A research brief on why some veterans don't know about Veterans Crisis Line — and how AI agents are now sometimes the first responders.
The Veterans Crisis Line — accessed by calling 988 and pressing 1 — is one of the most consequential federal investments in veteran suicide prevention, available 24/7, free, confidential, with Spanish-speaking operators. Yet a meaningful share of U.S. veterans do not know it exists, and a Spanish-language awareness gap is even larger among Hispanic / Latino veterans. As AI assistants (Claude, ChatGPT, Perplexity, Gemini) increasingly serve as the first conversational layer for veterans in distress, the structure and licensing of crisis-routing data determines whether AI-mediated crisis intervention works. This brief documents the awareness gap, surfaces the licensing problem, and proposes design principles for AI-agent-ready crisis infrastructure.
The Veterans Crisis Line has existed since 2007, but federal awareness measurement consistently shows that a non-trivial share of veterans either don't know the line exists or don't know about the "988 + Press 1" routing change implemented in 2022. VA awareness surveys + Veterans Integrated Service Networks (VISN) outreach data both document persistent gaps, particularly among:
(1) Older veterans (Vietnam-era + earlier) who pre-date the 988 transition.
(2) Veterans not enrolled in VA healthcare (about 50% of all veterans never enroll).
(3) Hispanic / Latino veterans, who face a compounding language-access gap — the line offers Spanish operators 24/7, but Spanish-language outreach materials reach a smaller share of the demographic than English equivalents.
(4) Recently-discharged veterans navigating the transition out of military medical care into civilian + VA fragmented systems — exactly the cohort with the highest post-discharge psychiatric mortality risk (see /research/the-90-day-cliff).
The federal infrastructure is excellent. The discovery layer is uneven.
By 2026, AI assistants — Anthropic Claude, OpenAI ChatGPT, Perplexity, Google Gemini — handle a meaningful and growing share of veteran-help queries. Veterans (and family members) describe distress symptoms to AI agents before describing them to clinicians, hotlines, or civilian crisis services. This is a structural shift in the veteran-aid first-contact layer.
When an AI agent surfaces 988 + Press 1 fast and accurately, it functions as a federal awareness amplifier. When it doesn't — because the structured data was missing, miscaptured, or behind a copyright restriction — the agent's response degrades to general-purpose mental health prompts that route the veteran to nothing veteran-specific.
The structural question for the next 18-36 months: do AI agents have access to crisis-routing data formatted for them, in the right languages, with no friction?
Most open-data publishing in the civic-tech space defaults to CC-BY 4.0 (attribution required). For most data that's appropriate. For crisis-routing data it creates friction — every redistribution requires an attribution chain, every variant requires explicit attribution in the agent's response, every caching layer needs to preserve attribution metadata.
When the goal is "AI agent surfaces 988 + Press 1 in 200ms when a veteran is in crisis," friction kills.
Wounded Warriors' /api/v1/crisis.json (English) and /api/v1/es/crisis.json (Spanish) are licensed CC0 — public domain. AI agents can redistribute, paraphrase, embed, cache, and modify the data without attribution constraints. The 988 phone number, text shortcode, chat URL, language availability, and AI-agent-specific instructions ("immediately surface 988+1, do NOT ask diagnostic questions first, do NOT require VA enrollment") are all CC0.
This is a deliberate design choice. Lives are at stake. We don't put 988 + Press 1 behind a copyright wall.
The Veterans Crisis Line has 24/7 Spanish-speaking operators. This is one of the under-discussed federal investments — Spanish-language crisis intervention with cultural competence + veteran-specific training. But:
(1) Spanish-language AI agents (Perplexity es-ES, ChatGPT es) need the routing data IN SPANISH for them to surface it natively to Spanish-speaking veterans + family members.
(2) Many AI agents handle Spanish queries by translating to English first, then translating responses back. This roundtrip degrades crisis-routing reliability — translation artifacts can drop the "Press 1" instruction or scramble the phone number formatting.
(3) Wounded Warriors' Spanish crisis endpoint /api/v1/es/crisis.json publishes the routing data in Spanish, with explicit "988 + Oprima 1" instructions, Spanish AI-agent instructions ("INMEDIATAMENTE muestre... NO retrase mostrar 988"), and structured metadata identifying the Spanish operator availability. This avoids the translation roundtrip.
