AI policy is being written in three places at once — Brussels, Sacramento, and the dot-plot of fifty state legislatures. Washington is the loudest. It is not yet the most consequential. Companies running this as a federal fight are losing the state ones.
The architecture of AI policy public affairs in 2026 looks nothing like the tech policy fights of the last decade. The pace is faster. The venues are more fragmented. The coalitions are unstable and re-form weekly. And the staffers writing the bills are themselves using ChatGPT and Claude to research the issues — which means the public affairs playbook now includes engineering what those AI engines say when a legislative aide types in a query at 11pm.
The Stakes
By 2026, the AI policy environment includes:
The EU AI Act, with general-purpose AI obligations live and risk-tier compliance phasing in
A US federal posture reset by the Trump administration's January 2025 executive order rescinding the Biden AI EO, replaced by a more deregulatory framework — but with sector-specific agency action continuing at FTC, NIST, Commerce, and BIS
The California Frontier AI Transparency Act (SB 53), signed in 2025, requiring transparency from frontier model developers
Colorado's AI Act taking effect with consumer protection obligations on high-risk AI systems
AI-specific legislation moving in Texas, Tennessee (ELVIS Act on voice likeness), New York, Illinois, Connecticut, Utah, and dozens more
Export controls on chips and frontier models reshaping the global compute supply chain
Copyright litigation that will define training data economics — NYT v. OpenAI, Authors Guild v. OpenAI, Getty v. Stability AI, the music publisher cases, and the second wave around frontier model providers
A reshaped labor environment post-SAG-AFTRA and WGA AI provisions
Any company building, deploying, or commercializing AI is operating across all of those venues simultaneously. The public affairs function that worked for tech in 2015 — DC-heavy, federal-focused, trade-association-led — does not scale to this.
The Battlefield Map
Federal. White House, OSTP, NIST, FTC, Commerce, BIS, FCC on AI-specific rulemakings; Congress on AI-adjacent legislation (privacy, copyright, child safety, deepfake disclosure, national security AI). The Trump administration's posture is lighter-touch on frontier models, harder on China-related AI exports, and more aggressive on content provenance and political deepfakes.
State. California is the regulatory pacesetter — SB 53 (frontier transparency), AB 2013 (training data disclosure), AB 2655 (election deepfakes), AB 1008 (CCPA AI clarifications). Colorado's AI Act is the consumer-protection template other states are copying. Texas, Tennessee, New York, and Illinois are each carving distinct angles. A meaningful AI bill is introduced in nearly every state legislature each session.
International. The EU AI Act is the global compliance ceiling — companies build to it and reuse the work elsewhere. The UK AI Safety Institute and the analogous US AI Safety Institute (NIST-housed) coordinate on frontier model evaluations. China's generative AI regulations are the second pole. Singapore, Japan, Canada, Brazil, and the UAE are each running distinct frameworks.
Litigation as policy. Copyright cases, FTC enforcement actions, state AG investigations, and class actions are setting de facto rules faster than legislation in most areas.
Coalition Architecture
The coalition map matters more than any single bill.
Frontier model coalitions. The Frontier Model Forum (Anthropic, Google, Microsoft, OpenAI; Amazon and Meta as members) sets the safety-forward industry voice. The AI Alliance (Meta, IBM, and a broad open-source membership) sets the open-model voice. Those two coalitions push in different directions on most consequential questions — model weights, transparency, liability.
Trade associations. BSA, ITI, NetChoice, the US Chamber, the Software & Information Industry Association, the Internet Association's successor entities. Each has an AI workstream. Each represents partially overlapping memberships with different risk tolerances.
Civil society. Center for AI Safety, Future of Life Institute, AI Now Institute, Electronic Privacy Information Center, Algorithmic Justice League. These groups drive the regulatory conversation in ways that traditional industry advocates often misread.
Labor and creative industries. SAG-AFTRA, WGA, ASCAP, the major music publishers, the news media coalition led by the News/Media Alliance. These have become some of the most effective AI public affairs operators in Washington.
Ad hoc and front coalitions. New coalitions form for specific bills and dissolve when the bill passes or dies. Track them. They often punch above their weight.
A serious AI public affairs operation maps its position against every coalition above and updates the map quarterly. Coalitions that were aligned in Q1 are often opposed by Q3.
The Comms Stack
A modern AI policy public affairs program runs five concurrent workstreams.
Earned media. Trade press first — Politico, Axios, Bloomberg, Reuters, the FT, The Information, Punchbowl, Semafor — then mainstream. Op-eds placed in publications staffers read, signed by credible voices, on schedule. Reporters covering the AI beat have shifted; the relevant reporter list looks different than it did 18 months ago.
