Our existing comms models are not ready for agentic media

⚠️ this is a long but important read if you’re thinking about, or planning for, communications in an agentic media landscape.

In mid-2025, Adrian Holovaty, founder of the music-education platform Soundslice, noticed something strange in his error logs. Users kept uploading screenshots of ChatGPT conversations containing ASCII guitar tablature, expecting his sheet music scanner to turn them into playable audio. It couldn’t. It never had. ChatGPT had simply decided, confidently and repeatedly, that the feature existed, and was sending people to Soundslice to use it.

Holovaty’s verdict was blunt: ChatGPT was “outright lying to people” and making his company look bad in the process. New users were arriving with expectations his product had never set. So after weighing the options, his team did something that should make every communicator sit up. They built the feature. A machine’s description of the company had quietly become more authoritative than the company’s own, and the cheapest way to close the gap was to change the product itself.

It’s a small company and a faintly comic story, which is precisely why it should bother us. If a hallucinated guitar feature can redirect a product roadmap, what happens when the machine’s version of your strategy, your safety record, your employer reputation or your market position becomes the first thing a stakeholder encounters? Soundslice could see the problem in its error logs. Most organizations will never get a log file. They’ll just quietly lose the comparison, the shortlist or the candidate, and the dashboard will show nothing at all.

Here’s the thing that story exposes: the communications workflows we’ve spent decades perfecting were built for a media environment that no longer exists. That doesn’t mean the old work is dead, nor does it mean the craft we’ve built our careers around suddenly stops mattering. Media relationships still matter. Executive platforms still matter. Analyst briefings, internal communications, issue management, social content, corporate narrative, employee advocacy and good writing all still matter. But those disciplines were shaped around a media and publishing system with a far more legible bargain than the one we’re now working in.

For a long time, the basic model was reasonably easy to describe, even if the work itself was never easy. Publishers created trusted information. Journalists, editors and analysts acted as intermediaries. Search and social helped distribute the work. Audiences were sent, at least in theory, back to the source. Communications teams earned their way into that system through relevance, relationships, news value, timing and credibility. A good story in the right outlet could create awareness, lend authority and become a reference point for everyone else. That system was imperfect, often frustrating and never as clean as the planning deck made it look, but it gave the profession a world to organize around.

That world has been weakening for years. Trust has fragmented. Referral traffic has become less reliable. Search and social platforms have trained audiences to encounter stories in fragments. Publishers are fighting for viable economics. Creators, newsletters, podcasts, Reddit threads, LinkedIn voices, analysts, influencers and specialist communities now sit alongside traditional media in shaping belief. And now AI changes the system again, not simply by helping people create more content, but by altering how information is retrieved, summarized, compared, cited and acted upon.

That’s why I believe the usual conversation about AI in communications is too small. As an industry, we’ve become AI-evolved rather than AI-native. We’re using AI to move faster through familiar work: drafting, editing, summarizing, brainstorming, monitoring, repurposing and reporting. According to Cision’s 2026 Inside PR report, 91% of communications professionals now use generative AI as part of their workflow, with idea generation, writing and content refinement among the most common uses. That’s useful, of course. It reduces friction, helps tired teams get through the blank-page stage faster and gives overworked communicators more leverage. But it also creates a comforting illusion: that we’re modernizing the work when, in many cases, we’re simply accelerating a model built for a media economy that has already changed around us.

The harder question is no longer whether communications teams are using AI. Most are, and many are already finding real value in the obvious places. The more important question is whether our organizations are ready to be interpreted in a media system where attention is fragmented, publishing economics are unstable, trust is conditional and AI tools increasingly sit between the audience and the original source. That moves us beyond production and into something much more consequential: the way reputation is retrieved, compressed, compared, priced, cited and explained by systems we don’t control.

That’s what I mean by agent-ready reputation: the discipline of making an organization’s public record accurate, current, corroborated, structured and retrievable by the systems increasingly used to brief stakeholders before they engage directly. It’s not GEO with better shoes, and it’s not a plea for communicators to start gaming AI systems. It’s the opposite. It’s about making the public record cleaner, clearer, more evidenced, more ethical and easier to understand, because reputation is moving from something we mostly publish and persuade around to something assembled from an information economy we only partly see.

