AI search didn’t just change SEO. It quietly removed the economic reason most informational content existed in the first place — and that should worry the LLMs more than it worries anyone else.
“There are some decades where nothing happens, and then there are weeks when decades happen.” — Lenin
Lenin was talking about revolutions. I think about that quote every time I open ChatGPT or Claude now.
For about twenty years, the open web ran on a deal that almost no one wrote down but everyone in marketing understood. A user types a question into Google. Google sends them to a page. The page answers the question, and on the way to the answer, the publisher gets a chance to show an ad, build brand affinity, capture an email, or pitch a product. Informational content was the entry ticket. Monetization came later in the funnel.
That deal is breaking in front of us, and most companies are still pretending it isn’t.
The trade that built the web
If you’ve ever been told to write a blog post titled “What is SEO?” or “What is an LLM?” or “How are footballs made?” , you have lived inside the trade. The post itself rarely paid for itself directly. Nobody buys software off a definitional explainer. The post existed because it ranked, and ranking pulled strangers into a property where the company could finally introduce itself.
The classic move was the listicle. “Best CRM software in 2024.” “Top 10 project management tools.” The company writing the post was, conveniently, also #1 on the list. It was a soft con, but a productive one: the user got a real comparison, the publisher got a real lead, the search engine got a real answer to index. Every party in the system had an incentive to keep producing more of it.
Multiply that across every B2B SaaS company, every DTC brand, every media site, every affiliate. That is most of the written internet of the last two decades. Top-of-funnel content was never really about education. It was a customer acquisition channel that happened to be educational.
What AI search actually broke
ChatGPT, Claude, Perplexity, Gemini — they all do the same thing from the user’s perspective. They answer the question. Inline. With citations sometimes, often without. The user never lands on the page that wrote the answer.
This sounds like a small UX change. It is not. It is a structural break in the economics of informational content.
If a user can ask “what’s the difference between RAG and fine-tuning” and get a competent two-paragraph answer in the chat window, the marginal value of writing the 47th blog post titled “RAG vs. Fine-Tuning Explained” collapses. The traffic doesn’t arrive. The lead doesn’t get captured. The product never gets pitched. The post becomes a write-only exercise, useful as training data for the model, useless as a customer acquisition channel for the author.
The incentive that funded the production of informational content for twenty years has been quietly removed. Not banned, not deprecated. Removed.
The competition layer disappears too
The thing people miss is that this isn’t just bad for individual publishers. It is bad for the long-run quality of the answers themselves.
The reason there are good explainers on the web is that ten companies fought to write the best one. Competition for the click is what produced the second draft, the better example, the clearer diagram, the updated 2024 numbers. When the click goes away, so does the competition. So does the second draft.
Informational content won’t die completely, universities will still publish, hobbyists will still write, a handful of companies with strong brand budgets will keep doing it for reasons unrelated to immediate ROI. But the volume and the rate of improvement both fall off a cliff. The web stops getting better at explaining itself.
The existential problem this poses for the LLMs
Here is the part that should be keeping the model labs up at night.
LLMs are, in a real sense, fine-tuned on the output of the very system they are dismantling. The reason a chatbot can give you a competent answer about RAG, or footballs, or the Treaty of Westphalia, is that thousands of humans were paid, directly or indirectly, to write good explainers, because the click economy made it worth their while. Take away the click economy and you take away the production line.
The first generation of frontier models was trained on a web that was being aggressively cultivated for human readers. The next generation will be trained on a web where the incentive to cultivate has been pulled. You can see where this goes. Models trained on models trained on models, with fewer and fewer fresh human explainers entering the corpus. The technical term is model collapse. The marketing term is “why do all the answers feel the same now.”
The chatbots need the open web more than the open web needs the chatbots, and the open web is figuring that out slowly.
What this means if you’re a marketer right now
If you’re running content for a company in 2026, the honest answer is that a lot of what worked in 2022 doesn’t anymore, and pretending otherwise is expensive. A few things I’d push on:
Stop measuring TOFU content by traffic alone. Generic explainers that used to pull search traffic will keep losing it. The pages that still earn their keep are the ones with a point of view, original data, or a perspective the model genuinely cannot synthesize from the rest of the web.
Take brand authority seriously as a distribution strategy. When a user asks an LLM “what’s the best tool for X,” the model surfaces the brands it has the strongest, most consistent signal about. Getting mentioned in the answer is the new ranking. That mention is earned through the same things brand has always been earned through — PR, original research, podcast appearances, partnerships, customers talking about you in places models read, just with a much sharper payoff than before.
Write less, and write things only you can write. A weekly explainer of yesterday’s news loses to the chatbot every time. A field report from your customer base, a benchmark you ran, a contrarian take from someone with skin in the game, those still work, because they’re primary sources. The model needs them. So do humans.
Treat being cited by LLMs as a real channel and instrument it. Track which models mention you, in what contexts, with what framing. This is the new SERP and almost no one is monitoring it yet.
The honest summary
Top-of-funnel content didn’t die because AI is better at writing it. It died because the click, the thing that paid for it, stopped arriving. The chatbots ate the entry ticket and left the rest of the funnel intact, which is the part nobody seems to be saying clearly.
It’s a strange moment. The companies that own the chatbots are also the ones quietly eroding the supply of training data they depend on. The companies that produce the training data are watching their distribution model evaporate. The users are getting faster answers and a thinner web. Nobody is obviously winning except, in the short term, the inference layer.
Lenin’s line keeps coming back to me because this really is one of those weeks-where-decades-happen moments. The mechanics of how content gets produced, distributed, and rewarded on the internet are being rewritten in real time, and the playbooks built for the previous regime aren’t going to survive contact with this one.
If you’re trying to figure out what your content and brand strategy should look like in the AI-search era — what to keep doing, what to kill, and how to actually get cited by the models that are now sitting between you and your customers — that’s the question we work on at Zeta Theorem. I’d be happy to talk.
Leave a Reply