Doc Searls Weblog

Hyperlinks subvert hierarchy
—Thesis #7, The Cluetrain Manifesto

Big AI subverts everything, including hyperlinks, which are what make the Web a web.

With Big AI, you no longer surf from searches to sources across an ocean of links. You ask questions and get answers from the world’s largest Magic 8-Balls. They top the new hierarchy, which subverts and subordinates the Web.

Go to any of the Big AI chatbots—ChatGPT, Claude, Gemini, Meta, whatever—and ask a question. If it pauses for a moment, there’s a good chance it’ll say “Searching the Web.”

But it’s not. It’s composing an answer synthesized from what it has harvested from the Web, plus a vast amount of other forms of ingested human expression. It may or may not point back to sources on the Web. And if it does, it’s only providing footnotes.

Pew reports that people reading AI summaries of Google searches click on links only about 8% of the time. When search results come without an AI summary, that number is 15%. But both numbers are down from 100%, back when links were all that search engines produced. As a result of that change, publishers report losses of 20-90% in traffic and revenue in the past year alone. Many small publishers are now gone.

And what about the services and institutions that kept the Web both ours and durable (such as those with logos in the windows of the abandoned house image above)? Keep that question in mind while I ask Google’s new AI Mode to compare itself, side-by-side, to Google Search:

The stuff in the Google Search column is being steadily deprecated by what’s happening in the Google AI Mode column. Gemini explained that to me in January, when I interviewed it for Where Are We? That post begins,

While the Web isn’t a place, we know it as one. The language we use to describe it is concrete and structural: domains at locations, with sites that are built and have pages that are authored, and contents that are uploaded and downloaded.

For a generation or more, we have searched through the Web’s vast and structured library using Google, whose original mission was “to organize the world’s information and make it universally accessible and useful.” When we searched Google, we relied on its PageRank system to bring us lists of pages ranked by the number and quality of inbound link sources, and how likely a given site was to contain keywords, text strings, and other search terms.

That’s now gone. Old pages are missing from many searches. AI guesswork is replacing formerly normal search results. The look of Web structure in search results is just emulation.

This change matters a lot to me. I’ve been writing and publishing on the Web for thirty-one years, and almost all my work is still out there on solid static sites. And now Google fails to find many of those pages, despite having been on the Web for decades and having lots of inbound links. My friend Dave Askins tells me the same malady afflicts much of his own archival work: it’s present on the Web, but absent from Google searches. Why?

I asked Google’s Gemini that question. Here’s what it said:

The “disappearing” of the old web is a result of a massive shift in Google’s priorities over the last few years.

While you might be looking for a specific historical document or an old forum thread, Google’s systems are now aggressively tuned to filter out what they consider “stale” or “unhelpful” content…

In the race to be a ‘helpful assistant,’ Google has stopped being a ‘universal library.

PageRank is still there, it tells me, but subordinated. To what?

Of course: advertising.

Taking my clues from Keyords are dead, by Shalom Gonzalez in Search Engine Land, I asked Gemini to tell me “how searches by the helpful assistant might constantly be looking for what the person might eventually want to buy, so they can be advertised at when the right time comes.”

Gemini replied,

This is the “secret sauce” of the 2026 search model. If the old search engine was a librarian, the new one is a personal shopper who is taking notes on everything you do to predict your next “need state. The transition you’re sensing is the move from Reactive Search (answering what you asked) to Predictive Search (answering what you’re going to ask).”

This fucks the Web.

And you too. You are no longer the reader, the researcher. Google thinks what you want from the Web’s library no longer matters. Nor does the library itself. Google search is now guesswork based on what its AI thinks really matters, plus what it thinks you might want to buy. These assumptions are based on what Eli Pariser, in The Filter Bubble, calls “a bad theory of you.” Only worse, because it’s now also a bad theory of the Web. Here’s Gemini again:

While you might be looking for a specific historical document or an old forum thread, Google’s systems are now aggressively tuned to filter out what they consider “stale” or “unhelpful” content. Here is why those old archives are vanishing from your search results

OMFG.

Sixty-five years ago, The Twilight Zone aired an episode about what’s happening here. It was called “To Serve Man” and ended this way:

Everybody in the surveillance-fed advertising fecosystem already regards personal privacy as a bug, not a feature. With Big AI, the plan is to modernize that fecosystem by moving human cattle onto corporate ranches, where they can be observed more closely than ever, and advertised at with far more accuracy. This, the thinking goes, should multiply the size of the $1 trillion advertising business.

Here is how Meta tempts me to move onto its private AI ranch. Unlike my friends there, I’m not going. (Though I am on Facebook, for other—yes, morally compromised—reasons. That’s where this ad showed up, when I had just closed a creepy Meta-generated AI “reel.”)

For a look at how the new AI-expanded fecosystem works, see—

• Google is Now a Buyer’s Assistant, by Denine Harper, and
• Chatbot Ad IDs Share Data To Google, Microsoft, Other Analytics Providers, by Laurie Sullivan in MediaPost.

Laurie’s main source is Tracking Conversations: Measuring Content and Identity Exposure on AI Chatbots, a report by four researchers at UC Davis. Among much else, they say,

We find that 17 of 20 chatbots share information with at least one third party. Three chatbots share plaintext conversation text, including both prompt and response snippets, with Microsoft Clarity through session replay. Fifteen chatbots share conversation URLs or chat identifiers with third-party advertising, analytics, or social endpoints. Several chatbots expose user identity through support widgets, analytics, advertising, and session replay tags; in some cases, hashed emails are shared.

Here is how those data flows look:

This is exactly how things already work for most websites (run a PageXray on any site or page to see a visualization of personal data flows that looks much like the one above). The difference is that websites at least throw a cookie notice in your path, either to force consent to being tracked or to prompt you to click on “choices” that might stop some tracking. (Which, mostly, they don’t. You get tracked anyway.) With Big AI, all this tracking is already permitted by oxymoronic “privacy” policies you’ve don’t read.

We can only fight this by building an intention economy that’s bigger than the attention one, and to make personal privacy a feature rather than a bug. In 2012, Harvard Business Review Press published a book about making that economy happen, called The Intention Economy: When Customers Take Charge.

It said there are two ways. One is by replacing empty corporate promises with privacy contracts that people proffer, and sites and services agree to. The other is by building on genuine trust with scalable ways for customers and companies to do business and communicate with each other. Many more of those ways can be imagined when customers have full agency rather than what little they get as human cattle on corporate ranches.

We now have a standard on which an intention economy can be based: IEEE 7012-2025, nicknamed MyTerms.

We also have a way for customers and companies to start working together toward making the intention economy happen, based on MyTerms. It’s the IEEE’s Industry Connections program and its Individual Defined Privacy Terms Roadmap:

Our (Customer Commons, MyData Global, and the MyTerms Alliance) goal is to bridge the gap between the 7012-2025 standard and real-world adoption. The Industry Connections activity can help define and pilot practical training, develop strategic road maps, identify and implement certification programs, and document an industry white paper advancing standard awareness and understanding.

The motivation is simple: the standard has major implications across industries and consumers. We need a robust, yet agile framework for a rapid market introduction, iterated with expert feedback, ensuring broad stakeholder buy-in.

You can join by filling out this form, and by encouraging others to do the same, starting with the services and institutions in the top image above.