John Herrman

if ChatGPT is just Google Search results and Google Search results are just Reddit answers, who's flying the plane?

Everything from chatbots’ personification to their recurring issues with engagement-boosting sycophancy suggests that entertainment, not just utility, is what keeps a lot of users around. As a communications strategy, nudges project ambivalence about excessive platform use and raise the possibility that users may be misusing products. As a bit of marketing, they’re likewise helpful: Our product is so compelling we have to tell our users to take a break. Nudges are perhaps best understood, though, as attempts to demarcate liability. They imply concern for users’ well-being and offer tools to help them manage it. But they also make it clear that, should these new products drive you sort of insane, that’s ultimately your problem, not theirs. You were warned! Or at least you were nudged.

AI companies want us to think their products are like "The Entertainment" from Infinite Jest, capable of chatting us into terminal inertia
nymag.com/intelligence...

genuinely sorry about that

If you spend a lot of time on your smartphone - particularly if you spend a lot of time on social media — you've been nudged before. Apple's software is full of nudges, from reminders to turn down the volume or move the screen farther from your face to the aspirational behavior modification implied by its Screen Time monitor features; Google's smartphone software has its own "Digital Wellbeing" suite with similar features. TikTok allows its users to set daily limits and trigger reminders to take a break and to get some sleep; the app recently started testing "Well-being Missions," or tasks intended to "help people develop long-term balanced digital habits" and to "encourage and reinforce mindful behaviors."

New nudges also offer clues about how people are actually using these products. While OpenAI begins its announcement with an assurance that its "goals are aligned with yours" and that "instead of measuring success by time spent or clicks, we care more about whether you leave the product having done what you came for," lengthy engagements with chatbots - encompassing everything from therapeutic chat to fantasy role-play - are clearly a large and growing use case. Everything from chatbots' personification to their recurring issues with engagement-boosting sycophancy suggests that entertainment, not just utility, is what keeps a lot of users around. As a communications strategy, nudges project ambivalence about excessive platform use and raise the possibility that users may be misusing products. As a bit of marketing, they're likewise helpful: Our product is so compelling we have to tell our users to take a break. Nudges are perhaps best understood, though, as attempts to demarcate liability. They imply concern for users' well-being and offer tools to help them manage it. But they also make it clear that, should these new products drive you sort of insane, that's ultimately your problem, not theirs. You were warned! Or at least you were nudged.

as a tech company, you haven't *really* made it unless you're blaming your customers for using your product too much nymag.com/intelligence...

🤒

“the wife”

incredible things happening in the tesla subreddits

Generative-engine optimization - also known as answer-engine optimization, GEO, or, if you're feeling limber, LLMGEO - is best understood as an aspirational term, a way for marketers to assure their employers and clients that, in a world where people spend their days chatting with fast-developing Al chatbots that have eaten the entire web and can regurgitate it on demand, there are still ways to get an edge for their brands. Chatbots are trained on the web, continuously scrape the web, and often still link to the web; some have search features built in or call on search engines in the course of conversations with users. In other words, the optimizers still have hope, and early folk wisdom is taking shape, making the rounds, and finding its way into practice. "LLMs, like Google Al mode or ChatGPT, will use what is called a fan-out technique with lots of queries covering every angle," says Solis. "Then they will match these variations not with whole pages but with passages, or chunks," she said. In response to a question, in other words, a chatbot will tend to summarize and excerpt, with citations rather than prominent links. If you want to get cited, she says, you should publish content with that in mind. A lot of SEO-driven content "was very wordy," she said, which doesn't help with being scraped by AI. Now, she said, publishers should "structure the content in an easier way to be grabbed" - in citable chunks, with clear authorship.

Mass-pivoting SEO specialists have a partner in venture capital: In the last two years, dozens of startups have popped up and collectively raised hundreds of billions of dollars around chatbot analytics, optimization, and marketing. Profound, an analytics platform that promises to help clients get "mentioned" by Chat GPT, Gemini, Meta AI, Perplexity, and others - including Deepseek - has raised tens of millions of dollars from major VC firms and counts among its clients major international brands. "We're at the inflection point where people don't need to visit websites," says CEO James Cadwallader. "ChatGPT visits on my behalf, a new webpage is created, these are the citations, this is where it came from, and no one cares," he says. "Answer engines hijack, or steal, the relationship."

