You know Herman, I was looking for a specific recipe the other day, a very particular type of sourdough starter that uses local honey from the Jerusalem hills. I am talking about that deep, thyme-scented nectar you can only find near the Sataf springs. I spent about twenty minutes scrolling through these massive blog posts, jumping over advertisements for stand mixers I already own, and closing those annoying "join my newsletter" popups that blur the entire screen. I was trying to find the actual science behind the fermentation—how the specific glucose levels in that wild honey interact with the local yeast strains—not a three-thousand-word story about the blogger's childhood summers in a cottage. And I realized, halfway through my frustration, that I was still searching like it was two thousand eighteen. I was using these rigid, clunky keywords like "Jerusalem honey sourdough fermentation science," hoping the algorithm would be kind to me and filter out the fluff. I was treating the internet like a broken vending machine where you have to kick it just right to get what you want.
It is funny how muscle memory works, isn't it? Herman Poppleberry here, by the way. I still catch myself doing that "Pigeon English" thing that Daniel mentioned in his prompt. You know the vibe—just shouting nouns at a screen and hoping for the best. "Honey. Sourdough. Science. Jerusalem." It is like we are trying to speak a dead language to a machine that has already moved on to poetry. But today's prompt from Daniel is really hitting on a fundamental shift in the human-machine interface. He is asking about the transition from that keyword-based library catalog approach to the conversational, semantic search world we are living in now in February of twenty twenty-six. And honestly Corn, I think we are at the point where the traditional search engine results page—the "ten blue links" we grew up with—feels like looking at a dusty physical phone book while someone is trying to hand you a high-end smartphone.
It is a perfect analogy. Daniel mentioned in his message that a few years ago, search was just keyword matching and it was totally polluted by marketing through search engine optimization. It was an arms race between people trying to provide information and people trying to sell you a mattress. Now we have these massive language models that understand intent. But I want to push on that. Is it just that they understand what we mean, or is it that they are actually doing the heavy lifting of synthesis for us? Because when I finally gave up on Google and just asked a model about the sourdough, it did not give me a list of blogs. It gave me a three-paragraph explanation of how the specific enzymes in the honey—specifically the diastase—interact with the wild yeast to accelerate the breakdown of starches. No ads, no fluff, no life stories. Just the answer, synthesized from four different scientific papers and a baker's forum.
Right, and that is the "semantic" part of semantic search. In the old days, a search engine was essentially just a very fast, very literal index. It looked for the string of characters you typed and tried to find the pages where those characters appeared most frequently or most prominently. It did not know what a "sourdough" was; it just knew the letters S-O-U-R-D-O-U-G-H. But with modern models, we are talking about vector embeddings. Instead of matching words, the system is matching concepts in a high-dimensional mathematical space. If you talk about "something cold to eat in the heat," the system knows that "gelato" and "gazpacho" are conceptually close to that query, even if the word "cold" or "heat" never appears on the page. It understands the "shape" of your curiosity.
That is the part that still feels like magic to most people, even now. But Daniel brought up a great point about the "up-to-date" problem. For a long time, the knock on large language models was that they were frozen in time. They knew everything up to their training cutoff, but they did not know what happened ten minutes ago. He mentioned Gemini grounding as a solution that is already working. How does that actually change the stakes for the average user? If the model can now see the live web and verify its facts against the current second, does the blue link even matter anymore?
It changes everything because it solves the "hallucination" problem to a massive degree. This is what we call Retrieval-Augmented Generation, or RAG. When you ask a grounded model a question today, it does not just rely on its internal weights or what it "learned" two years ago. It actually goes out, performs a traditional-style search in the background, pulls in the most relevant snippets of text from the current internet—news sites, weather reports, stock prices—and then uses its reasoning capabilities to synthesize an answer based on those fresh facts. So, you get the best of both worlds: the massive, real-time index of a search engine and the conversational synthesis of an artificial intelligence. It is like having a research assistant who can read the entire internet in half a second and then explain it to you over coffee.
So if the AI is doing the "search" in the background and just giving me the "answer" in the foreground, what is the purpose of the search engine itself? Daniel asked if there is still a role for a search engine that discovers and links websites. I mean, if I never see the website, does the "link" even exist for the user? If I am getting the "hummus plan" for my afternoon in the Old City directly from the AI, I am not clicking on a "Top Ten Hummus Spots" article. I am just walking to the shop the AI suggested.
That is the existential question for the open web, Corn. I think we have to decouple the "search technology" from the "search interface." The technology—the crawlers that find new pages, the indexers that organize them, the spiders that navigate the dark corners of the internet—that stuff is more important than ever. The AI needs that data to stay grounded. It needs a map of the world to talk about the world. But the "interface"—the page with ten blue links and five ads at the top—that is what is dying. We are moving from a "search engine" to a "discovery engine" or an "answer engine." The search engine becomes the plumbing, and the AI is the faucet. Most people don't care about the pipes; they just want the water.
