Why and How PR Pros Should Adapt Their Content Strategy for AI

About 18 months ago, in a PR strategy meeting with a client, a new member of the corporate communications staff had what she thought was an epiphany: “We need to raise our profile in the LLMs.” And to do so, she reasoned, “we must generate more branded news coverage.”

Hmmm, yes and no, but not so fast.

The playbook that defined public relations for decades is being rewritten in real time. While traditional earned media relations remains valuable, a seismic shift is underway — one that requires PR professionals to fundamentally rethink how they create, distribute, and measure content in an age when AI algorithms, and LLMs in particular, increasingly serve as the primary information brokers between brands and consumers. We’re talking AEO and GEO: Answer Engine Optimization and Generative Engine Optimization.

A recent Muck Rack study analyzing over one million AI-generated citations reveals something both surprising and sobering: what your brand stands for matters less than whether AI systems choose to amplify it. Citations profoundly shape AI responses — enable them, and the answer changes entirely. AI systems are making editorial decisions about what to reference, and those decisions determine whether your company is part of the conversation or not.

The Shift in Web Physics

The old internet physics — search engines visit your site and send visitors in return — is collapsing. On a recent visit to New York, my Mom’s 80-year-old cousin, visiting from Cincinnati, kept pulling out her iPhone to ask ChatGPT questions.

Seth Besmertnik, CEO of Conductor, explained to me and to those at the C3 2025 conference that the economics have inverted. ChatGPT crawls websites roughly 60,000 times and sends back just one visitor. Anthropic bots crawl approximately 1,500 times per single referral. The implication: website traffic has become the wrong success metric.

As Catherine Perloff reported in The Information, this is why companies from Mexican hotel chain Tafer Hotels & Resorts to Fortune 500 firms are seeing 25% traffic declines, without corresponding revenue hits (for now). Hotel executives now recognize that visitors arriving through AI systems spend more time on sites and are more likely to complete purchases than casual Google searchers. Traffic quality, not volume, is what matters.

The real opportunity lies in content impressions within AI and search interfaces. As Besmertnik put it: “Every time someone is looking for something, a little window in their mind opens — an opportunity for organizations to inject information, wisdom, knowledge.” The question is no longer “How do we drive traffic to our website?” It’s “How do we ensure our content is cited as authoritative when AI answers questions about our industry?”

“There’s not enough content for AI because the way people use AI is very different than how they use Google.” Users don’t search for “best running shoes.” They ask: “What’s the best running shoe for someone with flat feet who can’t wear Nikes?” That specificity demands fundamentally different content architecture.

Reverse-Engineer the Algorithm

Many PR pros don’t realize that you can actually ask ChatGPT to show you its algorithm. Eric Schwartzman, a New York-based digital marketing consultant specializing in AI optimization, explained in a conversation that the process is pretty straightforward.

When a client wasn’t appearing in ChatGPT’s answers and his competitors were, Schwartzman asked the AI platform directly: “What criteria did you use to rank competitors in this vertical?” ChatGPT provided specific weightings: 25% board certification, 20% review quality and quantity, 15% experience and case volume, and so on. Schwartzman then examined top-ranked competitors and found their homepages contained all these elements — using those exact words.

The insight: LLMs aren’t mysterious black boxes. They’re scraping content voraciously for structural patterns and topical signals. Your job is to feed them what they’re looking for.

Schwartzman recommends starting with schema markup, the standardized metadata language at schema.org. “LLMs are scraping schema,” he noted. “That’s where they’re getting everything right now.” Create an FAQ on your website — ask ChatGPT to generate common questions about your industry, then answer them comprehensively. Publish that FAQ with schema markup in the background. This makes your content machine-readable and optimizable.

The strategy involves three types of keywords: variations (how people search differently for the same concept), entities (Wikipedia-recognized topics), and LSI keywords (latent semantic indexing — thematically related terms). Use these strategically across your owned content.

Journalism Still Rules

Here’s where traditional PR meets new reality. The Muck Rack study found that over 95% of links cited by AI systems are from non-paid media. More specifically, 27% of AI citations come from journalistic content. For opinion-based or recency-focused prompts, that number jumps to 49%.

This is music to PR practitioners’ ears — if they adapt. It suggests that the core skill of PR — securing earned media coverage — remains crucial. But there’s a catch. Not all earned media is created equal in the eyes of AI systems. High-authority outlets like Reuters, Axios, and the Financial Times appear frequently in citations, but consistency varies significantly by industry.

The implication: blanket media placement strategies won’t cut it. PR teams must identify which publications AI systems actually prioritize for their specific industry and topic.

Building the Content Pyramid

Sarah Evans’ “Inside-Out Story” framework remains a strategic backbone: owned media first, then earned, then AI visibility. Start with long-form anchor articles on your website structured for AI consumption. But Schwartzman adds critical specificity: this content must reside on your owned properties and then be amplified through all available channels.

Evans emphasizes the importance of framing headlines to reflect real prompts: “How can global brands prepare for AI-driven hiring?” rather than abstract positioning. Each anchor article should be supported by schema markup — FAQ and how-to formats are particularly valuable — and internally linked to supporting pages and glossary content.

Besmertnik’s research shows the most successful brands have content “widely cited by AI platforms and referenced in thousands of prompts.” Penn Medicine focused on lower-funnel content that AI picked up. LabCorp created proprietary research that became widely cited. Their success came from producing content at scale — not just one landing page, but dozens of pieces addressing granular personas and decision stages.

The distribution strategy is holistic. Schwartzman emphasized: “Before I spend time inviting people over, I want to clean up my house first.” Optimize owned media. Then build roads back to that content through social platforms affiliated with AI systems. Reddit and Wikipedia have favored status with LLMs. LinkedIn is affiliated with Microsoft and ChatGPT. Facebook data feeds Zuckerberg’s AI models. The principle: introduce information onto platforms where the LLMs you’re targeting actually source training data.