To help play it better, Profound offers something else to its clients. "Using a rote human brain to look through the data and do the content is no longer the path,, Cadwallader says. "You need to use technology now to create this content." That means Al-generated content created with new metrics and formats in mind. "We'll use state-of-the-art reasoning models to go crunch that data and replace client workflows, to create the content that's Al-optimized, very highly schemaed, and information dense." (For an idea of what AI-assisted, GEO-friendly content of the future might look like, look no further than Profound's own website or the outputs of virtually any "deep research" tool on offer by AI companies - it all reads roughly like the output of a chatbot). Al will help solve the problem of Al, in other words. With a "human in the loop," still, of course.

How is the SEO industry responding to the AI-driven collapse in Google traffic? By pivoting to "GEO," and thinking up ways to influence, trick, or introduce bias into chatbots — and, of course, by using AI themselves nymag.com/intelligence...

Yeah I see what you mean about the headline – slightly tongue-in-cheek! – but we also reported that ChatGPT is literally Googling, most best practices are the same, etc. Everyone still working with the same index

Let’s just say this would be an easy thing to deny and after a few days nether company did

I don't generally care to get preachy about "journalism" and what it is or isn't but I feel like a good baseline is the people a reporter interviews ought to exist

If OpenAI and Google Search are in bed together, the antitrust lawsuits just keep writing themselves.

No, totally. I do think a lot of clients are frantically looking for answers and need a story, though

This inevitable turn is why we spent so much time last year writing about the culture of the SEO industry - the concept was we had to describe the last days of disco clearly before everyone moved on!

such a great run of stories (linked the big ACL up top)

yeah, remixing so much of the same mega-corpus with similar techniques seems to be creating funny overlaps even without direct partnerships/off-books cross-scraping/generation

ha, we'll see!

In the search business model, three parties come out ahead: user, website, and advertisers (where G takes their cut) and it’s a non-linear relationship.

In chatbots, there’s no incentive to pay the content provider. Even with a mechanism like Cloudflare’s Pay-Per-Click, I just do not see it.

do we still tip @techmeme.com?

one factor seems to be that a lot of clients are going insane as well

But the traffic stats don't lie: It's a bloodbath. A web written to get attention from search machines may soon become a web written by machines for attention from Al, an index of crawlable footnotes written to be cited, of lists and diagrams yearning for a passing excerpt, or, perhaps, a moment in "position zero." Al firms, with comprehensive control over their users' experiences, will have leverage to simply charge for advertising, and nearly every one has announced its intention to do so. There will still be attention for clever marketers to skim from chatbots and Al search engines here and there, but competition for those leftover eyeballs will be intense. Everyone will be looking for an edge, and the world of GEO will claim to offer one. For his company, at least, Cadwallader sees a bright future. "If Jeff Bezos says bet on things that won't change," he says, "marketers will still be marketing."

the future of the web is so very bright that you can barely look at it directly

thanks Reyhan!

SEO Is Dead. Say Hello to GEO. Search-engine optimization now feels dated. Generative-engine optimization is all about trying to trick AI chatbots.

a weird and disorienting AI scoop in here: ChatGPT now appears to be using Google Search? Google declined to comment and OpenAI was like hey we love all our partners and everything with Microsoft is great by the way!! nymag.com/intelligence...

Like other SEO-to-GEO experts, he says that while traffic and mentions may be harder to get, much of what people should being doing is the same: publishing quality content that someone might actually find useful. He has deeper reasons to believe in continuity between the old and new worlds, however. Earlier this month, he published an investigation into ChatGPT's search functionality. OpenAI's partnership with Microsoft, along with various public statements, gave the impression that the company's products, when they needed to "search" the web, would use Bing, or search tools developed internally (for a while, he said, this led a lot of marketers to try to optimize for Microsoft's also-ran search engine, which had never previously cracked double-digit market share.) By creating decoy pages visible only to Google, lyer suggested he'd found "undeniable proof that ChatGPT uses Google search engine," querying OpenAI's biggest competitor before synthesizing results for its users.

It wouldn't be the first time an SEO specialist revealed something strange and surprising about how a major tech product actually works. In this case, lyer appears to be onto something. Asked about the experiment, OpenAI responded on background, emphasizing its ongoing relationship with Microsoft, suggesting that it used a variety of search providers, but not specifically denying that ChatGPT could be Googling. Informed of lyer's findings, Google declined to comment.