But wait, Herman, if the interface dies, the business model dies, right? This is the part that worries me. If I do not click the link, the website owner does not get a visit. If they do not get a visit, they do not make ad revenue. If they do not make money, they stop writing about Jerusalem honey sourdough science or testing the best hummus spots. Then the AI has nothing new to crawl. It is like the AI is eating the seeds of the very garden it needs to survive. Is this a parasitic relationship, or is there a way forward that doesn't end in a "content desert"?
You are touching on the "circularity problem" of the AI-driven web. If the AI eats the content and spits out the answer without sending traffic back, it eventually starves itself. But I think we are going to see a shift in what a "link" looks like. Instead of a list of results, we are seeing citations. If you look at how some of the newer search tools work, they give you an answer, but they have these little footnotes or sidebars. You can click those to see the source. It is more like a research assistant than a librarian. The librarian tells you which aisle the book is in; the research assistant reads the book and tells you the answer, but keeps the book open on the desk in case you want to check their work. The value for the website owner might shift from "ad impressions" to "licensing fees" or "referral credits." We are already seeing some major publishers sign deals with AI companies to allow their data to be used for grounding in exchange for a cut of the subscription revenue.
I like that distinction. But let's talk about the "Pigeon English" thing again. Daniel said he would choose the conversation every time. I agree for complex queries—like "help me plan a three-day hike in the Galilee that avoids steep inclines and ends near a winery"—but what about simple navigation? If I want to go to my bank's login page, I do not want a conversation. I do not want the AI to say, "Hello Corn, I see you want to manage your finances. Banking is a vital part of modern life..." No! I just want the link. Is there a risk that by making everything a "conversation," we are actually making simple tasks more tedious?
Absolutely. There is a concept in user interface design called "interaction cost." Sometimes, typing a single keyword and hitting enter is the lowest interaction cost possible. If I want to see the score of the Beitar Jerusalem match, I do not need a three-paragraph synthesis of the game's momentum and the historical significance of the rivalry. I just want "two to one." The future of search engines isn't just "chatting." It is about intent recognition. The system should know when I want a quick fact, when I want a navigation link, and when I want a deep, synthesized explanation. The best AI search won't always talk back; sometimes it will just show me the button I need to press.
So, to Daniel's question about decoupling the technology from the interface: could we see a world where search is just an invisible utility layer? Like, I am writing an email to a friend about a trip, and the "search engine" is just there, providing facts and links as I type, without me ever "going to a search engine" at all?
That is exactly where it is going. We are moving away from "destination search" to "embedded search." Think about how we use "Find My" on our phones or how we search for files on a computer. It is just a feature of the operating system. In the future, the "web" won't be a place you go to via a browser; it will be a data source that your local AI agent queries constantly. The "search engine" becomes the "API for the world's information." You might be wearing augmented reality glasses, and as you look at a building in the Old City, the search engine is running in the background, identifying the architecture, checking the historical records, and whispering the date it was built into your ear. You didn't "search" for it; the search happened because you looked at it.
That brings us to the SEO industry. Daniel asked if it is already redundant. I mean, for twenty years, people have made millions of dollars figuring out exactly how to make Google like their website. They used keywords, backlink strategies, and "cornerstone content." If Google is no longer the primary way humans consume information, what happens to all those people? Is SEO dead, or is it just evolving into something even weirier?
Oh, it is definitely not dead, but the old tactics are becoming useless. We are moving from Search Engine Optimization to what people are starting to call GEO—Generative Engine Optimization. Or AIO—Artificial Intelligence Optimization. The goal is no longer "how do I get to rank number one for this keyword?" The goal is now "how do I ensure that when an AI model synthesizes an answer about this topic, it includes my data and cites me as the authority?" It is a much more sophisticated game. You can't just buy a bunch of low-quality links and expect to win.
How do you even do that? You can't just stuff keywords into a meta-tag anymore. The AI is looking for meaning, for authority, for unique data. It is reading the content like a human, but a human who has read everything else too.
Exactly. You do it by being high-quality and highly structured. AI models love structured data. If you have clear headers, well-organized tables, and unique insights that aren't just copies of other websites, the model is more likely to "value" your content in its latent space. It is also about brand authority. If you are cited by other authoritative sources, the AI "trusts" you more. In a way, it is actually a return to what search was supposed to be before it got gamed: a meritocracy of information. If you provide the most helpful, verifiable answer to a specific problem, the AI will find you and use you as a source. The "spammy" side of SEO—the people who just built link farms and keyword-stuffed landing pages—they are in serious trouble. But the people who understand how to structure information? They are going to be more valuable than ever.