Focus on the Dominant Platforms

Ignore the noise about Claude, Grok, and Perplexity for now. As Besmertnik stated bluntly: “If you’re a brand, you should not be measuring or caring about how you show up in Grok or Claude. What you should care about is Google and ChatGPT.” These two command 99% of market share for consumer decision-making. That’s where your resources should flow.

This doesn’t diminish other models’ value for creative or utility purposes, but it clarifies where PR investment should flow. While your content team might use Claude to draft communications or Perplexity to research industry trends, your visibility strategy must prioritize the platforms where actual customers make decisions.

That said, the most dominant platform, ChatGPT, just changed the way it recognizes and recommends brands, as Profound’s Josh Blyskal noted in his thoughtful LinkedIn post about the structural change that took place on October 18. He wrote:

We pulled millions of ChatGPT responses, [sic] here’s what’s going on:

1. The avg. number of brands mentioned per response dropped from 6 to 3.5

2. At the same time, average brand visibility fell 31%

3. The competitive gap between 1st and 10th compressed by 23%.

The New Metrics

Besmertnik concluded his C3 presentation with a critical insight: “Priorities are measured by where you put money, time, effort.” Organizations winning at AI visibility aren’t tinkering incrementally. They’re reallocating budgets toward content production and distribution.

LabCorp shifted reporting from traditional KPIs (impressions, visitors) to: mentions, citations, and market share for key topics in AI search.

This represents a fundamental PR principle finally gaining leverage: your company doesn’t own the narrative, but it can own the authority and wisdom in your space. And when AI systems look for that wisdom, they should find you.

The New Gatekeepers was originally published in Adventures in Consumer Technology on Medium, where people are continuing the conversation by highlighting and responding to this story.

In 2016, I wrote a post building on Ben Thompson’s Aggregation Theory to explain how tech giants like Amazon, Facebook, and Uber became the internet’s front door, owning the interface, not the supply. These platforms thrived by removing friction, offering smooth user experiences, and capturing attention at scale.

But the landscape is changing. AI won’t just give us answers, it may soon act on our behalf. That shift demands a rethink of how platforms aggregate power. Enter Aggregation 2.0.

Aggregation 1.0: The Distribution Flip

In Aggregation 1.0, control came from owning distribution. Whoever controlled shelf space, airtime, or storefronts decided what people saw, and bought.

The internet changed that. Suddenly, anyone could reach everyone.

Power shifted to those who owned the interface, the digital places where people searched, shopped, and connected. These platforms made suppliers interchangeable, attracted demand, and scaled rapidly.

Ben Thompson’s original Aggregation Theory explained this shift:

• Suppliers lost leverage as access became universal
• Distributors became replaceable
• Aggregators controlled demand and built dominant positions

Google, Amazon, Facebook, Netflix, and Airbnb didn’t just organize the web, they became its default gateways.

The Fracture: What’s Changing

Aggregation 1.0 rested on three pillars:

1. The internet removed distribution barriers
2. Software made discovery easier
3. Platforms captured attention and intent

But three major shifts, accelerated by advances in AI, are redrawing the map:

1. AI as the Interface

Where platforms once competed to be the interface, AI may soon replace it. Instead of clicking through menus or search results, we’ll delegate goals: “book me a trip,” “lower my bills,” “keep me healthy.”

The interaction shifts from browsing to outcomes. The interface fades into the background.

2. Trust as the Bottleneck

Delegation requires more than smart systems, it requires trust. Systems must understand our goals, explain their reasoning, allow for opt-in control, and demonstrate alignment through transparent and auditable behavior. Trust isn’t an add-on, it’s the product.

3. Context as the Moat

In a world of autonomous agents, knowing the user becomes the edge. Platforms that understand behaviors, preferences, and constraints won’t just personalize, they’ll act. Context becomes the moat.

But the bigger change: supply may be sidelined, or disappear altogether in some cases.

In Aggregation 1.0, suppliers lost leverage. In 2.0, they risk losing visibility. But how much they’re sidelined depends on what happens after the user’s goal is fulfilled.

In some verticals, like finance, travel, or healthcare, the supplier still matters after the action. There’s an ongoing relationship: managing accounts, delivering services, providing care. AI can streamline the path to decision, but it often makes the best suppliers more discoverable — not invisible.

In other areas, especially content and media, the supplier’s role can vanish entirely. If an AI delivers a news summary or factual answer, the original publisher might never be mentioned. This is why legacy media organizations are fighting back: they’ve gone from gatekeepers to ghostwriters, stripped of visibility in the value chain.

Supply isn’t always erased, but in Aggregation 2.0, it’s increasingly abstracted.

Aggregation 2.0: The AI Playbook

The next generation of platforms won’t just offer choices, they’ll deliver outcomes by combining intent, context, and action in a seamless loop:

• Intent: What you want, even if you can’t fully articulate it
• Context: What the system know — your history, behavior, and constraints
• Action: What the system can do, with your permission

A Personal Finance Example

You want to lower your monthly expenses. The system understands your spending, credit profile, and upcoming bills. It flags unused perks, suggests better payment strategies, and, if you approve, activates offers or refinances high-interest debt.

Kudos represents one approach: combining what you’re trying to accomplish with your actual financial situation to recommend and trigger meaningful actions that unlock earnings and savings. Always with user control at the center.

You don’t just get advice. You get results, with transparency and consent.

A Travel Planning Example

You want to book a stress-free trip to Lisbon. The system knows your travel habits, loyalty accounts, budget, and calendar. It assembles options, proposes an itinerary, and books, with your sign-off. You skip the rabbit hole of planning.

A Healthcare Example

You’re managing a chronic condition. The system monitors your health records, prescriptions, and wearable data. It schedules necessary checkups, recommends affordable providers, refills prescriptions on time, and flags issues before they escalate. You stay ahead of your care, with less effort.

