGPT, and Gemini on free tier. The Researcher Power-User Workflow requires a corporate Microsoft 365 Copilot license with Researcher available.
• The Editor mechanic is intentionally adversarial. It rejects effort framing because effort framing is what kills most self-reviews. Do not soften your inputs to avoid pushback. The pushback is the value.
• This prompt is part of a series. Vol 32 covered The Presenter's Brief for talk tracks. Future SMPs will cover memos, reports, and emails. Each uses the same Editor mechanic with input fields tailored to the deliverable.
• For sensitive performance review content, do not paste proprietary project names, internal data, or confidential metrics into a tool your organization has not approved. Use roles, functions, and ranges instead of named systems and exact numbers.