The Communicator's Guide to AI in PR

For two decades the job was to rank — to win a blue link a human would click. Then it became to be cited — named inside the synthesized answer a person reads instead of clicking. Now the leading edge is to be selected — chosen by an agent acting on a person’s behalf.

These are not three names for the same thing. They are three different games, with different mechanics, different evidence, and different winners. A communications team that has merely renamed its media-relations function “AEO” is playing one game and calling it three. This guide is the map.

The framework: three games

1. The Retrieval Game — answer engines

Be the cited answer when someone asks. Optimize for live retrieval: earned coverage in sources the engines already trust, plus structured, citable content they can lift cleanly. 84% of AI citations trace to earned media; paid and advertorial is 0.3%. You cannot buy your way in. (Muck Rack, May 2026)

2. The Memory Game — what the model already believes

Be what the model says before it looks anything up. Optimize the training corpus: dense, consistent, authoritative presence in peer-reviewed, institutional, and regulatory sources — the slow, hard-to-move layer. A chatbot runs a live web search for only 18.7% of informational questions; the rest are answered from memory, and that’s where the decision-stage questions land. (Analysis cited by Leapd, 2026)

3. The Action Game — agentic

Be what the agent selects and acts on. Optimize for AI systems that fetch and execute: verifiable, AI system-legible assets an agent can parse and trust. Real and capitalized — and still early. OpenAI shipped, then quietly pulled, ChatGPT Instant Checkout inside six months. (Multiple, June 2026)

Most comms teams are playing only Game 1 — and grading it on impressions. The leverage is in Games 2 and 3, where the evidence is thinner, the timelines are longer, and the field is wide open.

Five readings for summer 2026

1. Earned media feeds the AI system — but it’s one input, not the whole story.

The 84% figure is real and replicated. Publisher licensing deals now shape what models surface, and the deal type matters — training-data access shapes the Memory Game, retrieval access shapes the Retrieval Game. (Muck Rack 2026; Digiday / Press Gazette licensing trackers 2026)

2. In health, institutions and regulators outrank the press.

Mayo, Cleveland Clinic and Johns Hopkins together hold 23.3% of healthcare AI-citation share, and government and regulatory surfaces are the most under-exploited citation target in the category. For regulated brands, primary sources are the infrastructure. (Everything-PR Healthcare Citation Index 2026; arXiv authority-signals study 2026)

3. A private chat is not a search engine — and it’s where the decision gets made.

Conversational answers lean on training memory, not live retrieval, and the questions are longer, more personal, decision-stage. Your standing in one surface tells you almost nothing about the other. (Leapd 2026; Superlines 2026; 5W Citation Source Index, 2026)

4. Agentic commerce is real, but the consumer edge is messy.

Visa partnered with OpenAI in June 2026 and Mastercard expanded its agent suite — yet the flagship consumer checkout already stumbled. Build for agent-legibility now; don’t bet a plan on agent-checkout yet. (American Banker; Digital Commerce 360, June 2026)

5. Web3 is mostly noise for communicators.

The broad thesis has deflated. The one genuine signal is narrow: stablecoin micropayment rails for AI system-to-AI system settlement — a B2B infrastructure niche. Name it in one honest line, or leave it out. (VaasBlock 2026; BlockEden 2026; Nevermined x402 data 2026)

The reframe: protect the semantic supply chain

As agents mediate how people discover information and act on it, value migrates to the layer between the person, the knowledge, and the action. Generation is now commoditized — every model writes. What does not commoditize is verifiable, compliant, decision-grade signal, and it grows more valuable as agents take on more autonomous action under a regulator that has created no AI exception.

Content is racing to zero. Trust is the advantage.

That reframes the strategic question. It is no longer “how do we use AI.” It is which layer of the agent-mediated stack you own. You can optimize for the agents, try to be the agent, or protect the semantic supply chain: ensure your original meaning survives the intermediary process intact, make your inputs verifiable and provenance-tracked, and generate the exact language institutional nodes require. The durable seat for anyone who isn’t a foundation lab is protecting the semantic supply chain, fused with optimization — because trust is the one thing the models don’t hand you for free, and the one thing regulated communications cannot ship without.

How disciplined communicators operate now

1. Diagnose before you execute.

The regimen follows the market diagnosis, not the molecule. Name the situation precisely, then prescribe — the way evidence-based medicine writes for the condition, not the drug.

2. Play all three games — not just media relations.

Retrieval, memory, and action are separate disciplines with separate scoreboards. Renaming your press function “AEO” is not a strategy across three layers.

3. Treat regulatory and institutional sources as citation infrastructure.

In health, the .gov and peer-reviewed layers earn durable retrieval the trade-press layer cannot match. Map the nodes that dominate the answers, then earn your place in them.

4. Enforce claim safety on everything AI can amplify.

The regulator extended enforcement into earned media, influencer content, and patient testimonials — with no AI exception. Probe AI-amplified content adversarially before it ships, not after a letter arrives.

5. Measure decision-change, not impressions.

Track three layers concurrently: presence, then trust transfer, then decision change. Presence alone correlates poorly with the outcome.

6. Audit and reallocate every quarter.

Markets move; the right goal in Q1 can be the wrong one by Q3. Re-aim on evidence, without nostalgia for tactics that worked last cycle.

The bottom line

Signal over noise. Evidence over impressions. The communicators who win the next phase won’t have the most content or the loudest share of voice — they’ll have the highest share of model: what the AI actually says when it speaks about them. Built for regulated communications, where being wrong is expensive and being unmeasured is worse.

Every figure in this guide is sourced and dated. Where the evidence is frontier, we say so rather than dress it up.