A former Google engineer told me that he'd discovered something strange: ChatGPT sometimes *uses* Google to retrieve websites now, not just Bing or internal tools. His evidence is solid, and neither OpenAI nor Google denied his findings when I asked

Generative-engine optimization - also known as answer-engine optimization, GEO, or, if you're feeling limber, LLMGEO - is best understood as an aspirational term, a way for marketers to assure their employers and clients that, in a world where people spend their days chatting with fast-developing Al chatbots that have eaten the entire web and can regurgitate it on demand, there are still ways to get an edge for their brands. Chatbots are trained on the web, continuously scrape the web, and often still link to the web; some have search features built in or call on search engines in the course of conversations with users. In other words, the optimizers still have hope, and early folk wisdom is taking shape, making the rounds, and finding its way into practice. "LLMs, like Google Al mode or ChatGPT, will use what is called a fan-out technique with lots of queries covering every angle," says Solis. "Then they will match these variations not with whole pages but with passages, or chunks," she said. In response to a question, in other words, a chatbot will tend to summarize and excerpt, with citations rather than prominent links. If you want to get cited, she says, you should publish content with that in mind. A lot of SEO-driven content "was very wordy," she said, which doesn't help with being scraped by AI. Now, she said, publishers should "structure the content in an easier way to be grabbed" - in citable chunks, with clear authorship.

Mass-pivoting SEO specialists have a partner in venture capital: In the last two years, dozens of startups have popped up and collectively raised hundreds of billions of dollars around chatbot analytics, optimization, and marketing. Profound, an analytics platform that promises to help clients get "mentioned" by Chat GPT, Gemini, Meta AI, Perplexity, and others - including Deepseek - has raised tens of millions of dollars from major VC firms and counts among its clients major international brands. "We're at the inflection point where people don't need to visit websites," says CEO James Cadwallader. "ChatGPT visits on my behalf, a new webpage is created, these are the citations, this is where it came from, and no one cares," he says. "Answer engines hijack, or steal, the relationship."

To help play it better, Profound offers something else to its clients. "Using a rote human brain to look through the data and do the content is no longer the path,, Cadwallader says. "You need to use technology now to create this content." That means Al-generated content created with new metrics and formats in mind. "We'll use state-of-the-art reasoning models to go crunch that data and replace client workflows, to create the content that's Al-optimized, very highly schemaed, and information dense." (For an idea of what AI-assisted, GEO-friendly content of the future might look like, look no further than Profound's own website or the outputs of virtually any "deep research" tool on offer by AI companies - it all reads roughly like the output of a chatbot). Al will help solve the problem of Al, in other words. With a "human in the loop," still, of course.

How is the SEO industry responding to the AI-driven collapse in Google traffic? By pivoting to "GEO," and thinking up ways to influence, trick, or introduce bias into chatbots — and, of course, by using AI themselves nymag.com/intelligence...

AI is killing the internet by Today, Explained

went on Today, Explained to talk a bit about life on the web as it gets completely devoured megaphone.link/VMP8234196136

Log in to X / X Log in to X to see the latest. Join the conversation, follow accounts, see your Home Timeline, and catch up on posts from the people you know.

x.com/search?q=cra...

Safe Superintelligence Inc. The world's first straight-shot SSI lab, with one goal and one product: a safe superintelligence.

sorry, it's ai twitter slang. basically means very good, exceptional, 100x, sort of crazy, etc ssi.inc

The excellent @jwherrman.bsky.social on Reddit and the end of the Internet (with a few notes of context on how the hell we got here from me) nymag.com/intelligence...

Writing this sort of memo in this style is, again, clearly intended as a signal: that the various people he's hired to revamp Meta's Al efforts are rubbing off on him; that people at other Al labs declining his billion-dollar offers should take him more seriously; that Meta's pivot to AI isn't a temporary diversion; to the Al world in general that he is one of them. At risk of overstating a formatting choice, to anyone mainlining Al news - particularly in the context of Zuckerberg's past venues for publishing essays, as Facebook Notes and Threads posts and official blogs - this sparse, black-on-white letter reads like yet another new outfit for the billionaire: Out with Rogan Zuck, in with "cracked" Al researcher Zuck. It's a clean start, see? Just look at the font! This was not lost on, for example, OpenAl employees:

The idea that Meta is charting a fundamentally different course for AI development and deployment than the companies it's following, mimicking, and poaching from is already a stretch. As has been the case with coding assistance, any monetizable feature that emerges from LLM development will be immediately and almost universally pursued by firms that have tens of billions of investment to recoup. But the promise that a Meta product will be personally empowering — that it will help you "be a better friend" or "become the person you aspire to be" - is profoundly unconvincing coming from Mark Zuckerberg, a notoriously cutthroat CEO who is one of the wealthiest people in the world and who as of February had a public approval rating far below Elon Musk's. AI CEO memos are strange and portentous documents to start with, hinting at either a coming era of unprecedented and terrifying disruption, collective executive delusion, or something in between. But the biggest issue with Zuckerberg's vision isn't one that can be papered over with another rebrand: It's that he's the one sharing it.

On Zuck's latest rebrand as a cracked plaintext AI guy nymag.com/intelligence...