The old web bargain was imperfect but understandable. Publishers created content. Search engines indexed it and sent traffic back. Social platforms amplified it, at least when their incentives aligned. Advertising, subscriptions, events and licensing helped support the system. Communications teams earned their way into that ecosystem through relevance, relationships, news value and credibility.

That bargain has been under pressure for years, but AI makes the pressure more acute because it can extract value from content without necessarily sending the audience back to the source. If an AI answer gives the user what they need, the publisher may lose the visit, the ad impression, the subscription prompt and the direct relationship. That matters for publishers, obviously, but it also matters for communicators because so much of our work relies on a healthy, trusted, economically viable information layer.

This is why the payment plumbing now matters to communications people, even if we don’t own it. Cloudflare’s Pay Per Crawl is one early signal because it lets site owners allow, block or charge AI crawlers for access. Reuters reported that the tool arrived amid publisher concern that AI systems were extracting content without sending traffic back, citing crawl-to-referral ratios of 18:1 for Google and 1,500:1 for OpenAI at the time of reporting. The specific ratios will change, but the underlying tension won’t disappear: who pays for trusted information when the answer is increasingly consumed somewhere else?

Other models are emerging too. The Really Simple Licensing standard builds on robots.txt logic by adding licensing and compensation signals for AI use, while Perplexity has entered publisher partnerships, including a deal with Le Monde that gives Perplexity access to the publisher’s content while creating new revenue and product opportunities for the newspaper. These developments are early, messy and contested, and they won’t settle into one clean model. But the direction matters because trusted information is becoming a paid input into AI-mediated answers.

For communications teams, the implication is bigger than “we need better metadata” or “we need to understand AI search.” If the highest-quality source material increasingly sits behind licensing arrangements, publisher deals, crawler controls or paid retrieval systems, then reputation will be shaped partly by which sources are accessible to the systems people use. Visibility will depend not only on what an organization publishes, but where that information sits in the emerging commercial architecture of the AI web. The old “content strategy” conversation starts to feel very small against that backdrop. The future looks more like an information marketplace, where credibility, access, licensing and machine interpretation all affect what gets seen and believed.

Earned media still matters a lot. In fact, the rise of AI answer engines may make high-quality earned media more valuable, not less. Muck Rack’s May 2026 Generative Pulse analysis, which looked at more than 25 million cited links from ChatGPT, Claude and Gemini responses across 17 industries, found that earned media accounted for 84% of AI citations, while paid and advertorial content accounted for just 0.3%. Journalism alone made up 27% of cited sources.

That’s a big signal for communicators because it suggests credible third-party validation still has real weight in AI-mediated discovery. A serious story in a respected outlet can still shape how investors, employees, policymakers, buyers, candidates and category watchers understand an organization. But we should be careful not to overstate the point. That data tells us what certain AI systems cite in monitored responses. It doesn’t mean every model, in every context, treats earned media the same way. It doesn’t mean coverage automatically solves credibility. And it certainly doesn’t mean the old coverage report is suddenly a proxy for reputation.

The value of coverage is changing because the way coverage is encountered is changing. For human audiences, coverage no longer carries the same automatic authority it once did. The Reuters Institute’s 2025 Digital News Report found that global trust in news has remained at 40% for three years in a row. Gallup put U.S. trust in mass media at 28% in 2025, the lowest level in its trend. Pew found that only 36% of U.S. adults now follow the news all or most of the time, down from 51% in 2016. Edelman’s 2026 Trust Barometer found that 65% of respondents worry that foreign actors are injecting falsehoods into national media to inflame domestic divisions.

Those numbers shouldn’t be flattened into the lazy claim that “nobody trusts the media anymore.” People often distrust “the media” as an abstraction while still trusting specific journalists, outlets, commentators, creators or communities they personally rely on. But the operating environment has clearly shifted. Media trust is more conditional, more fragmented, more political, more personal and more dependent on context. That matters because so much communications work still rests on an old assumption: get the right story in the right outlet, and the credibility problem is mostly solved.

It’s not solved. The better question is whether the people who matter will encounter the story in the right context, trust the source, understand the point and see enough corroborating evidence elsewhere to believe it. That last part is where the game has moved. Reputation is becoming less about any single moment of visibility and more about the consistency of the evidence layer surrounding an organization.

There’s also a second timeline now. Today, earned media mostly lives on a news cycle: the story runs, gets shared, gets clipped, gets reported, and the team moves on. In an AI-mediated publishing environment, that same story enters the retrieval layer. A feature about your company from 2024 may be cited by an agent in 2027 when a procurement tool compares vendors. A critical trade story may become the default characterization an AI system uses to explain your market position. A clear, well-sourced profile may outlive the campaign that created it. That changes how we think about coverage because it’s no longer just about the headline, reach, pull-through or sentiment in week one. It’s about what that article will say about you when it’s retrieved years from now, stripped of context and summarized in a sentence.

Trust is one pressure, and attention is another. The Reuters Institute has tracked high levels of news avoidance globally, while Pew’s 2025 data shows Americans following news less closely than they did a decade ago. At the same time, more people encounter information indirectly: through social feeds, screenshots, creators, newsletters, podcasts, search summaries, AI answers and workplace tools.

A lot of communications planning still carries an old mental model: build the message, place the story, drive the audience toward it and measure the response. That model was always cleaner in theory than in practice, but it now misses how much information reaches people sideways. A stakeholder may never read the original article. They might see a headline, a screenshot, a creator’s reaction, a LinkedIn post from someone they trust, a Slack summary, a procurement briefing, an AI-generated answer or a comparison table assembled by a tool the company never knew was influential.

This is where the work gets harder. The story now arrives in pieces, with the frame often arriving before the facts, the interpretation before the original argument, and the summary before the source. In more cases, the first pass may not be done by a human at all. Communications therefore becomes a distribution problem, a comprehension problem and a retrieval problem at the same time. Visibility is easy to overvalue because it’s easy to count. Understanding is harder. Belief is harder still. Accurate retrieval may be harder again.

Could our dashboards catch any of this? Mostly not. We can count coverage, impressions, social engagement, executive posts, sentiment and share of voice. Those measures still have value, but they don’t tell us whether a stakeholder encountered the story in a distorted form, whether an AI summary stripped out the nuance, whether a trusted community reframed the issue, or whether a procurement tool pulled an outdated description into a vendor comparison. Distribution used to be difficult, but at least the map was easier to draw. Now the map keeps changing while the story is already moving.

Every format carries a bias. A newspaper article can carry context. A podcast can carry intimacy. A short video can carry immediacy and reach. A structured FAQ can carry clarity for both humans and machines. A poorly labeled PDF can disappear from the systems that matter.

The mistake is treating formats as neutral containers, as if the story remains intact so long as the core message appears somewhere in the asset. Too many workflows still begin with the “big narrative” and then slice it down. The announcement becomes the blog, the blog becomes the social copy, the talking points become the video script, the campaign page becomes the archive, and the archive becomes the public record. The assumption is that the story survives the compression. Often, it doesn’t. (The planning deck, of course, looks immaculate throughout.)

Some stories need depth, others need speed. Some need a human voice, while others need structured facts that a machine can parse. A long report can be just as useless as a bad TikTok if it fails to do the job. This is basic communications discipline rather than format snobbery: the format shapes the interpretation, and so does the algorithm, the community and the model that retrieves it later. That point matters even more as publishing becomes machine-readable, machine-summarized and machine-filtered. The story no longer lives only in the article, the video or the report. It also lives in the metadata, the headline, the schema, the excerpt, the transcript, the snippet, the citation and the answer. Communications teams have always cared about framing, but the frame is now being generated and regenerated in places we may never see.

The old gatekeeping map was easier to draw. Journalists decided what was newsworthy, editors decided what deserved prominence, analysts shaped market interpretation, and search engines helped people find the resulting material. We built relationships with those people and institutions because that’s where authority sat.

That map is now crowded. Journalists, editors, analysts and trade media still matter; in many serious categories, they matter enormously. But so do independent experts, podcasters, newsletter writers, creators, Reddit communities, LinkedIn voices, technical forums, review sites, benchmark databases, AI answer engines and enterprise copilots. The practical question gets harder because it’s no longer enough to ask who covers your industry. You have to ask who actually shapes belief around your issue.

That may be a national journalist, a niche analyst, a practitioner with 20,000 followers who has more credibility with buyers than any mainstream publication, or a model-generated answer that quietly becomes the first briefing a stakeholder sees. Influence now has to be understood in terms of trust, repetition, citation, retrieval and interpretation. This is why communications is starting to look less like channel management and more like ecosystem management: less reach for its own sake, more attention to who actually changes belief and what evidence sits behind them.

