Getting Started with GenAI.mil - Dec 2025 Edition

Two weeks ago week, the Pentagon put Google's Gemini for Government on the desktops of 3 million military personnel, civilians, and contractors. If you have a CAC, you now have access to a frontier AI model cleared to handle CUI.

Most people will poke at it a few times, get underwhelming results, and move on. This isn't because the tool is bad—it's because talking to AI is a skill, and nobody taught you how to do it.

It's kind of like getting handed a professional camera. Point it at something and press the button, and you'll get a picture. But you won't get a good picture until you learn something about composition, lighting, and what all those other buttons do. The gap between "functional" and "effective" is where most people give up.

This piece is a first attempt to close that gap. I'll walk through what GenAI.mil actually is, what it can and can't do right now, and how to set yourself up for success. Then I'll give you ready-to-use prompts—including more sophisticated versions that show what's possible when you put in the work upfront. I'll close with specific applications for information and influence professionals, since that community has some distinctive opportunities with this platform.

What You're Actually Working With

GenAI.mil's current iteration runs on Google's Vertex AI infrastructure, specifically Gemini 2.5 Pro and Gemini 2.5 Flash, with "thinking" mode enabled. It's authorized for Controlled Unclassified Information (CUI) at Impact Level 5. Your data stays within the secure boundary and is never used to train Google's public models.

Key capabilities:

• Standard chat interface for queries and document analysis
• Web search grounded in Google Search (sandboxed for government use)
• Deep Research for comprehensive, sourced analysis on complex topics

Current limitations you need to know:

• No Gems or custom projects — You can't save persistent personas or instruction sets. Every session starts fresh.
• No memory across sessions — The system doesn't remember previous conversations or learn your preferences.
• No image generation — Text outputs only for now.
• No PII — Don't paste personnel records, evaluations, medical information, or other personally identifiable information.
• CUI ceiling — Unclassified work only, though it handles CUI appropriately.

Other models from Anthropic, OpenAI, and xAI are expected to join the platform over time. The habits you build now will transfer.

Three Things to Manage these limitations

Because GenAI.mil doesn't have memory or projects, it starts fresh every session. It will only work with the context you give it in a discussion.

To get the best results without driving yourself crazy, you want to save your personal context and your prompts so you can reuse them. The people who get the most from AI aren't the ones with the cleverest prompts; they're the ones who've built systems to quickly give the AI what it needs to be useful in their context.

Good Context + Good Prompt = Good Results.

When memory and custom projects eventually arrive on this platform (and I expect they will), the habits you build now will pay off. But don't wait. The tool you have today is more capable than you probably realize—you just have to meet it halfway.

1. Build Your Personal Context Block

GenAI.mil starts with fresh memory every session. A personal context block is a short text file (200-400 words) that you paste at the start of conversations where you need personalized output. It tells the AI who you are, what you're working on, and how you like to receive information. You can also build a longer file, something I outline in my AI for Diplomats article.

Save it somewhere you can grab it quickly. I keep mine in a file synced across my phone and desktop.

A simple structure:

ABOUT ME:
- Role: [Your position, organization, and scope]
- Primary audiences: [Who you write for and brief]
- Current focus: [Key projects or priorities]

MY PREFERENCES:
- Tone: [Direct? Formal? Technical?]
- Format: [Narrative prose? Bullets? Executive summary style?]
- Length: [Concise? Comprehensive?]
- Pet peeves: [Jargon? Passive voice? Excessive hedging?]

Pro tip: Ask Gemini to help you create this. Start a session with: "Interview me to help create a personal context file I can paste into future AI sessions. Ask me 5-7 questions about my role, audiences, and communication preferences, one at a time."

2. Start a Prompt Library

When you craft a prompt that works well, save it. When you see a prompt structure that gets good results, save it. Over time, you'll build a personal playbook of what works for your specific needs. This article gives you a head start on some prompts that I played with in Gemini.

3. Organize Your Reference Files

Build a folder of documents you frequently reference or might want to upload as context:

• Your organization's style guide or writing standards
• Key policies or directives you work with regularly
• Templates for recurring products
• Background on ongoing projects

When you start a session, you can upload a relevant document and tell Gemini to use it as context.

For a deeper walkthrough on preparing yourself for AI-augmented work, see my Beginner's Guide: https://community.bastionworks.co/ai-and-diplomacy-beginners-guide/

Five Things Almost Everyone at DoW Can Use This For

These are the use cases I'd point any defense worker toward on day one. For each, I'm giving you a basic prompt and a more sophisticated version. The sophisticated ones may not be perfect for you--that's intentional. They are intended to spark your own thinking and show you what becomes possible when you invest in context-setting upfront.

Copy these. Modify them. Make them yours.

1. Summarize What You Don't Have Time to Read

You know the drill: forty-page directive lands in your inbox, and you need to brief someone on it by tomorrow. AI can reduce that to a focused summary in seconds—but the quality depends entirely on whether you tell it what you specifically need from the document.

Basic prompt:

Summarize this document in 1 page. Focus on key decisions, action items, and deadlines.

[Paste document text or attach document]

Sophisticated prompt:

CONTEXT: I'm a program manager in a PEO supporting Army acquisition programs. I need to brief my O-6 division chief on this policy update tomorrow.

TASK: Summarize this document with the following focus:
1. What changed from the previous version of this policy
2. Compliance requirements and deadlines that affect acquisition programs
3. Any ambiguities or areas where we'll need to seek clarification
4. Recommended actions for our office

FORMAT: Executive summary (half page), followed by a bullet list of action items with deadlines.

DOCUMENT:
[Paste document text or attach document]

2. Draft the First Version of Written Products

Here's where the time savings really show up. I don't use AI to write final products—I use it to get past the blank page. Memos, talking points, information papers, SOPs: describe what you need, give it context, and iterate from there. You still own the final product; you just got there without staring at a cursor for an hour.

Basic prompt:

Draft a memo recommending we extend the deadline for the quarterly report submission by two weeks. Professional tone, one page max.

Sophisticated prompt:

CONTEXT: I'm a branch chief in J3 at a combatant command. I need to send a memo to my division chief (O-6) recommending we push the quarterly readiness report deadline by two weeks due to data quality issues from component commands.

TASK: Draft a decision memo with the following elements:
- Purpose statement (one sentence)
- Background (why we're in this situation—component commands submitted incomplete data)
- Options: (1) Extend deadline 2 weeks, (2) Submit on time with caveats, (3) Submit partial report
- Recommendation with rationale
- Risks of recommended option and mitigation

TONE: Direct, no jargon, action-oriented. My chief prefers bottom-line-up-front and hates passive voice.

LENGTH: 1 page maximum.

3. Get Smart on Unfamiliar Topics, Fast

This is where Deep Research earns its name. When you suddenly need to understand something outside your lane—an emerging technology, a policy area, a partner nation's capabilities—it can produce a comprehensive, sourced overview that would have taken you hours to assemble manually.

Basic prompt:

Use Deep Research to explain zero trust architecture and how DoW is implementing it. Include sources.

Sophisticated prompt:

CONTEXT: I'm an IT portfolio manager preparing a briefing for senior leadership on our command's zero trust implementation roadmap. They'll want to know how we compare to DoW-wide progress and what peer organizations are doing.

TASK: Use Deep Research to produce a briefing-ready analysis covering:
1. DoW's zero trust strategy and key milestones (cite the actual strategy documents)
2. Current implementation status across the department—what's on track, what's lagging
3. How military services and major commands are approaching implementation differently
4. Common challenges organizations are facing and how they're addressing them
5. Recommended resources for our team to consult

FORMAT: Structured briefing document with clear sections. Include source citations I can verify.

4. Translate the Same Information for Different Audiences

You probably do this constantly: take a technical finding and explain it to leadership, take a policy update and explain it to your team, take contractor jargon and make it comprehensible to the program manager. AI handles this reframing well—especially if you're specific about who the audience is and what they care about.