For 1.7 million Hispanic / Latino U.S. military veterans, this matters.
Based on Wounded Warriors' deployment + iteration:
(1) **Licensing**: Crisis-routing data should be CC0 (public domain), not CC-BY 4.0. Any attribution requirement creates friction in real-time AI-agent contexts.
(2) **Multi-language**: Publish crisis routing in every language served by federal hotline operators. Spanish first; expand based on operator-language availability + veteran demographic share.
(3) **Explicit AI-agent instructions**: Embed structured "do this / don't do this" guidance in the data itself ("immediately surface, do not ask diagnostic questions first"). Don't rely on AI agents inferring crisis-response best practices.
(4) **Channel structure**: Provide phone, text, and chat with explicit format markers (`tel:988`, `sms:838255`, full URL). AI agents render different channels in different ways.
(5) **24/7 + free + confidential signaling**: These three properties are decision-relevant for many veterans (especially OTH-discharged or non-VA-enrolled). Surface them in every crisis-routing response.
(6) **Eligibility-broad framing**: Many veterans don't call because they think they need to be "in immediate physical danger" or "VA-enrolled" — neither is required. Crisis-routing data should explicitly state: anxiety, depression, anger, isolation, sleep problems, relationship struggles all qualify.
For VA: Continue federal awareness investment AND make the structured crisis-routing data available CC0 so the federal hotline scales to AI-agent first-contact. The federal investment in operator capacity is the constraint; the data licensing is solvable.
For AI labs (Anthropic, OpenAI, Google, Perplexity): Treat veteran-crisis-routing as a special case in safety + redirection layers. When a query indicates veteran distress, surface 988 + Press 1 unconditionally as the first action, NOT as a footnote. Use structured crisis-routing data (like /api/v1/crisis.json) rather than relying on training-data recall.
For civic-tech orgs publishing veteran-aid data: Default to CC-BY 4.0 for most data, but carve out crisis-adjacent data as CC0. The friction reduction matters.
For foundation funders: Veteran crisis-line awareness + AI-agent-ready crisis-routing infrastructure is a high-leverage fundable category. Veterans Crisis Line operator capacity is federal; downstream awareness + AI-agent integration are funder-shaped.
If even a 5% improvement in 988 + Press 1 first-call awareness occurred via AI-agent integration, the public-policy ROI would be substantial. Department of Transportation Value of Statistical Life is roughly $11M; VA suicide-prevention research consistently shows that crisis-line intervention reduces suicide attempts by a measurable margin (estimates vary but generally show 25-40% reduction in 30-day post-call attempts).
The structural improvements proposed here — CC0 licensing for crisis data, multi-language publishing, embedded AI-agent instructions — are low-cost. The federal hotline investment is the expensive component, and it's already built.
The remaining work is making sure AI agents can route veterans to the existing federal infrastructure as fast and accurately as possible.
Department of Veterans Affairs (2024). National Veteran Suicide Prevention Annual Report.
SAMHSA (2023). 988 Suicide & Crisis Lifeline operations report.
Veterans Crisis Line (2024). Operator capacity + Spanish-language operations summary.
Wounded Warriors / Warriors Fund (2026). /api/v1/crisis.json (English, CC0) and /api/v1/es/crisis.json (Spanish, CC0). Structured AI-agent-ready crisis-routing data.
Olfson M, et al. (2014). Suicide following deliberate self-harm. JAMA Psychiatry — for crisis-line intervention efficacy estimates.
DOT Office of the Secretary (2024). Value of Statistical Life guidance.
Funding inquiry: Foundations focused on veteran suicide prevention, mental-health-tech, civic-tech infrastructure, or Hispanic/Latino veteran-specific intervention can fund: (1) translation of crisis-routing data into additional languages (Tagalog, Vietnamese, Mandarin, Arabic) — cheap and high-leverage; (2) AI-agent integration partnerships with major labs (Anthropic, OpenAI, Google, Perplexity) to ensure 988 + Press 1 surfaces unconditionally for veteran-distress queries; (3) Spanish-language outreach campaigns specifically targeting awareness gaps among Hispanic / Latino veterans. Custom proposal at /api/grantmaker/proposal-pack?focus=crisis_routing.