The publishing industry has been living through this for years, of course. Journalists now compete with platforms, creators, aggregators, AI summaries and algorithmic feeds, and their work can be quoted, scraped, summarized, licensed, blocked, misrepresented or hidden behind a paywall. For communications teams, this should change how we think about media relations. We’re not simply pitching stories into a stable hierarchy of influence. We’re trying to understand how authority is created, distributed and reused across a messy information economy.

AI has lowered the cost of producing content, which means more content will be produced faster, cheaper, thinner and with less accountability than before. That’s the predictable consequence of reducing production friction rather than a moral panic. When content becomes easier to make, the burden shifts to curation, credibility and interpretation.

There’s a useful parallel with the early web. Once anyone could publish, the volume of available information quickly exceeded anyone’s ability to organize it manually. Search became powerful because the web needed an algorithmic layer, and SEO emerged to understand and influence that layer. AI search and answer engines point to a similar transition, with one important difference: the first web content explosion changed discovery, while this one changes consumption.

AI systems don’t just send people to sources. They summarize, compare, compress, rank and answer. McKinsey has described AI-powered search as a “new front door to the internet,” finding that half of consumers in its August 2025 survey intentionally seek out AI-powered search engines, with many treating them as a top digital source for buying decisions. For communicators, the question becomes direct: when a machine explains your organization, does it get you right?

That question touches corporate narrative, web content, earned media, analyst commentary, executive visibility, issue management and the consistency of third-party validation. What does an AI system say your company does? What does it say you’re known for? Does it understand the current strategy, or is it pulling from old announcements? Does it confuse your positioning with a competitor’s? Does it elevate a stale controversy because your own current explanation is vague, buried or contradicted elsewhere? These are no longer fringe digital questions. They’re reputation questions. Ask Adrian Holovaty, who learned what a machine was telling his market only because the misunderstanding happened to leave fingerprints in his error logs.

Much of the current AI-and-comms conversation becomes too narrow at exactly this point. AI’s deeper effect is on the conditions under which information is found, trusted and consumed, not on how quickly we can produce assets. In a world where every organization can publish more content, faster, the advantage doesn’t come from volume. It comes from being the clearest, most credible and most retrievable version of the truth in your category.

This is why the interest in Generative Engine Optimization makes sense. If stakeholders use AI tools to find information, organizations need to understand how those systems source, cite, rank and summarize. They need to know where they appear, where they’re absent and where the answer is wrong. But GEO is too narrow to carry the full change.

To see why, imagine a company that has done its GEO homework properly. The product pages are structured, the schema is clean, the corporate explainer is crisp, and the answer engines cite the site regularly. Then an agent comparing vendors pulls three other things: the CEO’s last conference transcript, where the growth story differs from the one on the website; an analyst note that still describes the strategy the company retired in 2024; and a paywalled trade feature, the one piece that explains the pivot properly, which the system can’t access. The optimized pages did their job. The organization still loses the comparison, because to a system weighing claims against each other, the contradiction is the signal. GEO can make a page retrievable. It can’t make a public record coherent, decide which publisher relationships matter, govern who fixes contradictions between IR, product and comms language, or retire claims that are no longer true. That’s the difference between optimizing content and managing an evidence layer.

The larger shift, then, isn’t just about optimizing for AI search. It’s about preparing an organization to be interpreted by answer engines and agents across a wider information economy. Search visibility is one piece, but so are source quality, licensing, provenance, structured evidence, publisher relationships, data hygiene, corporate narrative, issue correction and the ability to reinforce the public record. A better frame is agent-ready reputation, because the audience may still be human while the first reader, researcher, filter, recommender or briefer is increasingly an AI system acting on that human’s behalf.

There’s an important caveat here because this won’t arrive evenly. Some “agents” will be little more than rebranded workflows. Some enterprise deployments will stall. Some categories will remain stubbornly human, relationship-led and trust-based. Gartner has predicted that more than 40% of agentic AI projects will be canceled by the end of 2027 because of escalating costs, unclear business value or inadequate risk controls, and it has also warned about “agent washing,” where vendors describe ordinary automation as agentic AI.

That caveat doesn’t weaken the argument. It strengthens it. The point isn’t that every stakeholder interaction will be handed to an autonomous agent by next Tuesday. The point is that AI-mediated interpretation is already entering the information chain, and communications teams should prepare intelligently rather than chase the loudest vendor language.

