Using AI Prompts to Convert Messy Project Notes into Star Resume Bullets
Ads
When you feed raw project notes into an LLM and ask for resume bullets, the model fills gaps by inventing metrics. It writes things like "boosted revenue by 40%" or "reduced latency across 12 teams" even when your notes mention neither. Those fabrications damage your credibility during interviews and background checks.
You can prevent invented metrics by splitting the task into two passes. First, force the model to extract verified verbs and scope boundaries directly from your notes. Then prompt it to format those extracted facts into STAR (Situation, Task, Action, Result) bullets, requiring placeholders for any missing numbers.
Why standard resume prompting invents accomplishments
Large language models predict probable completions. Resume datasets are saturated with strong percentages and quantitative impact markers. When your input lacks numbers, the model generates plausible-sounding figures to match that typical pattern.
To stop this, separate fact extraction from stylistic formatting. If an accomplishment lacks a baseline metric or measurement, instruct the model to return a bracketed placeholder like [Insert verified %] instead of estimating a number.
The two-pass bullet extraction workflow
This workflow treats your raw project notes as the only factual source. You can run it in any chat interface.
The production prompt to run tonight
Copy and paste this prompt into your chat tool alongside raw notes from a project, sprint log, or quarterly review:
You are a technical resume editor. I am providing messy, unorganized notes from a work project.
Your task:
1. Extract only the verified tools, core responsibilities, and direct actions mentioned in the text.
2. Convert these into 3 to 4 resume bullet points following the STAR format (Action Verb + Context/Task + Action + Measurable Result).
3. CRITICAL CONSTRAINT: Do not invent any numbers, percentages, team sizes, or outcomes not explicitly stated in the source text. If a metric is implied but unstated, output '[Insert verified metric]' in its place.
4. Highlight any technical tools mentioned.
Here are the raw notes:
[PASTE YOUR RAW NOTES HERE]Auditing the output before applying
Once the model returns bullets, audit each line. Read through and check every noun, verb, and number. If you cannot explain the concrete work behind a tool or metric during an interview, remove it or rewrite it to match what you did.
Interviewers check claims by asking for details about edge cases and individual contributions. An accurate bullet with modest scope is easy to defend; an inflated claim falls apart quickly under questioning.
Frequently Asked Questions
What if my original notes contain no quantifiable metrics?
Direct the prompt to focus on scope and operational changes instead of numbers. A defensible outcome can describe resolving a dependency, standardizing a manual runbook, or delivering a feature against a production deadline.
Why should I avoid using AI to generate the initial project notes?
An LLM has no record of what you built, debugged, or organized. Pulling notes from your actual commit history, sent emails, or project tickets ensures your source facts remain accurate.
Does this structure work for non-technical roles?
Yes. The STAR format and the anti-hallucination constraints apply equally to operations, sales, customer support, and administrative positions.