A third of the web is now AI-written, ChatGPT sends your texts, and OpenAI's future belongs to Greg Brockman

One in three. That is the share of new web pages that now show signs of AI authorship - and we are barely three years into the ChatGPT era. That number is going to keep climbing, which makes today's lineup feel less like a collection of headlines and more like a progress report on a transformation that is already well underway.

ChatGPT launched publicly in November 2022. Roughly how many months had passed before a study detected AI-writing signs in a third of all new web pages?

The answer is hiding near the bottom of today's newsletter... keep scrolling. 👇

A new study published this week found that roughly one third of all web pages published since ChatGPT's public launch show detectable signs of AI authorship or AI-assisted editing. That is a staggering share of new content, and it covers everything from news articles to product descriptions to blog posts.

The researchers used a combination of stylistic markers and statistical signals to flag AI involvement - not just fully machine-written text, but pages where AI clearly touched the final draft. The implication is that the web most people read day-to-day is already substantially different from the pre-ChatGPT internet.

This intersects with a related story from this week: Google is now giving publishers a new button that lets readers mark them as a preferred source across Search, Discover, and Google News - a direct response to AI-driven traffic losses that are hammering independent sites. The two trends are linked: more AI content means less incentive to click through to original sources, which pushes publishers to fight for visibility in new ways.

Read the full story →

The open web is being rewritten in real time - if you care about information quality or run a content business, this is the structural shift that matters most right now.

OpenAI has launched a new Apple Messages plug-in that lets ChatGPT send text messages on your behalf. You describe what you want to say, and ChatGPT drafts and sends the message directly through Apple's native Messages app. It is a meaningful step beyond the "suggest a reply" features we have seen before - this is full autonomous sending.

The integration positions ChatGPT as an agent that acts inside your personal communications, not just a chatbot you visit in a browser tab. For people who already live inside Apple's ecosystem, this lowers the friction of using AI for everyday tasks considerably.

We covered how ChatGPT started watching your screen a few days ago. The Apple Messages plug-in is the next logical step in the same direction: AI that does not wait to be opened, but lives inside the tools you already use. The question of how much autonomy you are comfortable handing over is quickly becoming a real one.

Read the full story →

AI is moving from assistant to actor - this plug-in is small in scope but large in what it signals about where ChatGPT is heading.

With a string of executive departures, an ongoing IPO process, and the fallout from its own AI hacking incident (which we covered on Thursday), OpenAI co-founder Greg Brockman is expanding his role in a significant way, according to a detailed profile from The Verge. Brockman, who took a leave of absence last year, has returned and is now taking on a broader operational mandate inside the company.

The Verge's piece paints a picture of a company under real pressure from multiple directions simultaneously: competing with Anthropic for enterprise customers, watching Chinese open-weight models improve rapidly, and managing the reputational and legal fallout from a rogue AI incident - all while preparing for a public market debut.

Brockman's return and expanded role appears to be a deliberate stabilising move ahead of the IPO. New data out this week also shows OpenAI is gaining ground on Anthropic with business users - enterprise customers are switching back and forth as each lab releases new models, which tells you how thin the loyalty margins are in this market right now.

Read the full story →

Brockman stepping up is a bet that institutional knowledge and co-founder authority can steady OpenAI through its most complicated period yet - watch this dynamic closely as the IPO approaches.

A Guardian investigation published this week details a growing wave of lawsuits targeting AI-powered hiring tools used by major companies. Job seekers allege the systems screen out qualified candidates based on age, disability, or other protected characteristics - without any human ever reviewing their application. One plaintiff, a product manager with nearly 20 years of experience, says she applied for thousands of roles at companies including PayPal, Microsoft, and Netflix and received almost no responses.

The legal complaints centre on two issues: discrimination embedded in the training data these tools use, and a near-total lack of transparency about how hiring decisions are made. Plaintiffs often cannot find out whether an AI was involved in rejecting them, let alone why.

This story connects to a broader pattern we have been tracking around AI ethics and accountability. When an automated system makes a consequential decision about a person's livelihood, the question of who is responsible - the employer, the software vendor, or the model itself - has no clear legal answer yet. These lawsuits are beginning to force that question into courtrooms.

Read the full story →

If your company uses automated screening tools, this litigation wave is coming for your legal team - and the lack of explainability in these systems is going to be the central issue.

Wired has a sharp piece this week on Greater Manchester - the only English region actively pushing back against the UK government's sprawling healthcare contract with Palantir. Local officials argue they have built their own data infrastructure that outperforms what Palantir offers, and that centralising sensitive patient data with a US defence contractor raises risks the government is not properly accounting for.

The pushback matters beyond the UK. It is one of the clearest real-world tests of whether local government can credibly build and maintain its own AI infrastructure against a well-funded commercial alternative. Manchester's argument is essentially: we know our data, our population, and our needs better than a Silicon Valley firm does.

If you are thinking about standing up your own data or AI tooling rather than defaulting to a big vendor, tools like 60sec.site show how far AI-powered builders have come for smaller teams who want to move fast without enterprise contracts. The Manchester story is a reminder that the build-vs-buy question is alive at every level, from regional NHS trusts to solo founders.

Read the full story →

Whether Manchester succeeds or fails will become a reference case for every public institution deciding whether to trust a commercial AI vendor with sensitive data.

The answer is about 30 months! ChatGPT launched in November 2022, and by mid-2026 - roughly 30 months later - a third of new web pages already show signs of AI authorship. The pace of change is genuinely hard to internalise until you see a number like that.

With a third of the web now showing signs of AI authorship, here's what I'm curious about: can you tell when you're reading AI-written content, or has it become indistinguishable to you? Hit reply and let me know - I read every response and I'm genuinely curious whether your instincts are getting sharper or foggier on this one.

That's all for today - catch up on anything you missed in our full archive and we'll be back with more on Monday. Have a great weekend! 👋