For years, the front door to journalism was a search bar. News lived where audiences typed questions, tapped app icons, or scrolled curated feeds. But during INMA’s Media Tech & AI Week in San Francisco, executives heard something striking: Those doors are multiplying — and in some cases, disappearing altogether.

Across five days of study tour visits and conference discussions, a new information architecture emerged. Search engines are becoming answer engines. Feeds are morphing into companions. And conversational agents are starting to do what browsers once did — not simply retrieve stories, but mediate the relationship between people and information.

What follows are the week’s main takeaways for how AI is reshaping discovery, attribution, and value for news publishers.

A new era of discovery

The audience pathways that once looked stable now vary by device, region, and interface. 

Chartbeat’s global data showed that in many Android-heavy markets, Google Discover already outpaces traditional search as a source of news traffic. 

Discover is about 20% and Google search is 6%-7%, said John Saroff, CEO of Chartbeat, noting the divide is “actually a device story.” On iPhones, Discover is only 3% of iOS audience, while search remains around 12%. On Android, by contrast, Discover is all over the phone.

That distinction may sound technical, but it’s strategic. For news publishers, performance is no longer only about keywords or content freshness. It’s about the operating-system mix in each market — which algorithmic feed dominates the local handset base and how content is rendered within it.

The same Chartbeat presentation revealed how AI itself is joining that distribution mix. Referrals from ChatGPT may still be small but have risen five times since July 2024, Saroff said, signalling conversational entry points are beginning to send measurable traffic. For the first time, executives could quantify the earliest effects of answer-engine discovery.

The shift from consumption to companionship

If Chartbeat mapped the data, Microsoft sketched the product vision. Its team previewed Copilot Discover, a next-generation feed that pairs articles, videos, and community commentary with a chat companion. 

“It’s about content consumption with companionship,” said Aparna Lakshmi Ratan, partner product director at Microsoft AI (MSN) — an experience designed to let users talk alongside the content, not just scroll through it.

Behind that interface is a dual memory system: an “MSN memory” that learns from a user’s browsing and feed behaviour, and a “Copilot memory” built from their chat interactions. Combined, they enable a level of personalisation that moves beyond headlines and into intent — why the reader came, what they asked, and what they might ask next.

For news publishers, the implications are two-fold. First, attribution must travel with content as it enters these semi-closed, conversational experiences. Second, the unit of engagement is no longer just a pageview or a click but the conversation itself. Measuring brand presence inside those dialogues will become as important as measuring traffic outside them.

A marketplace for agentic content

Microsoft also introduced the Publisher Content Marketplace, a step toward giving publishers visibility — and compensation — when their work appears in AI agents or copilots. The concept answers two common questions, as Nikhil Kolar, vice president of product and engineering at Microsoft, said: “How often is [my content] being used? Are you being paid for it?” It also aims to support the “long tail” of AI builders beyond the big model companies.

That framework reflects a broader industry goal: to replace opaque crawling and training with transparent licensing and payment. As Sara Trohanis, head of strategic partnerships at the Associated Press, emphasised in another session, “Licensing models ultimately protect journalism’s value and protect publisher value … we do not do every deal that comes up very purposefully.” Selectivity, not volume, is the marker of sustainable participation.

The week’s discussions made clear that discovery and rights are converging. To appear in answer engines and agentic interfaces, news publishers need content that is not only findable but also readable by machines, attributable by design, and covered by explicit terms of use.

From answer engines to agents

At the product layer, the conversation is moving from static responses to agentic behaviour — AI systems that act on behalf of users rather than merely reply to them. 

Jessica Chan, head of publisher partnerships at Perplexity, described its evolution “from an AI answer engine” to an agentic browser called Comet. “We do not build our own foundation models. Perplexity acts as a RAG [retrieval-augmented generator],” pulling in licensed and open sources to create a composite answer.

This architectural shift means the competition for audience attention is happening at the orchestration level. Agents decide not only which sources to cite but also how to fulfil a user’s request — summarising, comparing, booking, buying, or explaining. News becomes one ingredient in a blended output.

The conversation at OpenAI added another dimension. Executives described ChatGPT apps as a new layer of distribution — effectively an “operating system for interaction” where publishers could build in-chat products, not just distribute stories. The takeaway: The next generation of discovery may happen inside a chat, not a browser.

Local use cases for the agentic age

Agentic AI may sound abstract until it becomes practical. 

Nexstar Media Group offered an early example: an experimental agent trained on local data to answer real-time questions such as traffic or weather conditions. “It needs to be able to answer those questions … or else it’s not useful,” said Jeff Moriarty, chief product officer at Nexstar.

That use case points toward the most immediate opportunities for media: build agents where you already have proprietary, time-sensitive data. Traffic, transit, power outages, school closings, and other local signals translate naturally into conversational services that audiences will value and return to. In effect, a publisher’s archives and feeds become fuel for micro-agents — small, focused AI services that turn trusted information into everyday utility.