Aggregation Theory 1.0 was about capturing demand by becoming the default interface, organizing suppliers and commanding time. Attention was the currency.

Aggregation Theory 2.0 is about earning trust, by continuously turning intent and context into valuable actions. Outcomes may become the new currency.

What Winners Will Do

The next generation of dominant platforms won’t compete on clicks. They’ll compete on outcomes.

They’ll:

• Prioritize outcomes over engagement
• Design interfaces that feel like companions, not dashboards
• Align business models with user success (e.g., save money, not just spend time)
• Let you maintain agency, even as systems act on your behalf

And they’ll evolve in stages:

• Short-term: Smarter suggestions with more data
• Mid-term: Actions with user review and approval
• Long-term: Delegation with clear safeguards

The Risks

As systems begin to act, not just advise, the consequences get bigger. Four major risks stand out:

• Overreach: Acting without full context or transparency. Target used purchase data to predict which customers were likely pregnant and sent them maternity-related coupons. In one case, a father was upset that his teenage daughter received such ads, only to later learn she was indeed pregnant. While technically accurate, the approach felt intrusive and sparked backlash about privacy and consent.
• Opacity: If users can’t understand why something happened, trust collapses. Systems must be explainable.
• Bias: AI trained on skewed data can reinforce discrimination, sometimes invisibly. In domains like financial services, using AI for underwriting could worsen existing inequalities if not carefully managed.
• Accountability: When things go wrong, people expect someone to own it. Air Canada’s chatbot gave wrong info on bereavement fares. A court held the airline responsible, not the bot.

Automation without transparency erodes trust. Delegation must be earned.

The Next Trillion-Dollar Opportunity

Aggregation 1.0 gave us access. Aggregation 2.0 will give us outcomes.

Winners will:

• Combine intent, context, and action to deliver clear value
• Build systems that are safe, transparent, and user-first
• Earn trust through consistency and clarity

Google built one of the greatest businesses of all time by capturing intent and monetizing attention. But in this new model, the very strength of its original approach becomes a constraint. Its core flows haven’t been rethought, not because of a lack of insight or talent, but because of how deeply its business model is locked into Aggregation 1.0. The hesitation to disrupt its own interface shows how tightly incentives can shape what evolves, and what doesn’t.

The next trillion-dollar platforms won’t just show us options. They’ll help us get things done, on our terms, with our permission, and in our best interest.

They won’t just organize the internet. They’ll help us live better, with less friction and more focus on what matters.

But as context becomes the moat, it’s worth asking: will a single platform like Google or OpenAI realistically have the context to act well across every domain?

Understanding what you want to eat tonight is very different from knowing how to refinance your debt or manage your health. The deeper and more personal the context, the harder it is to stretch across use cases without losing precision — and trust.

That’s why Aggregation 2.0 might not lead to one winner that does everything. Instead, we may see a new wave of vertical platforms — each deeply embedded in a specific context — emerge as trusted agents in their domain.

Aggregation Theory 2.0: The Trillion-Dollar Outcome Economy was originally published in Adventures in Consumer Technology on Medium, where people are continuing the conversation by highlighting and responding to this story.

After nearly a decade away from long-form public writing on Medium, I’m jumping back in. There’s simply too much happening right now — and so much of it intersects with the questions we wrestle with every day as we build Kudos. Agentic shopping is one of those topics, and lately, the questions I used to hear once a year are landing in my inbox daily: is this really how we’ll shop in the future, and how does Kudos fit in?

AI shopping agents have quickly become one of the buzziest ideas in tech. Visa, PayPal, and Amazon each made recent moves that point to a future where AI doesn’t just help you shop — it shops for you. That promise has fintech and VC circles buzzing. But as with past hype cycles — from Virtual Reality to Internet of Things— the excitement often arrives well before mass consumer adoption. Investors race in, headlines soar, and expectations rise. But adoption tends to happen slowly… then all at once. Just look at how quickly ChatGPT reached millions once the right product met the right moment. Agentic shopping may follow a similar arc — but only if it solves real problems and earns consumer trust along the way.

As someone building in this space, I think it’s worth slowing down and asking: what’s real, what’s coming, and what still needs to be figured out?

What Just Happened: Visa, PayPal, Amazon — and Now OpenAI and Perplexity Step In

Let’s break down what’s actually new — in plain terms — with each of these announcements:

Visa: Opened up its payments infrastructure to AI agents with tokenized, AI-ready cards. The incremental step here is that Visa is giving agents the tools to act more like trusted users — enabling secure transactions without breaking the trust rails that banks and merchants depend on.

PayPal: Released a developer toolkit that makes it easier for AI agents to plug into payments, shipping, and invoicing workflows. The new piece is its hosted MCP (Model Context Protocol) server — giving developers a way to inject real-time context into agent actions, like what product is being purchased and where it’s going.

Amazon: Launched a beta that lets its AI not just recommend but buy on your behalf from third-party sites. What’s new here is full end-to-end delegation — Amazon’s agent doesn’t just hand off to another store, it completes the transaction within your Amazon app.

OpenAI: Introduced a shopping experience inside ChatGPT powered by its browsing capabilities and informed by real-time search results. Users can now ask for product recommendations and get detailed, up-to-date shopping suggestions — often linking directly to places they can purchase. The new piece isn’t just better search — it’s OpenAI starting to stitch together a real commerce layer within the assist

Perplexity: Blending AI search with affiliate shopping results and monetization. It’s still early, but they’re getting closer to not just recommending what to buy, but earning money on those decisions — a stepping stone toward full agent-led purchases.

It’s worth noting that agentic shopping isn’t entirely new. Technically, consumers have been able to order products through Amazon Alexa for years. But voice commerce never truly took off — and the reasons are instructive. Limited product context, clunky interactions, and trust issues held it back. People weren’t ready to hand over purchasing power to a device that couldn’t show comparisons, clarify tradeoffs, or adapt well to nuanced intent. In contrast, today’s agents are built on more powerful models and more flexible infrastructure — capable of parsing complexity, learning preferences, and potentially delivering more intuitive handoffs. The shift now isn’t just about functionality — it’s about meeting users where they are with a system that earns their trust step by step.

