Using AI to Write NCOERs (Without Getting Yourself in Trouble)
AI tools like ChatGPT, Claude, and Gemini can dramatically improve evaluation bullets when used responsibly. They can also get you flagged, end careers, or compromise sensitive information when used carelessly. This guide shows you how to use AI as an editor for content you wrote — not a ghostwriter that invents accomplishments.
Applies to: NCOER (DA 2166-9) · OER (DA 67-10) · Counseling (DA 4856)
- No PII or CUI in public AI tools. Names, SSNs, DOD IDs, unit designations below brigade, deployment locations, medical details — none of it. Public ChatGPT/Claude/Gemini are commercial services not approved for CUI.
- The rater observes; AI does not. AR 623-3 makes you (the rater) personally responsible for the evaluation content. AI cannot observe performance. Don't let it invent accomplishments.
- AI is an editor, not an author. Write the bullet yourself based on what actually happened. Use AI to tighten grammar, strengthen verbs, and check brevity — not to generate content from nothing.
- Follow your local command policy. Some units prohibit commercial AI for any official work. Your S6 or G6 has the current approved-tool list and authorized data levels.
What AI Can Help With
After you draft a bullet from observed performance and sanitize the PII out, AI is excellent for:
- Rewriting passive voice into active — common bullet weakness
- Suggesting stronger action verbs — "led", "directed", "engineered", "transformed" vs. weak "did", "helped", "worked on"
- Tightening to fit character limits — the system enforces strict per-line limits
- Restructuring into A-V-E or 5W+Impact format — Action, Verb, Effect; or Who/What/Where/When/Why/Impact
- Eliminating filler words — "very", "really", "in order to", "actually"
- Checking for ambiguity — does this bullet clearly communicate what happened?
- Suggesting 3 alternative phrasings — pick the one that sounds most like your voice
What AI Should Never Do
- Generate the bullet from scratch with no observed performance behind it
- Invent specific awards, numbers, or accomplishments to make a bullet sound better
- Process any text containing names, SSNs, DOD IDs, unit identifiers below brigade, or deployment locations
- Replace your direct observation as a rater (you must have seen the performance)
- Generate scoring or box-check recommendations (those are the rater's judgment, not AI's)
- Process classified information of any kind, period
The Safe Workflow (5 Steps)
Sanitization Checklist
- Soldier names — first, last, nicknames
- Rank in combination with a name (e.g., "SGT Smith")
- SSN, DOD ID number, EDIPI
- Unit designation below brigade level (battalion, company, platoon, squad)
- Specific FOB, COP, or installation names
- Country / region of deployment when paired with dates
- Operation names that are not publicly released
- Medical conditions or specific profile content
- Security clearance level or compartmentation
- Family details (spouse name, dependent count if combined with identifiers)
- Email addresses, phone numbers
- Vehicle bumper numbers, callsigns, frequencies
Sample Prompts (Safe Format)
Notice these prompts ask for specific improvements and never contain identifying information. The placeholder [sanitized bullet] is where you paste your already-sanitized text.
Rewrite this NCOER bullet using active voice and stronger action verbs. Keep it under 150 characters. Do not add details that aren't already present: [sanitized bullet]
This bullet is too long. Tighten it to under 130 characters without removing any of the specific accomplishments or numbers. Do not invent new details: [sanitized bullet]
Restructure this bullet into Action-Verb-Effect format (what was done, how it was done, what the result was). Keep all numeric details exactly as they are. Do not add accomplishments I didn't mention: [sanitized bullet]
Give me 3 alternative phrasings of this bullet. Each must preserve every fact I included. Don't add awards, numbers, or outcomes I didn't write. Each version should be under 150 characters: [sanitized bullet]
Why Over-Reliance Is Risky
Senior raters read hundreds of bullets a year. Patterns that scream "AI did this":
- Buzzword soup — "seamlessly orchestrated", "spearheaded transformative initiatives", "leveraged synergies"
- Uniform sentence structure across multiple bullets — all start with the same verb pattern
- Sudden voice change from the rater's usual style — this year reads nothing like last year
- Vague impressive-sounding outcomes with no specific numbers or measurable results
- Identical bullets across multiple soldiers — happens when raters paste a template prompt and don't edit enough
Related Resources
Frequently Asked Questions
Is using AI to help write NCOERs against Army policy?
There is no blanket Army-wide prohibition on using AI tools to assist with writing administrative documents like NCOERs. However, several constraints apply: (1) AR 623-3 requires the rater to base the evaluation on direct observation, not invented content. (2) DoD CIO has issued guidance restricting how generative AI may be used with government information. (3) Some commands have stricter local policies. (4) Putting Controlled Unclassified Information (CUI), names, SSNs, unit identifiers, or operational details into a public AI tool is an information disclosure issue. Always check your local command policy and your unit S6 before using a specific AI tool for official work.
Will my senior rater or HRC be able to tell I used AI?
Sometimes — yes. AI-generated text tends to over-use certain phrases ("seamlessly", "spearheaded", "leveraged synergies"), uses uniform sentence structure, and lacks the specificity that comes from observation. Senior raters who review hundreds of bullets often notice when something doesn't sound like the rater's usual voice. Treat AI as an editor, not an author — your written voice and the specific accomplishments should be yours.
What information should I never paste into a public AI tool?
Never paste: soldier names, SSNs, DOD ID numbers, unit names below brigade level, specific deployment locations, classified information, operational details, medical information, security clearance details, or any Controlled Unclassified Information (CUI). Genericize first — replace "SGT Smith of 1-501 IN at FOB X" with "the rated NCO" before pasting anything.
Are there approved AI tools the Army provides?
Yes — your S6 or G6 will know what is currently available. Army-approved AI environments are typically vetted for handling government information and may be approved for CUI work where commercial tools (public ChatGPT, Claude, Gemini) are not. The list of approved tools and their authorized data levels changes; ask your IA/Cybersecurity office for the current authorization.
Can I just have AI write the whole NCOER for me?
No — and you should not. AR 623-3 makes the rater personally responsible for the content. AI cannot observe a soldier's performance. If an AI invents accomplishments that did not happen, you (the rater) are signing a false evaluation, which is a UCMJ issue, not a process improvement. Use AI to refine bullets you have already written from direct observation.
What if my unit prohibits public ChatGPT?
Follow that policy. Some commands restrict commercial AI tools entirely; others restrict them only for processing CUI. If commercial tools are restricted, you can still use the workflow on this page with an approved tool, with offline tools (e.g., word processor grammar check, thesaurus), or with a peer who is willing to review.
Does this guidance apply to OERs and counseling too?
Yes — the same principles apply to OERs, the OER Support Form, and DA Form 4856 counseling. The rater is responsible for the content; AI can help with grammar, structure, and verb choice; PII and CUI must never go into public tools; and the underlying accomplishments must be real observations, not AI inventions.