Basic prompt:

Explain this technical security finding in plain language for a non-technical executive.

[Paste finding]

Sophisticated prompt:

CONTEXT: I'm a cybersecurity analyst who needs to brief a vulnerability finding to three different audiences today: (1) my technical team, (2) the system owner who's a program manager without deep technical background, and (3) our O-6 for the risk acceptance decision.

FINDING:
[Paste the technical finding]

TASK: Create three versions of this finding:

VERSION 1 - Technical team:
- Full technical detail, CVE references, specific remediation steps
- Include what we need to test after patching

VERSION 2 - Program manager:
- Plain language explanation of what the vulnerability means for their system
- Mission impact if exploited
- Timeline and resources needed for remediation
- Decision they need to make

VERSION 3 - O-6 risk acceptance brief:
- 3 sentences max: what it is, what's the risk, what's the recommendation
- Frame in terms of operational risk, not technical risk

5. Prepare for Meetings and Actually Capture What Happened

Meetings eat enormous time, and the value often leaks away because nobody captured the decisions and action items clearly. Use AI on both ends: before, to generate agendas and prep questions; after, to turn your messy notes into a structured summary you can actually send to attendees.

Basic prompt:

Here are my notes from today's meeting. Extract all action items with owners and deadlines.

[Paste/attach notes]

Sophisticated prompt:

CONTEXT: I just left a 90-minute IPR (In-Progress Review) on our modernization initiative. Multiple stakeholders, several decision points, and a lot of discussion. I need to send a summary to attendees within 2 hours.

MY RAW NOTES:
[Paste notes—can be messy/incomplete]

TASK: Produce a structured meeting summary with:

1. DECISIONS MADE
- List each decision with enough context that someone who wasn't there understands it

2. ACTION ITEMS
- Format: [Action] | [Owner] | [Deadline]
- Flag any items where owner or deadline wasn't clearly established

3. OPEN ISSUES / PARKING LOT
- Topics raised but not resolved
- Questions that need follow-up

4. KEY DISCUSSION POINTS
- Brief summary of major topics discussed (not a transcript—just the highlights)

5. NEXT STEPS
- When's the next meeting
- What needs to happen before then

FORMAT: Clean, professional, ready to send as an email to attendees.

For Information and Influence Professionals: Five Specific Applications

If you work in public affairs, strategic communications, psychological operations, or civil affairs, you have some distinctive opportunities here. GenAI.mil's combination of Deep Research and web grounding is genuinely useful for the kind of environmental scanning, audience analysis, and message development work that used to eat up enormous time.

You, like I, probably have a full suite of technical tools and vendors you use to help do this. You should still test these prompts. Some tools provide raw data--which may actually be helpful as a source for GenAI prompts. Some tools provide summaries or written analysis. GenAI's deep research capabilities should raise your expectations for what you get out of paid tools and services. If what you are getting isn't as useful as a GenAI deep research task... you should have a conversation with your vendors.

A few caveats before we dive in: These prompts produce starting points for analysis, not finished products. They're for unclassified work only. Apply your professional judgment. Validate sources. Route products through appropriate review processes.

That said—these can meaningfully accelerate your workflow.

1. Map an Information Environment Quickly

Before you can communicate effectively in a region or on an issue, you need to understand the landscape: who are the key voices, what narratives are circulating, how does information flow, what do people trust? This kind of environmental scan used to require days of research. Deep Research can give you a solid first draft in minutes.

Prompt:

CONTEXT: I'm an information operations planner supporting [REGION/MISSION]. I need to quickly understand the information environment before we begin planning.