A polished demo is one thing — but building something consumers actually trust and want to use is another. Closing that gap requires not just technical execution, but a deep understanding of real user needs and behavior.

The Hype vs. Consumer Reality

There’s no question these announcements are designed to spark interest — and that’s a good thing. Investor excitement helps fund the teams building what might become the future of how we shop. For VCs and corporate strategy teams, agentic commerce checks every box: AI, fintech infrastructure, commerce disruption. But for actual consumers, especially when it comes to financial decisions? The leap from AI advisor to AI actor is a big one.

Consumers aren’t just looking for convenience. They need to feel safe handing off decisions — especially financial ones — to software. And most people aren’t there yet. Giving a bot access to your wallet, preferences, and shipping address requires a level of trust that goes far beyond UX polish — especially when it comes to high-stakes shopping and financial decisions. That level of trust becomes even more critical when you’re delegating major shopping and financial decisions — which demand a fundamentally different level of confidence, consistency, and reliability that most AI experiences haven’t yet earned. Asking a bot a quick question or getting help with a bit of “vibe coding” is lower stakes by comparison.

That trust builds over time. The current agentic shopping pilots are likely to succeed first in narrow use cases: reordering household items, booking routine services, or checking out at familiar merchants. These are low-stakes, low-emotion transactions. High-context decisions like booking travel, buying clothes, or purchasing gifts will be much slower to hand off. In fact, despite the mobile boom, many of these larger or more involved purchases still happen on desktop — where consumers can more easily compare options, apply discount tools, and feel a greater sense of control. A recent Wall Street Journal piece even noted that retailers are frustrated by how often big-ticket purchases still happen on laptops rather than mobile. That’s not just inertia — it reflects the intentionality and perceived stakes involved in these decisions.

Where It Really Starts: Consumer Intent and Control

The agentic journey doesn’t begin with a prompt — it begins with intention. Before an AI can buy something for me, it needs to know: what do I care about? What constraints should it follow? What tradeoffs am I willing to make?

The industry tends to skip this step. The hard part isn’t building the API calls or integrating the checkout flow. It’s capturing nuanced consumer preference in a way that’s structured enough for the agent to act — without requiring a 30-minute configuration flow.

This is where a lot of real-world adoption will stall. Most consumers don’t want to fine-tune their shopping agent. They want it to just “know” — but that requires behavior tracking, trust in data sharing, and clear value delivery. And all of that depends on incremental wins — ones that prove to the consumer that the value exchange makes sense and the system is reliable. It’s those repeated, successful moments of assistance that move a user from curiosity to confidence.

The Infrastructure Problems Beneath the Surface

Once you’ve earned a user’s trust and they’re ready to delegate, the next set of challenges are no longer emotional or behavioral — they’re technical. While the consumer-facing agentic experience seems simple — “just have the bot buy it for me” — what happens under the hood is far from trivial. These challenges aren’t hypothetical. Back in 2022, we built a working prototype of a 1-click checkout experience — what we internally called a “supercharged autofill” — that automated every step in the checkout process on a handful of sites, including all Shopify checkouts. It worked, but not perfectly or consistently.

Once an agent understands your intent, the challenges don’t go away — they just shift to infrastructure. Some of the biggest roadblocks ahead that we need to work through for a commercially viable product include:

• Bot prevention: CAPTCHAs, 2FA, and browser fingerprinting all exist to stop exactly what AI agents are trying to do — automate checkout. Even well-meaning agents will trigger fraud systems.
• Checkout diversity: Every merchant site is different. Unless there’s a universal commerce API standard (there isn’t), agents have to custom-handle thousands of UI flows that change often or rely on fragile automation.
• Payment complexity: Modern checkouts support a wide range of payment methods — from credit and debit cards to BNPL (Buy Now, Pay Later), gift cards, and digital wallets — each with their own rules and edge cases. Many also include layered incentives like coupons, rewards, and cash back. For agents to truly replace manual checkout, they need to navigate these nuances reliably while aligning with each consumer’s preferences and financial goals.
• Accountability and returns: If an agent makes the wrong call, who’s responsible? What’s the return process? These questions need product, policy, and support solutions baked in.

Some companies tried to fake it till they made it — and crossed the line into deception. Take Nate, for example.. Instead of solving the hard technical challenges of agentic shopping, they masked the complexity by quietly relying on human workers — reportedly in the Philippines — to complete purchases manually. It was framed as automation, but behind the scenes, it was people, not software, doing the work.

The Work That Actually Matters

AI might end up being a bigger shift than the internet itself — and that’s not hyperbole. The scope of change it promises touches everything from how we search to how we buy, save, and decide. So yes, the excitement is real.

As a founder building an AI-powered wallet, I get that excitement. It’s easy to picture a future where AI handles the mess of online shopping. But I also know where these systems break:

• They break when the consumer doesn’t understand what the agent is optimizing for — and the outcome feels arbitrary or misaligned.
• They break when the consumer doesn’t see enough value to justify the effort — even if the tech works, the payoff has to be obvious.
• They break when behavior change is required, but the user isn’t ready or doesn’t trust the system enough to try something new.
• They break when merchants don’t play along — rejecting automation through anti-bot systems or refusing to integrate.
• They break when the transaction fails at the last mile — whether that’s a CAPTCHA wall, 2FA challenge, or flagged payment that halts the flow.

What we’re seeing now are the first primitives. Visa is solving the tokenization and trust layer. PayPal is building APIs that make agentic checkout programmable. Amazon is testing UI flows and user comfort. It’s the right sequence — but we’re still early. And to be honest, some of this depends on how much is truly being built versus what’s being positioned to satisfy investors or check the AI and agentic boxes. A roadmap milestone isn’t the same as a production-ready system — and building trustworthy delegation takes more than a press release or a slick demo.