That sounds optimistic, Herman. But I worry about the "black box" nature of it. With Google, we at least had some idea of how the algorithm worked—we knew about PageRank, we knew about mobile-friendliness. With a large language model, even the engineers don't always know why it chose one source over another in a specific synthesis. If I am a small business owner in the Old City, how do I "optimize" for an AI that might have a hidden bias or just happens to like my competitor's writing style better?
That is the big challenge. Transparency in these models is going to be the next big regulatory battleground. We are already seeing calls for "algorithmic accountability." But from a technical perspective, the "optimization" is going to be about providing the best "answers," not the best "keywords." Daniel mentioned that the search results were often "polluted by marketing." In the AI era, marketing has to look like helpfulness. If your website is purely sales fluff, the AI will probably ignore it because it doesn't help the user's query. But if your website provides a really useful tool, or a deep analysis, or a unique dataset, the AI will find it very attractive as a source. It is about being a "node of truth" in a sea of noise.
So, is the SEO industry redundant? I would say the "spammy" side of it is. But I think we will see a new kind of "search consultant" who focuses on "model influence." They will look at how different models—Gemini, GPT-five, Claude four—perceive a brand. They might even use AI to test how other AIs respond to their content. It is going to be a very strange, high-speed game of cat and mouse.
I agree. I think we will see "Model Audits" where a company pays a consultant to find out why they are being left out of the AI's synthesized answers. The consultant might say, "Well, the models think your data is unreliable because your citations are circular," or "You need to use more Schema-dot-org markup so the AI can parse your pricing table more accurately." It is SEO, but with a PhD in linguistics and data science.
Let's go back to Daniel's point about the "library catalog" versus the "conversation." He said he would choose the conversation every time. But I think there is a hidden danger there that we should address. When you look at a library catalog, you see the breadth of what exists. You see the books next to the one you were looking for. You see the different perspectives. In a "conversation" with an AI, you often get a single, unified answer. Do we lose the "serendipity" of search? Do we lose the ability to see the dissenting opinion? If I ask about a controversial historical event in Jerusalem, and the AI gives me a "balanced" synthesis, I might miss the raw, passionate arguments from both sides that I would have found by clicking through five different websites.
That is a profound point, Corn. The "synthesis" can feel like "consensus," even when there isn't any. It smooths over the cracks. If I ask an AI "what is the best way to solve this economic problem," and it gives me a very confident, synthesized answer, I might not realize that there are five other major schools of thought that disagree. The traditional search engine, for all its flaws, forced you to see multiple sources. It forced you to do a bit of the synthesis yourself. The risk of the conversational future is that we become intellectually lazy. We stop looking for the "library" and just listen to the "librarian." We might get the "what," but we lose the "why" and the "who else says otherwise."
And that is why I think the "traditional" search engine—or at least the ability to see the raw index—won't go away entirely. Power users, researchers, and people who actually care about the truth will always want to "see the receipts." We might start our journey with a conversation, but we will often want to end it by looking at the source material. I think the search engines that survive will be the ones that make it easiest to toggle between the "answer" and the "sources."
I think you are right. We are going to see a "layered" search experience. Layer one is the AI agent that gives you the quick answer. Layer two is the synthesis with citations. Layer three is the deep dive into the original sources. The "search engine" of the future is the system that manages all three of those layers seamlessly. It is not about replacing the links; it is about putting them in context.
So, if we look at the next five years, what does the daily experience look like? I'm sitting here in Jerusalem, I want to find a new place to get hummus that isn't the usual tourist spots. I don't go to a search bar. Do I just talk to my glasses? Do I just think it?
It is probably a mix of things. You might ask your voice assistant, "Hey, where is the best hummus within a ten-minute walk that has a courtyard and is open on a Tuesday?" The AI doesn't just give you a link to a Yelp page. It says, "There is a place called Abu Shukri, but for a courtyard vibe, you should try this hidden spot in the Christian Quarter. Here are some photos people took yesterday, and by the way, they changed their hours last week, but I checked their latest social media post and they are definitely open today. Do you want me to put it on your map and check the walking route for construction?"
See, that is the "grounding" Daniel was talking about. It is pulling from multiple live sources—social media, reviews, maps, news—and giving me the "hummus plan," not just a "hummus search." It is agentic. It is doing the work for me.
Exactly. And the "search engine" is the thing that facilitated that entire transaction in the background. It found the social media post, it found the map data, it found the reviews. The "traditional" search engine is becoming the "nervous system" of the internet, while the AI is the "brain." You don't interact with your nervous system directly; you interact with your brain. But without the nervous system, the brain is blind and deaf. The technology is more vital than ever, but the interface is becoming invisible.