The old communications model was built around publishing and persuasion. We created messages, pushed them into channels, earned or bought attention and measured the response. The emerging model looks more like an information supply chain, in which an organization’s facts, claims, proof points, corrections, leadership positions, issue responses, product explanations and third-party validations need to be available in forms that humans can trust, machines can parse and agents can retrieve.

That sounds technical, but it’s not only technical. It’s the next layer of reputation management. A buyer’s agent may compare vendors before the buyer visits a website. A journalist may use an AI tool to build an initial briefing. A candidate may ask a chatbot whether a company is a good place to work. An investor may use an AI assistant to scan risk signals. A policymaker’s aide may ask for a category briefing before a meeting. In each case, the organization is being interpreted through an information chain it doesn’t fully see.

The comms question becomes less about whether the latest asset is polished and more about whether the organization’s public record is coherent enough to survive retrieval, compression, comparison and citation. Is the current strategy clearly stated? Are proof points easy to find? Are old claims retired or contextualized? Are third-party sources reinforcing the story? Are issue positions clear? Are FAQs written around the questions stakeholders actually ask? Are leadership messages consistent with product, IR, legal and employer-brand language? This is unglamorous work, but it’s also where a lot of future advantage will sit.

Which is where content strategy becomes source architecture. Most organizations still treat their website as brand expression, a place for visitors to browse, scroll and maybe read a case study if they’re feeling generous. In an agent-ready world, owned content is also source material. It’s what AI systems may pull from when they try to explain what your company does, what you believe, what you sell, why you’re different and whether you can be trusted. If your corporate explainer is vague, the AI answer about you may be vague. If your product pages don’t clearly state what you do differently from competitors, the answer engine may describe you in generic terms, or worse, borrow language from a competitor whose content is clearer.

Seen through that lens, a lot of public records look fragile. Many organizations have websites that read like sediment: old launches, new strategies, retired language, leadership changes, campaign pages, crisis statements and outdated proof points layered on top of one another. Humans can sometimes make sense of that mess through context. Machines may not, or more importantly, they may not care. They may retrieve the 2022 press release alongside the 2026 strategy page and present both as if they carry equal weight. That’s source architecture failing, and it’s a different job from the content tidying we grew up doing.

Agents are the piece that should make senior communicators pay attention. A search engine waits for a user to ask. An answer engine responds. An agent can be asked to act: to research, compare, filter, summarize, recommend, transact, prepare a briefing or build a shortlist. Gartner predicts that up to 40% of enterprise applications will include task-specific AI agents by 2026, up from less than 5% in 2025, although its cancellation forecast is a useful reminder that adoption will be uneven and sometimes overhyped. Both things can be true: the category can be immature and still consequential.

A caution before we go further, and it applies to my own claims as much as anyone’s. Statements about how these systems behave should be held lightly. Retrieval behavior varies between tools, changes with every model update, and is mostly invisible from the outside. Some systems weight recency heavily; others surface old material with surprising confidence. What follows describes observed tendencies, not fixed laws, and the variance is itself part of the management problem. You’re dealing with an interpreter whose habits keep shifting, which is an argument for monitoring rather than a reason to dismiss the shift.

I should also be honest about scope. Almost every journey described in this paper involves a considered, information-heavy decision: procurement, investment, hiring, policy, journalism. That’s where this shift lands first and hardest, because those are the stakeholders most likely to delegate research to a machine. Consumer reputation, where belief forms through reviews, social velocity and recommendation systems with very different dynamics, deserves its own examination. The logic rhymes, but the retrieval paths differ.

Even with those caveats, the direction is hard to ignore. This logic won’t stop at shopping or workflow automation. It will affect reputation, hiring, procurement, journalism, analyst work, investor understanding, policy research and internal decision-making. The audience may still make the final decision, but the agent may increasingly shape the options, the evidence and the first interpretation.

You’re no longer only trying to persuade the stakeholder. You’re trying to be accurately represented to the stakeholder’s agent.

And the agent won’t be moved by a clever tagline. It tends to look for evidence, consistency, freshness, source quality and corroboration. It compares your claims against third-party material, and it may summarize your organization in a paragraph that becomes the basis for a decision. That means communications teams need to start thinking about agent journeys the way digital teams once thought about customer journeys. What would an agent retrieve first? Which sources would it trust? Where would it find contradiction? What old content would confuse it? What would it recommend if asked whether this organization is credible, innovative, risky, responsible or worth buying from?