TASK: Use Deep Research to produce an information environment assessment covering:

1. MEDIA LANDSCAPE
- Major news outlets (state-controlled, independent, international)
- Reach and credibility of each among target population
- Social media platforms dominant in the region and usage patterns

2. KEY VOICES AND INFLUENCERS
- Government spokespersons and official channels
- Journalists and commentators with significant reach
- Civil society voices, religious leaders, cultural figures
- Social media influencers with relevant audiences

3. DOMINANT NARRATIVES
- What narratives is the government promoting?
- What narratives are circulating in independent/opposition media?
- What external actors are active in the information space and what are they pushing?

4. INFORMATION VULNERABILITIES AND OPPORTUNITIES
- Credibility gaps in official messaging
- Topics where audiences are seeking information but not finding credible sources
- Historical examples of effective/ineffective communications in this environment

INCLUDE: Source citations for all major claims so I can verify and dig deeper.

2. Analyze Adversary Narratives and Develop Responses

When someone is pushing a narrative that undermines your mission, your instinct might be to respond immediately. Resist that. First, understand what you're dealing with: what's the structure of the argument, what emotions is it leveraging, where are the weak points? AI can help you slow down and analyze before you react.

Prompt:

CONTEXT: I'm analyzing adversary messaging related to [TOPIC/OPERATION/ISSUE] to support counter-narrative development.

ADVERSARY NARRATIVE TO ANALYZE:
[Paste examples of adversary messaging—articles, social posts, official statements]

TASK: Produce a structured analysis with:

1. NARRATIVE STRUCTURE
- Core claim(s) being advanced
- Supporting arguments and evidence cited (whether valid or fabricated)
- Emotional appeals being leveraged (fear, pride, grievance, etc.)
- Target audience(s)—who is this designed to influence?

2. VULNERABILITIES
- Factual inaccuracies or misrepresentations
- Logical inconsistencies
- Contradictions with adversary's other statements or actions
- Gaps between claims and audience's lived experience

3. STRATEGIC INTENT
- What behavior or belief is this narrative trying to produce?
- How does this fit into broader adversary information strategy?

4. COUNTER-NARRATIVE OPTIONS
- Direct rebuttal approaches (with risks of amplification)
- Alternative narrative approaches (reframing without direct engagement)
- Inoculation approaches (prebunking anticipated future claims)
- For each option: target audience, key message, appropriate channel, risks

5. RECOMMENDED APPROACH with rationale

NOTE: This analysis supports planning. Final products require human judgment, legal review, and appropriate approval.

3. Understand Your Audience Before You Start Messaging

Generic messaging fails. You know this. But really understanding an audience—their values, their information habits, their historical experiences, what makes them trust or distrust a source—takes significant research. This is exactly what Deep Research is built for: synthesizing multiple sources into an audience profile you can actually use.

Prompt:

CONTEXT: I'm developing a communication campaign targeting [SPECIFIC AUDIENCE] in [COUNTRY/REGION] regarding [TOPIC/OBJECTIVE].

TASK: Use Deep Research to produce an audience analysis covering:

1. AUDIENCE PROFILE
- Demographics, but more importantly: shared values, priorities, concerns
- Historical experiences that shape how this audience processes information on this topic
- Current attitudes toward the US/US military (with nuance—not monolithic)

2. INFORMATION CONSUMPTION HABITS
- How does this audience typically get news and information?
- Which sources do they trust? Which do they distrust?
- Role of interpersonal networks vs. mass media vs. social media

3. CULTURAL CONSIDERATIONS
- Communication norms (direct vs. indirect, role of authority, humor, etc.)
- Religious or cultural sensitivities relevant to the topic
- Historical references or analogies that resonate (positively or negatively)
- Visual/aesthetic preferences if relevant to product development

4. MESSAGE TESTING CONSIDERATIONS
- Potential unintended interpretations of common US messaging approaches
- Themes that research suggests resonate with this audience
- Themes to avoid and why

5. RECOMMENDED MESSENGER PROFILE
- What kind of voice/source would be most credible to this audience on this topic?
- US official? Local partner? Peer voice? Expert? Cultural figure?

INCLUDE: Source citations, with notation on source quality (academic research, journalistic, polling data, etc.)

