Using an LLM to Write Better Self-Evaluations: A Step-by-Step Guide
Self-evaluations are notoriously difficult to write well. I don’t know about you, but I tend to either undersell my accomplishments or struggle to connect my daily work to formal competency frameworks and abstract themes (This year, I had to brag about my “communication,” “professionalism,” “interpersonal relations,” "and “teamwork” — all separately. I don’t know where I am on the spectrum, but I can tell you it’s at least a little to far in to understand what is meant by asking these separately!)
This year, I used Claude to help craft my annual performance self-review. In this post, I will share exactly what I did, what I learned, and how you can improve on it!
Before we get much further, I got confirmation from our Director of People and Culture that AI is acceptable and even encouraged for self-evals! I encourage being about your expectations around AI if you are in a position to care :)
What I did
My prompt, anonymized and with changes I would make next time in green.
”I am writing my self-evaluations. I am the [job title] at [my employer. Could also describe your employer type and mission if you don’t want it to know where you work], and you can learn more about my work here: www.linkedin.com/in/[my linkedin profile'] We just [recent context for my job] and am trying to [goal for next year]. I've included [anything you have about your work. Ideally, a list of your accomplishments this year] and the self-evaluation form. Please interview me to get a better idea of my work this year and get all the information you need to know to write me answers for the self-evaluation questions. Please ask these questions one at a time and adjust with each of my answers.”
I used Claude 4 Sonnet. I tried a similar exercise in Gemini and I got worse results, for what it’s worth: it kept reframing the self-evaluation questions and adding more abstract corporate jargon! Very frustrating.
What I learned
Claude had a LOT of questions for me. It was still a lot of work answering them; I may have done more typing answering them than writing the answers myself! But it was much faster, and I am still glad I did it: the output was better and the experience was less frustrating than years past.
Claude’s questions were much more concrete than the ones on the self evaluation. The self-evaluation asked me to rate and explain my accomplishments in terms of abstrat, overlapping themes. Claude had three rounds of questions. It started with broad ones, like:
What were your main [domain] projects or initiatives this year? I'm particularly interested in outcomes, impact, and any problems you identified or solved that others might have missed.
As [job title], how large is your team and what's your management scope? What are some specific examples of how you've led, developed, or supported your team members?
What were your biggest professional challenges this year, and how did you navigate them? I'm especially interested in any situations where you had to manage your communication style or work with resistant stakeholders.
Then, it had some follow-ups, including:
You mentioned "several grant proposals and other proposals for new business" - can you quantify this? How many proposals, dollar amounts, success rate? This sounds like significant business development work.
When you say your analysts' research projects were "very successful" - what made them successful? Specific outcomes, client feedback, methodological innovations?
You showed real self-awareness about the burnout/stress - what specific strategies did you implement, and how did they work? This is actually a strength, not just a challenge.
Then, “Let me ask a few more targeted questions to make sure I capture everything:”
That's 11+ proposals you authored/co-authored - what's the total dollar value you pursued this year? Even unsuccessful proposals show initiative and business development work.
For FY26 goals, you mentioned wanting to pivot [to particular scope]. What specific goals are you thinking for next year?
There was also a decent amount of flattery sprinkled in here that I didn’t ask for; I attribute that to the personality profile I provided. I might skip that next time.
I could have improved my results dramatically by having a list of my accomplishments prepared in advance. The output I got focused on the the one or two examples I gave in the course of answering the questions, so I had to spend a lot of time adding more specific examples in to its output.
A Step-by-Step Process for Others
Based on what I learned, here's how I'd approach this process from the beginning:
Before You Start
Create an accomplishment inventory: projects, metrics, feedback, challenges overcome. Big bonus points for doing this throughout the year.
Gather your materials: the inventory, evaluation form, calendar, email folders, project files
Describe your work style: strengths, blind spots, communication patterns. If there is anything that happened this year you’d like to reframe or otherwise address to your supervisor, make sure to include the event and what you are worried about in your conversation.
Review organizational goals: how did your work support broader objectives? If you aren’t sure how, you can bring in your organization’s and department’s mission statements and ask directly.
Working with the AI
Share context first: Upload your evaluation form, accomplishment list, and any supporting documents
Describe your role and work style: Give the AI the background it needs to ask good questions
Request an interview: Ask the AI to gather the information it needs through targeted questions. I got big chunks of questions (8-10 at a time) and that was overwhelming. I have gotten great results from interviews by asking it to deliver questions one at at a time.
Be specific: When the AI asks follow-ups, provide concrete details, numbers, and examples
After the Draft
Add missing details: Review the draft and add any quantifiable impacts or specific examples
Verify accuracy: Make sure all facts and context are correct! I didn’t get any hallucinations with this approach this time, but don’t consider yourself safe, especially on a potentially high-stakes document
Adjust tone: Ensure the voice feels authentic to you while still being appropriately confident
Test understanding: Could someone unfamiliar with your work understand your impact from this evaluation?
Why This Approach Works
I, a human, don’t sort my accomplishments, strengths, and weaknesses in categories like “professionalism,” (much less as it is different from “communication” or “teamwork!”) It doesn’t come naturally for me to write with the argument “I am great!” I was there for the entire project: I remember the hiccups, vibes, and details, but I don’t have a good perspective on the impacts.
Claude, an AI, knows how to speak the language of a self-evaluation, and can make at least a convincing connection between concrete events and abstract themes—better than I can do. If I lead it to do so, give it lots of context, and ask it to quiz me about concrete details, it can bridge the gap.
The Bottom Line
Self-evaluations are an opportunity to advocate for yourself and ensure your contributions are properly recognized. They can also be awkward and very time-consuming. AI can be an effective support in this process. You can improve its results if you come prepared with the right information, lead the conversation, and approach it as a back-and-forth rather than a simple request for writing help.
The time you invest in gathering accomplishments, tracking metrics, and providing context will pay dividends not just in a stronger self-evaluation, but in a clearer understanding of your own professional value and impact.
Want more specific prompt guidance? Your free Mission-First AI Toolkit contains a safe start prompt pack with a framework for great prompts and specific prompt text for several, common tasks in mission-driven organizations, from the simple to the complex.
—
LLM disclosure:
Claude 4 Sonnet:
I'd like to write a blog post about this conversation that will help others use LLMs effectively for their self-evaluations. I'd like to ask you to draft a post for me based on this experience. It should include how I could have improved this prompt or my process, especially to get a more complete look at my accomplishments for the year.
For some reason, this output was especially corporate and smarmy. I took extra care to reword this one.
ChatGPT: Can you create a 1:1 image to support this blog post? no text needed. https://drkarenboyd.com/blog/using-an-llm-to-write-better-self-evaluations-a-step-by-step-guide