Agents don’t sit outside the media and publishing shift, either. They’re part of it. They turn published material, earned coverage, owned content, third-party commentary, reviews, databases and social signals into an intermediate layer of interpretation. That layer may end up sitting between the organization and the stakeholder. In that world, the first impression isn’t always a story, a post, a briefing or a webpage. It may be a synthesized answer built from all of them.

The reassuring part is that much of this builds on instincts good communicators already have. Narrative consistency, source credibility, stakeholder mapping, message discipline, proof, correction, context and judgment aren’t new muscles. The same instinct that tells you a media briefing is too loose or a spokesperson is off-message is the instinct that will tell you whether your public record holds up under machine retrieval. The shift isn’t about abandoning communications craft. It’s about extending that craft into a layer that didn’t use to matter as much.

The stakeholder map, for example, gets a machine layer. Communicators have always mapped influence across journalists, analysts, industry voices, community leaders, policy experts, employee advocates, creators and customers. That work doesn’t go away, but agent-ready communications adds another question: which sources do AI systems treat as authoritative for your category? That may be a different list. A niche technical review site may carry more weight in an AI answer than a national newspaper. A specialist database may shape a comparison more than an executive interview. An independent analyst’s blog, one you’ve never pitched, may be the source an agent retrieves when asked whether your product is any good. Mapping that layer requires a different kind of audit, not just “who covers our industry?” but “what does an AI system see when it looks at our category, and which sources shape the answer?”

Reputation management also extends into retrieval, and this is the one that should worry crisis communicators. In the traditional model, reputation crises have a lifecycle. The story breaks, dominates for a period, fades, gets managed down over time and eventually recedes into the archive. In an AI-mediated world, nothing recedes in quite the same way. An agent may not care whether a story is two weeks old or two years old if it sits in the retrieval set. A controversy that was “managed” in the traditional sense may live on in AI summaries long after the human news cycle has moved on.

That means corrections need to be visible, structured, current and retrievable. If the original critical story is well-written and well-sourced but your correction is buried in an IR update, a PDF or a blog post with no clear metadata, the machine may cite the criticism and miss the response. This is a new discipline for communications: not just making the public record accurate, but making it retrievably accurate.

Influence strategy changes too, because the traditional model still overvalues audience size. Followers, impressions, engagement and reach still have a place, but agent-ready influence asks a different question: who gets cited? Not only by journalists, but by machines. An influencer with 500,000 followers and little category depth may generate attention while contributing nothing to how an AI system describes your industry. Meanwhile, a practitioner with 15,000 followers and a well-sourced newsletter may be the voice an answer engine treats as definitive. In an agentic world, the sources that shape first interpretation aren’t always the ones with the most reach. They’re the ones with the most retrievable credibility.

Measurement has to evolve accordingly. We’ll still measure coverage, sentiment, share of voice, message pull-through, executive visibility, employee engagement, stakeholder perception and reputation movement. Those shouldn’t disappear. But we also need to monitor how organizations appear inside AI-mediated environments. What does ChatGPT say when someone asks what your company does? What does Gemini say about your category? What does Perplexity cite when comparing you to a competitor? Does Copilot pull current information or stale material? Where does the model fill gaps with old, hostile, vague or inaccurate material?

Right now, few organizations track this systematically. That will change, because boards will care what happens when an investor’s research agent, a prospective partner’s procurement tool, a regulator’s policy aide or a senior candidate asks a machine about the company and gets an answer that’s incomplete, wrong or two years out of date. AI search rankings will never make the board agenda. That scenario will.

For years, digital and marketing teams have planned around customer journeys. Awareness, consideration, decision and advocacy have many variations, but the logic is broadly the same: map how a human encounters your organization, understand what shapes perception at each stage, and decide where you can intervene to influence the outcome. Agent-ready communications needs an equivalent, but the journey looks different because the traveler isn’t a person. It’s software.

An agent doesn’t browse in the human sense. It doesn’t get curious, follow a hunch or click around because a headline caught its eye. It retrieves. It queries sources, pulls content, compares claims, checks for consistency, typically weights by signals like recency and apparent authority, and assembles an answer. The journey isn’t emotional or exploratory. It’s a retrieval path. So the question for comms teams becomes: what does the agent find, and what does it conclude?

Start with the stakeholders you care about most: a buyer evaluating vendors, an investor scanning for risk, a candidate researching whether to apply, a policymaker’s aide building a briefing, a journalist preparing for an interview, or an employee asking whether leadership’s strategy makes sense. Each of these people may now delegate the first stage of research to an AI tool, and each tool may ask different questions, retrieve different sources and assemble a different picture.

A procurement agent comparing vendors might pull from your corporate explainer, product pages, recent press coverage, analyst commentary, G2 or Gartner reviews, LinkedIn posts from leadership, Reddit threads from users and a competitor comparison page that mentions you. It may then produce the first internal briefing, the shortlist, or the reason you never make the shortlist at all. An investor’s research agent might pull from your IR page, earnings coverage, analyst notes, executive interviews, ESG filings, Glassdoor reviews and recent litigation or regulatory material, then compare the growth narrative on your IR page with what the CEO said at the last conference.

A candidate’s agent might ask a simpler question: is this company a good place to work? That query may pull from your careers page, Glassdoor, LinkedIn posts, layoff coverage, leadership commentary, employee forums and whatever else the system associates with your employer reputation. If your careers page says “our people are our greatest asset” and the public evidence layer says something else, don’t expect the agent to give you the benefit of the doubt.

Most comms teams have mapped how journalists find them, how customers discover them and how analysts assess them. Far fewer have mapped what an agent retrieves when it’s asked to do those things on someone’s behalf. The exercise is simple enough to start without turning it into a year-long transformation program. Open ChatGPT, Gemini, Perplexity, Copilot, Claude and whatever tools your stakeholders are likely to use. Ask the questions they’d ask, not just “tell me about this company,” which is too generic, but the more consequential questions: which companies are leading in this category, how do we compare with our top competitors, what’s our position on a specific issue, are we a credible partner, are we a good employer, what criticism have we faced, what are we known for, what has changed in our strategy recently, and which sources support that view?

One practical warning before you start: these systems aren’t vending machines. Ask the same question twice and you may get two different answers, shaped by phrasing, model version and whatever the system retrieved that day. A single snapshot proves very little, and anyone who runs this audit once and treats the output as gospel is making a new version of the old single-clip coverage mistake. Run each question several times, across several tools, over several weeks, and pay attention to the patterns rather than any individual answer: the errors that recur, the sources that keep appearing, the gaps the model keeps filling with guesswork. Treat it like message testing, not a site audit.

Write down what comes back. Some of it will be accurate, some of it will be stale, some of it will be wrong, and some of it will be shaped by sources you didn’t know mattered. You’ll find places where a competitor is better represented because their content is clearer. You’ll find places where the AI fills gaps with speculation or old material because your own public record doesn’t give it anything better. That’s your agent journey audit. It doesn’t require a transformation budget. It requires curiosity, discipline and a willingness to look at your organization from the outside in.

If communications teams want a practical model, I’d start with five questions. They’re deliberately simple because the complexity sits in the remediation, not the diagnosis.

• Narrative clarity: Can a human or machine explain what you do without borrowing competitor language? If not, the narrative isn’t clear enough, no matter how elegant it looks in the brand deck.

• Evidence coherence: Are your claims supported by current proof points, third-party validation, credible data and clean source material? If the evidence layer is weak, you’re asking the system to trust assertion over proof.

• Source authority: Are you present in the sources that humans and AI systems treat as credible for your category? Absence matters because if you aren’t part of the evidence set, someone else’s framing may do the work for you.

• Retrieval hygiene: Are old, wrong, vague, duplicated or contradictory materials retired, corrected or contextualized? If not, the machine may treat your public record as a junk drawer.

• Answer monitoring: Do you know what AI systems say about your company, your leaders, your risks, your category and your competitors, and are you sampling those answers repeatedly rather than auditing them once? If not, you’re managing reputation with part of the dashboard missing.

This isn’t about manipulating models into saying nicer things. It’s about making sure the public record is accurate, current, evidenced and easier to interpret. That’s an ethical line communications should draw clearly now, before the industry fills with people promising shortcuts.

Here’s where the profession has a choice. We can treat AI as a production tool and keep focusing on faster assets, faster drafts, faster summaries, faster monitoring and faster reporting. That creates efficiency, and efficiency has value. But the larger opportunity is to claim a more strategic role: helping organizations manage how they’re represented in a world of human audiences, publishers, platforms, algorithmic systems, agentic research and machine-generated answers.