4. Track How a Topic Is Playing Across Different Information Ecosystems

The same event gets framed completely differently depending on where you look: US mainstream media, regional outlets, adversary state media, social platforms. Manually surveying all of these is exhausting. AI can give you a structured snapshot of who's saying what—which is often the first step in deciding whether and how to engage.

Prompt:

CONTEXT: I need to understand how [TOPIC/EVENT/ISSUE] is being covered across different information ecosystems to support situational awareness and planning.

TASK: Use web search and Deep Research to survey coverage and produce a synthesis:

1. US/WESTERN MAINSTREAM COVERAGE
- Dominant framing and key themes
- Notable voices and their positions

2. REGIONAL MEDIA COVERAGE (specify region)
- How are outlets in [REGION] covering this?
- How does framing differ from Western coverage?
- Local context being added or emphasized

3. ADVERSARY/COMPETITOR STATE MEDIA
- How are Russian, Chinese, Iranian (as relevant) state media covering this?
- What narratives are they promoting?
- What are they omitting or downplaying?

4. SOCIAL MEDIA DISCUSSION (where observable)
- What angles are getting traction?
- Notable hashtags, influential posts, viral content
- Differences between platform ecosystems if relevant

5. NARRATIVE GAPS AND OPPORTUNITIES
- What aspects are undercovered?
- Where are official sources losing the narrative?
- Where is there appetite for information that isn't being provided?

6. SYNTHESIS: Key takeaways for planning

NOTE: This is a point-in-time snapshot for initial orientation. Validate through established monitoring channels.

5. Accelerate Foreign Language Content Development

If you've ever waited three days for a translation to come back from the language office, only to find it needs significant revision... you know the pain. AI won't replace your linguists—but it can produce a working first draft, flag cultural localization issues, and back-translate so you can verify meaning was preserved. It compresses the review cycle significantly.

Prompt:

CONTEXT: I'm developing communication products for [TARGET LANGUAGE] speaking audiences in [COUNTRY/REGION]. I have English source content that needs to be adapted—not just translated, but culturally localized.

ENGLISH SOURCE CONTENT:
[Paste content]

TASK: Provide three outputs:

1. DIRECT TRANSLATION
- Accurate translation into [TARGET LANGUAGE]
- Maintain fidelity to original meaning

2. CULTURAL LOCALIZATION NOTES
- Phrases or concepts that don't translate directly
- Idioms or expressions that would sound foreign or awkward
- Suggested adaptations that convey the same meaning more naturally
- Cultural references that should be substituted
- Tone adjustments needed for this audience

3. LOCALIZED VERSION
- A version that conveys the same core message but reads as if originally written for this audience
- Flag any places where localization required substantive changes to meaning

4. BACK-TRANSLATION
- Translate the localized version back to English so I can verify intent was preserved

5. SENSITIVITY CHECK
- Any words, phrases, or concepts that could have unintended negative connotations
- Regional dialect considerations
- Anything that could be perceived as offensive, condescending, or culturally tone-deaf

NOTE: This requires review by a qualified linguist with cultural expertise before use. AI is a first draft, not final QC.

Where does this go next?

GenAI.mil will get more capable over the next year. More models will arrive. Memory and custom projects will come eventually. The prompts you write today will probably feel clumsy in a year. The biggest change is likely to be access to files, which might enable you to build a more sophisticated Personal Knowledge Management system that stores more of your context and work.

But here's the thing: the underlying skill—being able to clearly articulate what you need, provide relevant context, and iterate toward useful outputs—that transfers. It's a communication skill dressed up as a technology skill. And like any communication skill, you get better by doing it, not by waiting until conditions are perfect.

So pick one prompt from this article. Copy it. Modify it for something you're actually working on. See what happens. If it works, save it. If it doesn't, figure out why and try again.

That's the whole game.

For more on preparing yourself and your organization for AI-augmented work, you can follow on Linkedin or on my email newsletter. If you're interested in how I help organizations navigate technology-driven transformation, send me a DM to start the conversation.