On March 20, The Verge reported that Google is running an experiment replacing news headlines in search results with AI-generated ones — after quietly doing the same in Google Discover since January. The same day, the Wall Street Journal reported that OpenAI plans to merge ChatGPT, Codex, and its own browser into a desktop "super-app." Two platforms. Two approaches to the same problem. Google is rewriting what publishers produce. OpenAI is building a world where nobody needs to visit them at all.
The Headline
Google's relationship with publishers has been a slow extraction. Each phase took something more.
- 2000s Google indexes publisher content and links to it. Publishers get traffic. The exchange works.
- 2010s Google shows snippets, knowledge panels, and answer boxes. Some queries resolved without a click. Publishers lose traffic at the margins.
-
MAY 2024Google rolls out AI Overviews — AI-generated summaries above search results. Publishers warn it will threaten their business models.
- Apr 2025 Bloomberg: publishers say traffic has plummeted since AI Overviews launched.
- Jun 2025 Similarweb: Google's AI tools are significantly cutting organic search traffic to publishers.
- Jul 2025 Similarweb: since AI Overviews launched, news search clicks have declined. A Pew study confirms AI Overviews reduce clickthrough rates.
-
MAR 2026Google tests replacing publisher headlines with AI-generated text in search results.
The progression is worth naming precisely. Snippets summarized the article. AI Overviews answered the question. Headline replacement erases the publisher's voice from the result itself. The user no longer sees what the journalist wrote — they see what Google's AI decided the journalist meant. The publisher's most compressed expression of editorial judgment, the headline, is overwritten before anyone reads it.
This is different from showing a summary above the results. A summary adds content. A headline replacement removes it. The publisher's identity — their word choices, their framing, their editorial perspective — is replaced by a generic AI paraphrase. The Verge's own headline for this story calls it a "canary in the coal mine." That headline, on Google, might be rewritten to something like "Google tests AI headlines in search" — accurate, stripped of voice, functionally identical to what any other publication would write.
The Browser
OpenAI's approach to the same problem is more direct: eliminate the search step entirely.
If ChatGPT is how users find information, and Codex is how they write code, and the browser is how they access the web — and all three are one application — then the user's entire digital workflow happens inside OpenAI's interface. Search becomes a feature of chat. Browsing becomes a feature of the agent. The publisher's website is something the AI visits on the user's behalf, extracts what's relevant, and presents as a response.
Google's experiment says: when users search, they'll see Google's version of the headline instead of the publisher's. OpenAI's super-app says: users won't search at all. They'll ask. The distinction matters because it determines whether publishers are rewritten (Google) or bypassed (OpenAI). The outcome is the same — the publisher becomes invisible to the user — but the mechanism differs.
And the same day, OpenAI told MIT Technology Review it plans to build "an autonomous AI research intern" by September. The company isn't just consolidating its consumer products. It's building AI that does AI research — inside the same platform. The super-app is not just a product. It's the container for a self-improving system.
The Levy
Also on March 20, Mistral CEO Arthur Mensch proposed a revenue-based levy for AI model providers in the EU — a tax on AI companies proportional to the content they consume. The proposal would require AI companies to pay publishers for the training data that makes AI summaries and headline replacements possible in the first place.
The timing is not coincidental. The EU has been the jurisdiction most willing to force tech platforms to compensate content creators — from the Copyright Directive to the Digital Markets Act. Mensch's proposal is the first serious attempt to price what Google is doing to headlines and what OpenAI is doing to browsing: the consumption of publisher content without the traffic that used to be the payment.
Whether the levy arrives in time to matter is the question. Google's headline experiment is already running. OpenAI's super-app is already planned. By the time a revenue-based levy clears the European Parliament, the publishers it's designed to protect may have already lost the traffic that made them viable.
Two Roads
The publisher business model has always rested on a transaction: create content, attract attention, monetize that attention through subscriptions or advertising. Google and OpenAI are dismantling different halves of that transaction.
Google is dismantling distribution. For twenty years, search was how readers found articles. AI Overviews reduced the clicks. Headline replacement reduces the identity. The publisher's content still exists, but the pathway to it — the search result with the publisher's name and words — is being rewritten into something generic, something that could have come from anywhere.
OpenAI is dismantling the visit. The super-app doesn't rewrite the headline. It makes the headline unnecessary. The user asks a question, the AI provides an answer, and the article that informed that answer is never seen. No headline. No click. No visit. No ad impression.
March 20 is the day both approaches appeared on the same page. Google editing what publishers write. OpenAI building a product where nobody reads what publishers write. Two platforms, competing with each other, converging on the same future — one where the content layer that sustains journalism, analysis, and reporting becomes invisible to the people who consume it.