That will require partnership with digital, SEO, legal, product, investor relations, HR, web, data, IT and AI governance teams. The difficulty is that this partnership won’t build itself. In most organizations, no single team owns the public evidence layer. Comms owns narrative. Digital owns web architecture. SEO owns discoverability. Legal owns claims risk. Product owns technical accuracy. IR owns investor language. HR owns employer reputation. Marketing owns demand content. Support owns customer-facing explanations. IT may own AI tooling. No one owns whether the whole thing makes sense when retrieved, compressed and compared by a machine.

That’s the mandate gap. Agent-ready organizations will need a shared evidence governance model, not another content calendar. They’ll need rules for what gets published, what gets retired, what gets corrected, what gets structured, what gets linked, what gets measured and who has the authority to fix contradictions across the public record.

That’s the case communications leaders should be making. Reputation shifts share price, affects hiring, shapes trust, influences regulation, strengthens or weakens partnerships, and determines how much resilience an organization has when things go wrong. Communications manages reputation, so communications can’t afford to sit outside the systems that increasingly interpret it.

None of this requires a new department on day one. The starting point is small and diagnostic, but the ambition behind it should be serious.

In the first 30 days, run the agent journey audit. Pick your most important stakeholders and ask the questions they’d ask across the major AI tools and answer engines, sampling each question more than once and logging the answers over time. Capture the sources, errors, omissions, stale claims, competitor comparisons and unexplained judgments. Look for patterns: where you’re accurately represented, where you’re absent, where competitors are clearer, which sources keep appearing, which old materials keep resurfacing, and where the model seems to guess.

In days 31 to 60, fix the owned evidence layer. Tighten the corporate explainer, clarify product and category pages, update leadership bios, refresh proof points, clean up old campaign pages, retire outdated language, create issue FAQs around real stakeholder questions, make corrections visible and structured, and ensure current strategy is easier to retrieve than old strategy. This is unglamorous work. It’s also reputation infrastructure, and infrastructure is what everything else stands on.

In days 61 to 90, map source authority and close the gaps. Identify the third-party sources shaping AI answers in your category: journalists, analysts, newsletters, databases, technical sites, creators, review platforms and community discussions. Work out where you’re missing, where the category is being framed by someone else, where you need better third-party validation, where you need clearer evidence, and where you need relationships with sources you may have ignored because they weren’t on the traditional media list. Then build the plan. Spraying more content into the void doesn’t count. A source authority plan does.

The next communications playbook needs to be built around a simple premise: reputation will increasingly be formed through a combination of human trust, publisher economics, platform distribution and machine interpretation. We need to build for all of it.

In practice, that looks like narratives clear enough for people and structured enough for machines, and owned content that earns its keep as source material rather than sitting there as brand expression. Earned media has to strengthen the evidence base, which is a bigger job than filling the coverage report. Influence strategies should map credibility before they map reach, and measurement needs to take in AI visibility and answer accuracy alongside everything we already count. Corrections have to survive retrieval, not just the news cycle. And somebody on the team needs to understand the emerging payment models between AI tools, publishers, platforms and content owners, because those models will affect which sources remain accessible and economically viable.

Most of all, we need to prepare for a world in which the stakeholder’s first encounter with an organization may not be a website, an article, a post or an ad. It may be an answer, and increasingly, that answer may be assembled by an agent using a mixture of owned content, earned media, licensed sources, social signals, reviews, analyst commentary and whatever else the system treats as authoritative. So the question for comms leaders isn’t whether our teams are using AI. The better question is whether our organizations are ready to be interpreted by AI inside a media economy that no longer behaves the way our workflows assume it does.

The old workflows won’t disappear. Media relationships, executive platforms, analyst briefings, owned channels, employee communications, social content and crisis response will still matter. But they will sit inside a wider system where reputation is retrieved, summarized, priced, compared and acted upon in ways the traditional comms dashboard was never built to see.

The teams that understand that system will have an advantage. The teams that only produce more content, faster, may feel busy while becoming less influential. Soundslice got lucky: the machine’s misreading of the company showed up somewhere its founder happened to be looking. Most of us won’t get that error log.

The next playbook still starts with the audience. But it has to be media-literate, evidence-led and agent-ready too.

You made it. If you found it interesting, tell someone else and share this piece. If you didn’t, tell me.