Large language models are triggering the second major disruption in the publishing industry. To understand its scale, it’s worth remembering something uncomfortable: the industry still hasn’t recovered from the first one.
Platforms Took Discovery
For two decades, publishers ceded discovery to platforms. They adapted to SEO, chased social-media traffic, rebuilt audiences behind paywalls, and competed for attention against low-effort entertainment. Even in that “survival mode,” the losses were real: brand dilution, ad-blocker erosion, fewer readers reaching the paywall.
The industry survived that disruption, but it never won it. And now a new wave hits the exact same weak point: discovery.
AI Has Moved Discovery Out of Sight
Search engines still offered a predictable route to visibility, even if the terms were dictated by the platforms. AI assistants erase that predictability.
When people ask a question, the answer is assembled by an intermediary system whose sourcing process is largely invisible to both readers and publishers. The “path” from original reporting to the final response becomes a black box.
And young audiences are moving into this behaviour at speed. Already, 15% of under-25s in global markets use AI chatbots for news weekly (Reuters Institute 2025), and the trend accelerates each month.
When Discovery Moves, the Entire Value Chain Shifts
Discovery sits at the top of the funnel. When it drifts out of publisher-controlled spaces, traffic, monetization, and brand leverage all start to collapse.
AI summaries reduce the need for readers to click through at all. Pew Research Center found that only 1% of people who read an AI-generated summary clicked the cited source. Google’s own tests show the same dynamic: with “AI Overview,” users clicked standard links in 8% of visits vs. ~15% without it.
These aren’t short-term fluctuations. The industry keeps treating this as a quantitative problem (“traffic is down”), when the real issue is qualitative: traffic is no longer the channel through which value flows back to publishers.
And this is where many publishers take comfort in familiar defences:
Ads still work. Brands still matter. Paywalls still protect revenue.
But none of these hold in the new discovery architecture.
Why Advertising Becomes Less Valuable
Advertising relies on one simple thing: publishers must be able to predict how many people will visit their site and see the ads. That predictability is what advertisers actually buy.
But when more news consumption happens through AI-generated summaries, in contexts publishers don’t control, that predictability fractures. You can’t promise the behaviour patterns agencies need.
The result is that advertising is becoming unreliable, and unreliable revenue is impossible to build a business on.
Why Brand Power Weakens
The power of a publisher’s brand has always depended on readers arriving at a distinct, recognisable experience: the by-line, the site design, the signature editorial voice.
AI interfaces flatten that.
In practice, for a 20-year-old reader, a story from the Financial Times and a story from an unfamiliar outlet often look identical: both appear as paragraphs inside a chatbot answer. The editorial voice is abstracted away. The source becomes optional metadata. The distinctiveness disappears.
And when AI made mistake, the source is blamed as well.
A 2024 audit by the Global Disinformation Index (GDI) across 18 countries found that 45% of AI-assistant news answers contained sourcing or attribution errors. In follow-up interviews, users blamed not only the AI system but also the news outlet that appeared as the source.
The bottom line is that intermediaries now shape the reader’s trust more than the publishers who produce the work.
Why Paywalls Stop Protecting You
Paywalls are treated as the industry’s last line of defence. In reality, they protect far less than publishers assume.
First, the walls don’t stop the crawlers. Perplexity has been accused — with supporting evidence from Cloudflare and WIRED’s independent investigation — of scraping sites that explicitly disallowed it, including by ignoring robots.txt, cycling user-agent strings, and routing requests through obscured IP ranges. The wall is only as strong as the crawler’s willingness to respect it.
Second, the wall keeps out people more effectively than it keeps out models. A headline and two teaser lines don’t convey the depth or value of the reporting behind them. If readers can’t feel the difference between your journalism and the free noise around them, they won’t pay for it.
Finally, when discovery happens upstream inside AI interfaces, most readers never even see the paywall. You can’t defend a fortress no one approaches. A paywall only works when people arrive at it.
Content Becomes Raw Material, Not Final Product
This is the real break with the past.
Content is no longer a final product. It has become raw material for systems that extract value without returning it. The meaning, nuance, and editorial judgement created by journalists are intercepted upstream before readers ever encounter the original work.
Inside this shift is a deeper threat: AI models learn from everything, not just high-quality reporting. They absorb low-effort content, conspiracy narratives, political propaganda, and automated junk at the same weight as verified journalism.
Social platforms already struggle to slow misinformation moving at human speed. AI moves at machine speed and at machine scale.
The competition is no longer other publishers. It’s an infinite stream of auto-generated material that no newsroom can match in quantity. And because AI interfaces flatten all inputs into the same tone and level of confidence, truth and fabrication arrive dressed identically.
The Business Model Question
If publishers wait for this disruption to “settle,” it will be too late. The path is already predictable: they don’t just risk losing the discovery channel; they risk losing the editorial voice, the reader relationship, and the economic engine that sustains journalism.
Everything upstream is being captured by systems that can generate text without creating meaning. What remains is the one part of the value chain nobody else wants: the cost of producing the content.
The industry can’t rely on hope or defensive posture. It has to flip the table and rebuild its position on its own terms.
My take on this is straightforward: publishers need to rethink where their power actually comes from and stop competing in a race they can’t win.
Quality journalism has to separate itself from the ocean of pseudo-news by acting as a unified bloc, not as fragmented outlets fighting for scraps. That starts with a shared standard of verification, visible provenance signals, and a common label that marks “this is real reporting” and makes everything else look cheap by comparison.
They also need strict data discipline: protect first-party reader data, shut down leaky dependencies, and push back hard against platforms that treat publisher data as raw extraction fuel. And instead of bending to whatever interface the big tech companies dictate, publishers should align with ethical tech partners who preserve attribution, respect editorial integrity, and don’t weaponise user data against the people who create the content in the first place.
In the next articles, I’ll map the risks, the viable strategies, and the concrete models that can help publishers regain leverage before the gap becomes irreversible. Stay tuned!