The economics of access

Behind every new gateway lies an old question: Who pays? 

During the study tour visit to Cloudflare, the argument was that as the cost of building and maintaining large models climbs, fair-value licensing is the only sustainable solution.

This view resonated across the week. Whether through direct deals, collective standards, or emerging marketplaces, the news industry is aligning around the idea that access must have terms. Training data, snippets, and summaries are not free raw material but professional assets that require attribution and compensation.

The human factor in a machine-first world

Amid all the discussion about models, agents, and marketplaces, one theme kept resurfacing: In a machine-first world, trust is still the differentiator. However advanced AI interfaces become, audiences continue to rely on known, credible brands when the information stakes are high.

James Cham, partner at Bloomberg Beta, reminded participants that algorithms may excel at prediction but not at judgment. Publishers are not replacing judgment, they are building systems that still need trusted interpreters. As more discovery happens through conversational agents, that human layer of editorial credibility becomes a key signal for the machines themselves.

From the platform side, Danny Sullivan, director of Google Search, underlined the same point during his session on AI Overviews. Google is designing its generative-AI experiences to highlight verifiable sources and clear citations, he said that users need to understand where the information is coming from and why it’s being shown to them. That’s a signal to publishers that brand reputation must be made machine-readable — through schema, markup, and metadata that communicate authority.

Microsoft AI’s Ratan showed how that principle is already being implemented inside Copilot Discover. “The way we are approaching it is to really focus on the brand, the trust, and putting that front and center,” she said.

For news publishers, that’s a clear cue: Visibility and trust are now design choices.

During the end-of-the-week INMA podcast, The Debrief, INMA Product & Tech Initiative Lead Jodie Hopperton cautioned that as AI interfaces evolve, publishers can’t let their brands disappear behind the chat layer. Conference moderator Robert Whitehead stressed that provenance and trust signals must be embedded in code and standards — so the machines can read who we are. Moderator Nicki Purcell summed up the challenge succinctly: Audiences trust names they recognise, not answers from nowhere.

On the study tour, speakers at Hearst and CTGT added concrete examples of human oversight. Tim O’Rourke, Hearst’s vice president of editorial innovation and AI strategy, described how their AI systems are built to keep editors in the loop to ensure accuracy and accountability. At CTGT, Cyril Gorlla showed how AI transparency tools can audit the data and anchors inside models, demonstrating that trustworthiness is measurable, not just philosophical.

Together, these voices formed a clear consensus: Trust must be visible, verifiable, and programmable. In a landscape where AI systems are the new intermediaries, reputation can’t be assumed — it has to be encoded. That means verified bylines, brand identifiers, structured metadata, and ethical governance that signal credibility to both audiences and algorithms.

A call for coordination

The closing podcast featuring Jodie Hopperton, who curated the entire week and distilled the week’s urgency.

“This is the biggest moment we’re ever going to face in our careers, in our generation,” said Robert Whitehead, head of the INMA Digital Platform Initiative. The opportunity, they argued, is to “create standards and not replicate 26,000 isolated experiments … and to exploit the agentic AI era.” 

Big Tech companies, Purcell said, “don’t have all the answers … some of them have really small teams … and they’re just willing to experiment with us.”

That tone — collaborative but assertive — captured the mood of the week. Media companies no longer see themselves as passive subjects of platform change. They are early participants in shaping the next interface between humans and information.

What news companies should do now

1. Tune strategies by device and market: Android-dominant regions behave differently from iOS markets. Optimise for where Discover, not Search, drives reach, and monitor how AI referrals evolve.
2. Instrument for conversational attribution: If your content appears in Copilot, Perplexity, or other chat interfaces, ensure brand identifiers, headlines, and rights statements travel with it. Track how often your organisation is named inside answers.
3. Prepare for marketplaces and standards: Adopt machine-readable rights metadata and participate in frameworks that define payment and usage — from Microsoft’s marketplace to forthcoming industry standards.
4. Pilot utility agents: Use unique data — traffic, weather, local alerts — to build narrowly scoped agents that demonstrate public value and reinforce brand identity.
5. Be deliberate about licensing: Follow the AP’s example: Value your intellectual property, choose partners carefully, and reject agreements that dilute control or transparency.
6. Align editorial, product, and legal teams: Discovery, metadata, and rights are no longer separate silos. Coordination will determine whether journalism thrives inside the agentic ecosystem or becomes background noise.

Conclusion

By the end of Media Tech & AI Week, one phrase echoed through hallways and bus rides alike: The front door has moved. Search bars, app stores, and social feeds remain, but the most powerful gateways now speak back. They answer, recommend, and act — and in doing so, they redefine what it means to be found.

For news publishers, that shift is both challenge and invitation. The next stage of discovery will not be fought over keywords but over context, clarity, and control. Whoever learns to encode those values into the language of machines will decide how journalism is encountered in the decade ahead.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .