The Fractals Agentic PR & Communications Playbook: What we know to be true in 2026

The placement stopped working

PR has always been priced on one unit of output: the placement. Get the client onto a page a journalist wrote, because the page was where reputation got made. Two things broke that at once, and most of the industry is only watching one of them.

The channel is being replaced

Media relations meant getting your client's story onto a publisher's surface — a newspaper page, a homepage, a search result. Those surfaces are now being replaced by agent-mediated answers. When a journalist uses an agent to research a story, or a stakeholder asks an agent "what should I know about Company X", the old distribution chain of pitch → journalist → article → reader breaks at every step. The journalist may never read your pitch, the reader may never read the article, and the agent builds its answer from whatever sources it can reach. You have no visibility into whether your client is in it.

The journalist's inbox is now filtered by AI

Journalists use AI to filter and triage their pitches. The release you spent hours on may be read by an agent, compressed into one sentence and ranked against 200 others, with the journalist seeing only the top five. Every other PR professional is using AI to send more pitches than ever, so AI-written pitches meet an AI filter and human attention sits in the middle as the scarce resource.

Agents read publishers, and earned media is what they cite

Muck Rack's analysis of more than 25 million AI citations across ChatGPT, Claude, and Gemini found that 84% of citations draw on earned media — journalism, academic research, government sources, encyclopedic sites, and third-party corporate content. Paid and advertorial content accounts for 0.3%. That 84% share has held in a band between 82% and 89% across all three editions of the study since July 2025 — a pattern, not a snapshot.

Cloudflare's network data, drawn from more than 20% of the web, shows AI-related crawling at roughly half of all crawler requests, up from 22% in spring 2025. When an agent answers a question about a company, an industry or a product, it draws on whatever has been crawled and indexed most heavily, and that is overwhelmingly publisher content. A release on a wire and a post on the company's own blog are not in that pool. Your client's owned content stays invisible to AI until a publisher has carried it.

The measurement went with the placement

PR measurement was built on proxies: impressions, reach, share of voice, sentiment. In an agent-mediated world those proxies break, because the article may have been read by nobody and still cited by an agent answering 50,000 queries. Google complicates it further with a mixed-use crawler that combines search and AI access in one bot, which gives Google roughly twice the access to publisher content that other AI companies have and leaves site owners unable to tell which purpose the access serves. The metric that holds up is citation density: how often your client appears when an agent answers a question in their domain.

The channel, the craft and the scoreboard all moved at once. Here is the same function before and after:

One caveat on "evergreen". Muck Rack found that more than half of journalism citations point to articles published within the past 12 months, and citation volume drops sharply after the first six. Freshness is part of what gets you cited. Evergreen applies to the structured page you keep updating on a cadence, not to a single coverage hit that ages out.

Value moves up

Production got cheap. Judgment did not

AI can draft the release, build the media list, personalize the pitch, monitor coverage and write the measurement report. The production layer of PR — the mechanical work junior staff used to do — has collapsed in cost. It cannot decide which story is worth telling, who it actually matters to, or why now. What a client buys from a PR professional has moved from "can you write a press release" to "can you find the one story that changes how stakeholders think about this company".

The retainer stops being defensible

A fee priced on hours and output volume was defensible while production cost something. The client can now see that it costs nothing, so the retainer has to be re-priced against something else. What is left is strategic: which narrative to advance, how to structure content for agent citation, when to answer a reputational threat, and what to measure. That is a smaller volume of work and a much larger decision, which is a different commercial shape and a different conversation.

The junior generalist is already gone

The entry-level role that built media lists, drafted releases and compiled coverage reports is being automated, and the industry's own town halls now say so out loud. What survives is the person who can direct the systems, judge the output and add strategic judgment. The spec-writer replaces the drafter and the strategist replaces the coordinator.

Two policy shifts are moving the ground under this. Cloudflare's September 15, 2026 change alters the crawl economics the arguments above rest on: "mixed-use" crawlers, the ones used for search and AI together, are blocked from ad-supported pages unless the site owner opts in. That pushes AI companies toward licensing deals and raises the value of structured, owned content. And the UK's Competition and Markets Authority is pressing Google to separate its search and AI crawlers, after Cloudflare's data showed the combined crawler reaching roughly twice as much publisher content as its AI competitors. Both push the same way: the domain a company fully controls becomes the access point that matters.

The three things that decide your client's seat

This industry supplies the single most valuable input the machines cite, and it is still billing for that input by the hour. The market buying it is open and already priced, which means the terms are being set now. Three things decide whether your client is at the table.

1. Structure the client's content so a machine can find and cite it

Every piece of PR content falls into one of three kinds.

Answer-content. Structured information that answers the recurring questions agents ask about your client's industry: industry data, regulatory explainers, cost comparisons, specification guides. Built to be extracted and cited. Example: a logistics company publishing a structured comparison of APAC shipping costs by route and mode, updated quarterly and machine-readable, so it is the answer when an agent is asked "what does it cost to ship from Singapore to Jakarta?"

Signal-content. The delta, as it happens: what changed and what it means. Earnings, product launches, leadership changes, regulatory decisions. Timestamped and attributed, for agents that need to know what happened today. Example: a financial services firm publishing a same-day analysis of every central bank rate decision, structured with the old rate, the new rate, the reason and the market read.

Context-content. Strategic perspective an agent cannot synthesize because it depends on judgment, relationships and lived experience. The CEO's annual letter. The white paper that names an industry shift before anyone else does. The point of view that makes a model link to you rather than summarize you, because it knows it cannot reproduce the judgment.

The 84% figure is the tell: agents cite publishers, not press releases. So the most valuable thing a communications team can do is get the client's structured content onto a domain agents already trust, and increasingly onto the client's own domain structured well enough to be cited directly. The pitch that works is "can we build a data product together that your readers and AI agents will both cite", rather than "can you cover our announcement".

2. Publish it in more than one mode

Brian Morrissey of The Rebooting named this in October 2026: "Liquid publishing is here. Sometimes called headless content, the new approach to media is to be multimodal and have a business model that doesn't depend on people coming to you."

For a communications team that means one announcement becomes several artefacts rather than one. The structured page on the client's own domain. The data as a table a journalist can lift and a machine can parse. A plain-text version for anything that asks for markdown rather than HTML. An executive's bylined analysis placed with a publisher who already has the agent's trust. A short item for the newsletter and the feed. None of these is a summary of another, and each carries the same attribution.

The point is that the reader may arrive as a person or as an agent, and the work should be legible to both without being rewritten for either.

3. Move the fee off output volume

If the output got cheap, pricing on output is pricing the part that no longer holds value. The publishers who are being paid are the ones who restructured for the machine before the machines came asking: Cloudflare's chief strategy officer describes it as publishers using bot controls to "create reliable scarcity for their content, and then negotiate better deals".

The buyers are already in the market. Axios announced Axios Direct in September 2026, three feeds selling its newsroom's reporting to machine customers, client-direct rather than through a marketplace. Apple has offered publishers payment per use for Siri news against a nine-figure budget. Google has been paying about 100 publishers for the pages that shape its AI answers, on a "demonstrated value" basis with no published method — and the spread is the warning, with one publisher earning more than a million dollars a year against most earning under $1,000, which several said is less than 0.1% of their ad revenue.

That distribution is the whole stake. Arriving early is worth more than arriving at all, and a communications team that can put a client on the early side of it is selling something no amount of pitching volume can produce.

What to do on Monday

Audit your client's cite-ability

Start with the test that costs nothing and tells you the most.

• 1. Run the agent test on your client's existing content Feed domain-relevant queries to GPT, Claude and Gemini, and see whether the client appears and where. A client that is invisible to agents for queries in its own domain has a structural problem, and no amount of pitching fixes it
• 2. Build a structured content layer for every client Every client needs answer-content on its own domain: the three to five questions agents ask about the industry, each answered with structured content that is attributed and kept current. Treat it as citation infrastructure rather than a blog
• 3. Shift from pitching stories to building publisher partnerships The 84% figure is the tell: agents cite publishers, not press releases. The most valuable thing you can do is get your client's structured content onto a domain agents already trust. The pitch that works is "can we build a data product together that your readers and AI agents will both cite", rather than "can you cover our announcement"
• 4. Separate the bot-readable from the human-readable The structured page and the narrative page are different artefacts with different jobs, and one document cannot do both well. Publish the structured version on the client's own domain where they control it, and let the narrative carry voice for the reader
• 5. Build your own AI-native workflow Stop editing AI output word by word and write specifications instead: the five things this release must communicate, the three it must avoid, and the one action the reader should take. A PR professional working at level 3, directing and approving, or level 4, writing scenarios and checking outcomes, will outproduce an agency still at level 2 by three to five times per head. These are the levels of AI-native work, from the agentic media playbook
• 6. Develop a point of view for your agency or function Every agency and in-house team needs a public point of view about where its industry is going. That means a specific, falsifiable claim about the future of communications, not "we get results". A point of view is what makes clients choose you and what makes agents cite you
• 7. Train your team on judgment, not tools The most valuable skill in the PR function of 2027 is judgment rather than prompt engineering: knowing which story matters, to whom, and why now. Train your team to judge AI output against strategic criteria instead of producing it. The spec-writer replaces the drafter and the strategist replaces the coordinator

Then measure citations, not clips

Stop measuring: impressions, reach, AVEs, share of voice, sentiment scores.

Start measuring:

Citation rate. How often do GPT, Claude and Gemini cite your client for queries in their domain? Track it monthly, and segment by platform and query type, because the three engines cite at different rates and an aggregate number hides where the gap sits. ChatGPT cites sources in 96% of responses, averaging five; Gemini in 82% averaging eight; Claude in 55% but with thirteen sources when it cites at all. Each also draws from a different pool — Wikipedia for ChatGPT, PubMed Central for Claude, Reddit for Gemini.

Citation position. When cited, is your client the primary source or one among many? First-cited is worth more than fifth-cited.

Citation accuracy. When cited, are the claims about your client correct? Incorrect citations are the new misreporting, and they are harder to detect and correct.

Publisher domain authority for agent queries. Which publishers do agents trust most for queries in your client's space? This replaces the media list as the core targeting asset.

Content actionability score. For every piece of owned content: can an agent infer a specific action from it? Target: above 70%.

Spec-to-output ratio. How much time does your team spend writing specifications versus reviewing AI output? If the ratio is below 1:3, you are still operating at level 2 — AI drafts, human rewrites — instead of directing and approving. This ratio is the fastest way to see which level your team actually works at.