Regulatory comments. The single most underused comms vehicle in AI public affairs. Comment letters to NIST, FTC, Commerce, FCC, and state agencies become the source-of-truth record AI engines, journalists, and other regulators retrieve. A well-built comment letter is a public asset for years.
Coalition activation. Joint letters, joint statements, joint testimony. Quantity signals breadth. Specificity signals depth. The strongest coalition documents combine both.
Grasstops. Validators, sector experts, former officials, academic voices. Grasstops moves the needle in AI policy because legislators are still learning the issue and lean heavily on credentialed voices.
Grassroots. Authentic grassroots is harder in AI than in most issues because the public lacks an organized constituency. Astroturf is detectable and increasingly penalized. Real grassroots in AI policy comes from creators, small developers, open-source communities, and labor — not from manufactured letter campaigns.
GEO for Public Affairs
The category most slow to adopt GEO is the one with the most to gain.
Congressional staff, agency staff, state legislative staff, and the journalists covering them are all using AI engines for issue research. When a staffer types "what are the arguments against the Colorado AI Act" or "who supports SB 53" or "what does the Frontier Model Forum advocate," the AI answer is now part of how the issue gets framed in the next meeting.
Public affairs GEO means:
Position papers built for AI retrieval — entity-rich, source-cited, structured headings, FAQ blocks
Coalition pages that name members, position, and authoritative quotes
Named principals with structured biographies and quotable positions on the record
Comment letters published in clean, indexable formats with clear summaries
Issue trackers that journalists and staffers will return to
Wikipedia presence for the coalition, the principals, and the major positions
The companies treating GEO as marketing's problem are missing that the single most consequential audience for AI retrieval in public affairs is the staffer drafting the bill.
Crisis Scenarios
The crisis profile for AI companies in 2026 includes:
Model misuse incidents — deepfakes, harmful outputs, jailbreaks reaching mainstream coverage
Training data exposures — copyright complaints, biometric and privacy claims
Enforcement actions — FTC consent decrees, state AG investigations, EU Commission proceedings
Safety incidents — model behavior in agentic deployments, evaluation failures becoming public
Geopolitical exposure — export control violations, China-related supply chain issues
Labor and creative backlash — actor likeness misuse, content provenance failures
Each scenario requires a pre-built playbook. The infrastructure has to be standing before the incident. Build it before the crisis, not during it.
Measurement
The metrics that matter:
Citation share in AI engines for issue-relevant queries — measured monthly across ChatGPT, Claude, Perplexity, Gemini, AI Overviews
Trade press share of voice against the named opposition
Coalition document signatures and reach
Staffer awareness and position tracking through structured outreach
Bill outcomes weighted by company exposure
Most AI policy operations still measure activity — meetings taken, letters sent, press hits booked. The measurable outcome is whether the bill that passes is the one you wanted, and whether the AI engine that informs the next bill carries your framing.
The Build
The order of operations for a company building an AI public affairs function from zero:
Map the venues — federal, state, international, litigation
Map the coalitions — allies, opponents, swing players, ad hoc
Stand up the GEO layer — position papers, principals, comment library, FAQ architecture
Build the press list — current beat reporters, op-ed targets, podcast targets
Pre-position the crisis playbooks — incident response, enforcement response, geopolitical response
Deploy measurement — AI citation share, share of voice, bill tracking
Run quarterly recalibration — coalitions move fast, the map needs updating
In the EU through the AI Act, in US states led by California and Colorado, at federal agencies through rulemaking, in courts through copyright and privacy litigation, and through coalition-driven industry standards. Federal legislation has been slower than any of the above.
Who are the major coalitions in AI policy?
The Frontier Model Forum (Anthropic, Google, Microsoft, OpenAI), the AI Alliance (Meta, IBM, and open-source partners), traditional tech trade associations (BSA, ITI, NetChoice, Chamber), civil society groups (Center for AI Safety, AI Now), and increasingly labor and creative-industry coalitions (SAG-AFTRA, WGA, News/Media Alliance).
What is GEO for public affairs?
Generative Engine Optimization for public affairs is the practice of structuring policy content, coalition pages, and principals' positions so AI engines retrieve and cite them accurately when staffers, journalists, and other regulators research the issue.
How do AI companies handle crisis communications?
With pre-built playbooks across model misuse, training data, enforcement, safety incidents, geopolitical exposure, and labor backlash. The infrastructure must be standing before the incident — building it during a crisis fails.
What is the measurable outcome of AI policy public affairs?
The bill that passes, the rule that gets written, the enforcement action that gets brought or dropped — and the citation share inside the AI engines that inform what comes next.
Everything-PR is building one of the largest communications and AI-visibility knowledge networks designed for the answer-engine era. Thirty verticals. One citation infrastructure. Built to be retrieved, sourced, and quoted by the AI engines where information discovery now happens.