At Kudos, we’ve taken a deliberately assistive approach. We started by helping users in small but impactful ways — like intelligently recommending which card to use at checkout and autofilling it to maximize rewards. With our merchant Boosts, we show users how they’re earning more just by letting Kudos step in at the right moment — small wins that build confidence and trust.

Because we’ve already automated the decision layer at checkout, we believe the natural next step is layering in more automation — like 1-click execution — gradually, as we smooth out the remaining kinks to deliver a truly reliable experience. It’s a clear extension of what we already do, and we’ll introduce it carefully as the supporting technical and compliance details mature. We apply this same philosophy across the product. When we see strong engagement around a specific recommendation — like activating a card-linked offer — we explore how to take the friction out entirely by doing it for the consumer. That could mean turning a reminder into an automated action, or eventually helping users handle more complex tasks, like refinancing high-APR debt, once we’ve earned their trust to do so.

Our AI-powered insights are already live today — surfacing personalized, context-aware recommendations based on each user’s spending, card portfolio, and missed opportunities. These might include flagging unused card benefits, alerting users to upcoming annual fees, or suggesting when switching cards could lead to significantly better rewards. Each insight is designed to be timely, actionable, and easy to trust — helping users take the next best step without needing to dig through fine print. As users act on these insights and see results, we identify the right moments to step in more fully. Because it’s not just about what the agent could do — it’s about what the user is ready to delegate next.

It’s a gradual path — slow at first, then sudden once the right conditions align. But it will take deliberate effort to stay anchored in solving real customer needs and building trust. Consumers’ priorities and investor excitement don’t always align. What feels like a moonshot to Wall Street might still feel risky to the average user. Winning this space means earning trust first, and scaling ambition second.

What Comes Next

Real adoption won’t come from hype. It’ll come from behavior change — and behavior change comes from small, consistent wins. Agents that reorder something you already trust them with. Agents that ask for confirmation the first few times, then learn. Interfaces that help users set preferences without overwhelming them. And increasingly, agents that can guide users step-by-step through a flow — gradually earning their trust before moving toward all-in-one-click automation. Gradual guidance builds understanding and confidence, paving the way for deeper forms of delegation over time.

Ultimately, agentic shopping will be transformative — but only if we earn it. That means designing with patience, putting the consumer’s needs ahead of the hype cycle, and proving value through transparency, control, and outcomes they can see and feel.

Until then, the most agentic move we can make is to stay grounded — and build with realism, consumer empathy, and momentum.

Shopping Bots Are Coming — But Will Consumers Ever Trust Them With Their Wallets? was originally published in Adventures in Consumer Technology on Medium, where people are continuing the conversation by highlighting and responding to this story.

Modeled after “Best Girl” Asuka’s iconic mech, Pentax released the colorful, diminutive Q10 in 2013 as a tie-in to the Rebuild of Evangelion films. Weighing just 200g, Pentax’s Q system maintains the distinction of being the world’s smallest interchangeable-lens camera.

However, even by the standards of the day, its 1/2.3" sensor was tiny—limiting the bokeh and ISO performance relative to the similarly priced competition. What utility, then, could this camera have in 2025 beyond collecting dust on an otaku’s shelf? Is there a future for this fiery Pentax beyond Gen Z’s nostalgia for old tech?

A Sneaky Concert Shooter: the Pentax 06 Telephoto Zoom features the range of a 70–200mm full frame equivalent lens, and measures only two inches! Most lenses with that sort of reach will never make it past security. Who would suspect your toyish Evangelion collectible is capable of snagging close-ups of the band? The best smartphones of today can’t compare.

The Q10 hides a complementary surprise: it features one of the first in-body stabilization systems. This enables slower shutter speeds to be used with the telephoto, sidestepping the high ISO deficiencies of the camera. So long as the band isn’t hopping around like Mick Jagger, sharp results are possible even in dim scenarios.

To the Moon: another quirk of the design is the short flange distance between the lens and sensor. This comes into play in astrophotography, as the camera can be mated to a telescope without vignetting.

It gets better—the huge crop factor increases your reach, allowing for moon shots that nearly fill the frame. Even modest scopes become capable of close-ups with significant detail. The math is boring, but the resolving power of this combination exceeds modern solutions which cost thousands more.

A Night Out with Friends: should I bring my camera to the bar? Am I risking sticky buttons, scratched lenses, or worse? Photographers are familiar with the quandary—who wants to worry about their expensive gear? Thankfully, Q series cameras are affordable!

You’ll pay a premium for fancy robot models, but standard Q cameras can be had for well under $200. Grab the fast 01 Standard Prime, pop up the adorable flash, and you’ve got a capable, low-stress memory machine.

Bonus: the still-cheap Q7 model features an updated sensor with slightly better IQ in low-light situations. No longer does a spilled Negroni ruin an evening!

Street Photography: telephoto lenses are pariahs of street work, often unable to capture candid moments due to their noticeable and intimidating size. Once again, the Q’s Telephoto Zoom offers an alternative—providing reach in a subtle and playful package.

Buck conventional wisdom and achieve a tight perspective while avoiding ruffling feathers. Deep depth-of-field contributes to quick handling, despite the outdated autofocus and cramped controls.

In the age of planned obsolescence and disposable tech, it’s gratifying to stumble upon use cases for the tools of yesterday. Pentax created something unique with the Q series and it remains a showcase of the lesser-known benefits of small sensors. In short—wunderbar.

The Feisty German of Tiny Japanese Cameras was originally published in Adventures in Consumer Technology on Medium, where people are continuing the conversation by highlighting and responding to this story.

If you spend enough time online, you’ll encounter ads for telemental health company “BetterHelp.” Founded in the mid-2010s, BetterHelp offers online counseling with licensed therapists and has grown into the industry leader, following a boom during the COVID era. These days, it’s difficult to find a podcast not sponsored by the company.

Meanwhile, in-person therapy is harder to come by than ever. In search of an empathetic ear, I ran into one roadblock after the next; reputable counselors were swamped with patients and extended waitlists.

Driven by this shortage and enticed by an aggressive promotional discount, I decided to give online counseling a try. After a reasonably successful first video session, I attempted to engage my therapist using the platform’s built-in text chat functionality—and things quickly went awry.

I shared my thoughts on “The Courage to Be Disliked”, a misleadingly-titled book discussing Adlerian psychology. Within an hour or two, my therapist replied with a disconcertingly familiar block of text.

As the husband of a high school teacher, I‘ve seen my share of clandestine AI-generated writing. Large language models have a habit of rephrasing the prompt text; a sort of faux active listening. They parrot back words in an inoffensive way, while adding very little—just as my therapist had done.

Though I was near-certain this text had been written by AI, I ran it through a few detectors to verify my suspicion. While not infallible, these systems reinforced my hunch. I asked her point-blank: did you reply to me using an AI?

She admitted to “referring” to AI, without “revealing any client information”. This struck me as a half-truth; the reason I caught this AI usage was because the machine had written a lengthy paragraph. It was also rephrasing what I wrote—meaning she must have fed it my privileged communication to have generated that text.

Thankfully this happened after only one session, and regarding something relatively trivial. Imagine if I had built a meaningful relationship with this therapist, and then she swapped in the robot? Or if my writing had been about something deeply personal, or serious?

The legal and ethical implications of undisclosed AI use in therapy settings are significant. I first turned to BetterHelp’s Privacy Policy—was this behavior permitted?

BetterHelp has already been in hot water for selling data to advertisers, but what did their legalese say about communicatons between a therapist and patient? “Messages with your Therapist are not shared with any Third Party.” Clearly, the sharing of my writing with a large language model was in violation of this policy. While I’m no lawyer, it would also seem this disclosure of my writing would constitute a breach of therapist-patient privilege in my state.

If my intention was to chat with an AI about my feelings, I could’ve done so — at little or no cost. Instead, I invested the time and energy to engage a person of expertise. I felt an accute sense of betrayal when my thoughts were met with the impersonal voice of a machine. This artificial sleight-of-hand is not what BetterHelp users are signing up for.

I detail this experience not to criticize a specific therapist, but to call attention to the impact artificial intelligence is already having on mental health treatment and privileged communications. These AI tools have many potentially beneficial uses in healthcare, but strict disclosure and data safety policies must be enforced.

As the workforce stresses about potential job displacement, it’s disheartening to see a therapist outsource their own career—doubly so considering human empathy is the cornerstone of this occupation. Governments, corporations, and individuals must steer artificial intelligence use to avoid these kinds of dystopian outcomes and maintain institutional trust.

When reached for comment, BetterHelp referred only to the aforementioned Privacy Policy.

Artificial Empathy: My BetterHelp Therapist Took an AI Shortcut was originally published in Adventures in Consumer Technology on Medium, where people are continuing the conversation by highlighting and responding to this story.

In a recent post on resources for discovering podcasts, I noted that 2025 will be a year when those in the PR profession must reconcile the desire by clients for coverage in legacy media versus the potentially greater impact of a whole new cadre of earned media platforms. In the piece, I wrote:

“There is nothing like a prominent Wall Street Journal or New York Times feature, let alone a six-minute segment on NBC “Today” or PBS “NewsHour.” But do these traditional media outlets move the needle as much as a widely followed social media star, top-rated podcast, or heavily subscribed newsletter? Could Joe Rogan have more influence than NPR?”

It’s an important question for our times and the PR industry.

The growing hegemony of podcasts, newsletters, and Insty, TikTok, and YouTube stars may prove even more valuable than the productivity gains from AI and machine learning. A well-timed appearance on a top-rated podcast can have a greater impact than an ephemeral story in a paywalled legacy news outlet, with the last election being a primary example.

In her Nieman Lab piece, “Podcasting Becomes the Primary Media Strategy, not an Afterthought,” The Podglomerate’s Joni Deutsch observed:

“At a time where traditional media trust continues to hit historic lows, podcasts can help long-standing organizations establish stronger relationships with audiences and, in the process, drive real change in how people think and engage with the news.”

For years, The New York Times set the national news agenda. A prominent investigative report or fawning profile could easily lead to a Congressional hearing or the A-list, respectively. And while The Times still sets tongues-a-waggin’, its paywalled content does not travel as far and wide as a top influencer’s social media post or podcast episode. Diminished trust in mainstream media is another barrier, but that’s a topic for a different day.

These once-pedestrian content formats are truly ascendant with their ability to ratchet up digital eyeballs and a measurable increase in awareness, product sales, or even stock price. They have legs. It’s no wonder TikTok is doubling down on e-commerce.

As for real-time platforms, Twitter once had the unique capacity to create national story memes, in part due to the myriad journalists and (actual) verified users who loitered and amplified on the platform. No longer. An Elon-remade Twitter has X’d into a cesspool of misinformation, conspiracy, and anger, e.g.,

Thus, many in the fourth estate have abandoned his propaganda machine and migrated to Meta-owned Threads, only to learn that its algorithm suppresses breaking news in favor of titillating content. Strangely, Threads also hinders engagement with others and now, along with its sister platforms Instagram and Facebook, will follow X into the fact-free zone under the guise of free speech.

Most recently, BlueSky is enjoying its moment in the sun. Founded, but abandoned by Twitter co-founder Jack Dorsey, this more civil platform has usurped both X and Threads as a safe , real-time resource for journalists. In December, it surpassed 25 million registered users. Don’t ignore it.