That is a great way to put it. So, to Daniel's question about whether we can decouple the technology from the interface—we aren't just able to do it, we must do it for the web to survive. We need the technology to be more robust than ever to support these complex AI queries.
And I think we are going to see a lot of competition in that "invisible" layer. Companies like Google, Microsoft, and even newcomers like Perplexity or OpenAI are all fighting to be that "nervous system." The winner won't necessarily be the one with the best website, but the one with the best "grounding" capabilities. The one that can provide the most accurate, real-time data to the AI models. If an AI uses a search engine that gives it old data, the AI looks stupid. So the pressure on search engines to be fast and accurate is actually higher now than it was in the keyword era.
It is also worth thinking about the "local" aspect of this. Here in Jerusalem, search is often about physical reality—traffic, weather, security, opening hours of tiny shops that don't even have a website. The "search engine" of the future has to be hyper-local. It has to understand that "is it safe to go to the Old City right now?" is a question that requires a very different kind of search than "how do I bake a cake?" It needs to pull from police reports, news feeds, and even public cameras or sensors.
Right. Semantic search understands the "gravity" of a question. It understands that some questions are trivial and some are critical. The future of search is "context-aware." It knows where you are, what time it is, and what your past preferences are. If I search for "coffee," and I'm at home, it might show me how to fix my espresso machine. If I'm on Jaffa Street, it shows me the nearest cafe with a high rating from people who like strong dark roasts. It is moving from "what did you type?" to "what do you need right now?"
It almost sounds like the "search engine" is becoming a "life assistant." But that brings up the privacy concerns Daniel hinted at when he mentioned being "privacy conscious" and using DuckDuckGo. If the search engine knows my context, my location, my preferences, and my past behavior, that is a lot of data. Can we have this "semantic, conversational future" without giving up every shred of our privacy? Can I have a "hummus plan" without the AI knowing my exact GPS coordinates for the last five years?
That is the trillion-dollar question for the next decade. There is a massive push toward "on-device" AI, where the model and your personal data live on your phone, and it only sends "anonymized" or "masked" queries to the big search engines. That could be the middle ground. Your local AI knows you—it knows your favorite hummus and your walking speed—but the "search engine" in the cloud only sees a request for "best hummus in Jerusalem" without knowing exactly who is asking or why. We are seeing technologies like Private Cloud Compute and differential privacy becoming standard.
I hope so. Because as much as I love the convenience of the "conversation," I don't want the "librarian" to be keeping a permanent log of every single thing I've ever been curious about. I want to be able to ask a weird question without it following me around in the form of targeted ads for the next six months.
Well, the librarian might be a robot, but the robot's boss is still a corporation. We have to keep that in mind. But overall, I think Daniel's instinct is right. The "Pigeon English" era is ending. We are finally learning how to talk to our machines in our own language, and more importantly, they are finally learning how to listen. We are moving from a world where we had to learn the machine's language to a world where the machine has finally learned ours.
It is a massive shift. I think back to the early two thousands, typing "AOL keyword: something" into a box. We've come so far, yet the fundamental human desire is the same: we want to know something, and we want the most direct path to that knowledge. The path used to be a map; now it is a guide.
And that path is no longer a list of links. It is a dialogue. It is an exploration. It is, as Daniel said, a conversation. The search engine isn't dying; it is just finally growing up and becoming the intelligent partner we always wanted it to be.
I think that is a good place to wrap this one up. It is a lot to think about, especially for anyone who has built a career or a business on the old rules of the internet. The rules are changing, but the opportunity to provide real value is still there. If you are a creator, don't worry about keywords; worry about being the best possible source for that "nervous system" to find.
Exactly. Focus on being the best source of truth, and the AI will find you. That is the new SEO. It is about quality, not trickery.
Well, if you have been enjoying "My Weird Prompts" and our deep dives into these topics, we would really appreciate it if you could leave us a review on your podcast app or on Spotify. It genuinely helps other people discover the show and keeps us going. We are a small team, and every review counts.
It really does. We love hearing from you all, and your feedback helps us decide which prompts to tackle next.
You can find all our past episodes, including our episode memory and more context about the show, at myweirdprompts dot com. We have an RSS feed there for subscribers and a contact form if you want to get in touch with your own weird prompts. You can also reach us directly at show at myweirdprompts dot com.
And remember, we are available on Spotify, Apple Podcasts, and pretty much wherever you get your podcasts.
Thanks to Daniel for the prompt today. It definitely gave us a lot to chew on, much like a good piece of sourdough.
It did. Always good to talk it out with you, Corn. I'm going to go see if I can find that honey you were talking about.
Good luck, it is worth the hike. Alright everyone, thanks for listening to My Weird Prompts. We will see you in the next one.
Goodbye everyone!