Here’s a BlueSky post from Julie Brown, the Miami Herald investigative reporter who doggedly broke the Jeffrey Epstein story:

Then there are the countless newsletters that regularly arrive in your inbox. They reside on self-publishing platforms like Substack, Beehiiv, or Ghost and often feature the work of prominent journalists, many refugees of big-branded legacy media who’ve forsaken their once high and mighty perches to go it alone. Lester Wunderman, eat your heart out.

Some of my favorite newsletters include Casey Newton’s “Platformer,” Jedd Legum’s “Popular Information,” Oliver Darcy’s “Status,” Dan Perry’s “Ask Questions Later,” Taylor Lorenz’s “User Mag,” and any of those emanating from Axios, Puck, Semafor, or The Information.

As for AI, make no mistake: we’re witnessing a whirlwind of promising use cases for PR practitioners. They range from analyzing a journalist’s past coverage before an interview to condensing a laboriously long media pitch to its “Smart Brevity”-style essence. Here’s a piece on one AI platform that I’ve found especially beneficial to the clients of my firm.

What will not change for PR pros who compete in the earned media relations game is the requirement that they stay informed, inquisitive, and immersed in as much client-relevant content as possible, no matter how it arrives.

Finally, as we head into 2025, I’d like to give a shout-out to some of the newer media outlets that have embraced quality journalism. Paramount among them is ProPublica for its essential, First Amendment-protected investigative reporting that holds wrong-doers to account.

Happy New Year.

Peter Himler is the founding principal of Flatiron Communications, a NYC-based PR and digital media consultancy. Follow him on BlueSky or Threads.

The Mediaverse and the Year Ahead was originally published in Adventures in Consumer Technology on Medium, where people are continuing the conversation by highlighting and responding to this story.

Since the election, I’ve wondered whether Kamala Harris’s decision to pass up Joe Rogan’s #1-rated podcast may have cost her the Presidency. Her communications consigliere Stephanie Cutter later explained the ill-fated choice to Crooked Media’s “Pod Save America” podcast:

Yet, VP Harris did find time to trek to New York to appear on a half-century-old linear TV show that averaged 5.4 million viewers through 2024, most of whom were already in VP Harris’s court. Her appearance that evening drew 6.6 million, still 40% less than Rogan’s.

By contrast, Trump’s stultified appearance on the Joe Rogan Experience drew an audience of about 50 million and, more significantly, a pivotal Presidential endorsement from Rogan himself on the eve of the election. Ouch.

Unlike other communications counselors, I don’t relish the role of Monday morning quarterback, especially over the Democrats’ failed messaging and media strategy.

To her credit, the Vice President did join the #2-rated podcast “Call Her Daddy,” with an audience of one million, only to have that show’s host, Alex Cooper, diss the campaign to Andrew Ross Sorkin at the Dealbook Summit. The Harris PR team also pitched the Veep to “The Hot Ones,” the popular chicken-wing-eating show on YouTube, but they were turned down.

In either case, all this election conjecture is now moot given the world’s richest man's infusion of $277 million to Trump and the GOP, further drowning the American electorate in a sea of disinformation. Relatedly, Semafor reports that the king of disinformation himself, Rupert Murdoch, is now shopping for like-minded podcasts.

I digress.

In her Nieman Media Lab piece, “Podcasting becomes the primary strategy, not an afterthought,” Joni Deutsch writes:

“At a time where traditional media trust continues to hit historic lows, podcasts can help long-standing organizations establish stronger relationships with audiences and, in the process, drive real change in how people think and engage with the news.”

I thus want to use this space to explore where podcasts should reside in the PR pro’s “earned media” toolbox and share some useful resources for identifying the most appropriate pods for your company or client — of the millions now in production.

To this end, I surveyed the founders of four ascendant companies that offer searchable databases of podcasts. They include Anuj Agarwal, founder of MillionPodcasts, Greg Galant, co-founder and CEO of Muck Rack, Bradley Davis, founder and CEO of Podchaser, and James Potter, founder of Rephonic. First this chart:

Here is what they told me:

Q: Briefly describe your service and what makes it special:

MillionPodcasts is the ultimate solution for PR professionals, marketers, and outreach managers looking to connect with the right podcast hosts and producers. Our extensive and continuously updated database includes verified email contacts, social media profiles, and detailed podcast information across all niches and hosting platforms.

Q: What makes you special?

Dedicated Research Team: Ensures accurate, up-to-date contact details of hosts and producers.

• Advanced Filters: Allow tailored searches by popularity, activeness, location, duration, and more.
• AI-Powered Categorization: Simplifies discovering podcasts by niche or interest.
• Effortless Export: Generate CRM-friendly custom lists in CSV or Excel formats.

Q: Where are you headquartered and when were you founded?

MillionPodcasts is headquartered in Newark, Delaware, USA, and was founded in 2024.

Q: How would you quantify/describe your subscribers/users?

Our subscribers include PR professionals of all sizes, marketing teams (from small to large-scale operations), researchers, founders, and executives working in the podcast industry.

Q: What am I missing?

We have an in-house research team that goes beyond RSS feed emails, finding personal emails of hosts and producers to ensure comprehensive contact information.

Q: Briefly describe your service and what makes it special:

Muck Rack is AI-powered PR software built for how you work. What makes us special is that we’ve purposely built our AI-powered comprehensive and integrated platform from the ground up with public relations and communications professionals in mind to streamline the PR workflow to get you the right results, faster and more efficiently. We grow businesses by helping them generate positive media coverage, monitor mentions
to manage brand reputation and analyze the impact of PR on business outcomes.

Q: Where are you headquartered and when were you founded?

Muck Rack was founded in 2009 to connect journalists on social media. In 2011, the Muck Rack software platform was born, offering journalist profiles and a media database that gained a reputation for being the most accurate in the industry. We’re headquartered in the cloud! Muck Rack has been a remote-first company since our founding and became fully distributed in 2021. This means no offices and no physical headquarters.

Q: How would you quantify/describe your subscribers/users?

Muck Rack has two user groups: Our customers are primarily communications and PR professionals, with about 6,000 companies using our platform to generate earned media and analyze and report on their coverage. Founded as a journalist tool, we also have tens of thousands of journalists using Muck Rack’s free tools to showcase their work with online portfolios, analyze news about any topic, and measure the impact of their stories.

Our podcast database offers a variety of filters, including geography, languages, and domain authority, and users can search by a specific podcast name or keyword.

Q: What am I missing?

This question: What should PR pros consider when it comes to podcasts and incorporating them into their strategies?

The recent election results have sparked an awakening among corporate executives to rethink their approach to media relations, with an emphasis on nontraditional platforms like podcasts. But this shift is hardly news to PR pros. PR teams have been focusing on targeting niche outlets and audiences for years. At Muck Rack, our customers have been pitching podcasts since we introduced them to our media database in 2020 and have added 153,117 podcasters and 885 podcasts to their media lists just last month alone. The key here is that CEOs and business leaders are finally catching up to the value of niche targeting, which their PR teams have been trying to convince them of for years. Getting in front of the right, albeit sometimes smaller, audience means shifting KPIs from large-scale (impressions) to smaller-scale (engagement).

Q: Briefly describe your service and what makes it special:

Podchaser is the intelligence engine for the podcast industry, complete with 5.5M podcasts, podcast reach, audience insights, and 2M+ validated contacts

Q: Where are you headquartered and when were you founded?

Podchaser was founded in 2016 by a group of like-minded podcast enthusiasts who met via a humble Reddit thread. With the optimistic goal of creating the ultimate platform-agnostic podcast database, the small team worked to build a company from the ground up. The company has been fully remote and distributed since its founding. Podchaser was acquired by Acast in 2022.

Q: How would you quantify/describe your subscribers/users?

Hundreds of PR agencies, many dozen brands, and media organizations use Podchaser Pro. The majority of Pro users leverage podcast data for earned media, including guest booking and media monitoring. A growing number use Podchaser for podcast advertising use cases. The media orgs use Pro for research and insights, including competitive intel.

Q: What am I missing?

As traditional media continues to fracture and decentralize, the importance of podcasts has never been greater. We frequently hear from clients how Podchaser breathes new life into their businesses in the face of shrinking newsrooms and media contraction.

Q: Briefly describe your service and what makes it special:

Founded in 2020, “Rephonic is an all-in-one podcast outreach and research tool that gives access to detailed listener demographics and contact information for over three million podcasts. It simplifies the process of discovering relevant shows, pitching them at scale, and managing outreach campaigns with real-time data, saving users hours of manual work.”

Q: How would you quantify/describe your subscribers/users?

PR professionals, marketing agencies, individuals pitching themselves, businesses seeking sponsorship opportunities, and podcasters looking to cross-promote their shows. They range from solo entrepreneurs to teams at major brands.

Q: What am I missing?

Rephonic also has:

• Lots of advanced search filters (e.g. show me podcasts likely listened to by mechanics)
• Ability to create and collaborate on target lists with your team
• A concierge service to find better contacts
• Full-text episode transcripts

I‘ll leave you with this observation from Matthew Belloni of Puck in his piece on entertainment PR agency PMK’s defections and the lawsuit they spurred:

“You don’t need me to tell you that talent P.R. is a much different job today than it was even 10 years ago. Now, a well-timed Deuxmoi post can be far more effective than an EW cover. Chicken Shop Date is the new Tonight Show.”

Still, there’s nothing like a prominent Wall Street Journal or New York Times feature, let alone a six-minute segment on NBC “Today” or PBS “NewsHour.”

Do these traditional news organizations move the needle to the same degree as an influential social media star, top-rated podcast, or heavily subscribed newsletter? It’s an important question for these times and the PR industry.

If you can’t spring for a subscription to one of the above vendors, here is a (free) list of the Top 50 Podcasts in the U.S. on Spotify for 2024. But, oops, newbie Kylie (Mrs. Jason) Kelce’s “Not Gonna Lie” podcast just leaped ahead of Joe Rogan’s to take the top spot.

1. The Joe Rogan Experience
2. Call Her Daddy
3. This Past Weekend w/ Theo Von
4. Crime Junkie
5. The Daily (New York Times)
6. The Tucker Carlson Show
7. Huberman Lab
8. Armchair Expert with Dax Shepard
9. Smosh Reads Reddit Stories
10. Shawn Ryan Show
11. Up First from NPR
12. Serial Killers
13. The Journal* (Wall Street Journal)
14. SmartLess
15. Lex Fridman Podcast*
16. The Mel Robbins Podcast
17. Morbid
18. Rotten Mango
19. New Heights with Jason & Travis Kelce
20. What Now? with Trevor Noah*
21. Bad Friends
22. Stuff You Should Know
23. Anything Goes with Emma Chamberlain*
24. Dateline NBC
25. Conspiracy Theories*
26. Matt and Shane’s Secret Podcast
27. NPR News Now
28. Kill Tony
29. 2 Bears, 1 Cave with Tom Segura & Bert Kreischer
30. Hot Mess with Alix Earle*
31. Malevolent Mischief: True Stories of Horror
32. Distractible*
33. Science Vs*
34. On Purpose with Jay Shetty
35. The Ben Shapiro Show*
36. MrBallen Podcast: Strange, Dark & Mysterious Stories
37. Pod Save America
38. The Viall Files
39. 20/20
40. PBD Podcast*
41. Today, Explained
42. The Comment Section with Drew Afualo*
43. Cancelled with Tana Mongeau & Brooke Schofield
44. Rotten Mango Video*
45. The LOL Podcast*
46. Modern Wisdom*
47. Conan O’Brien Needs a Friend
48. The Diary of a CEO with Steven Bartlett
49. Andrew Schulz’s Flagrant with Akaash Singh
50. Murder in America

Podcasts Jump the Shark was originally published in Adventures in Consumer Technology on Medium, where people are continuing the conversation by highlighting and responding to this story.