The editorial argues the SERP has become a data-collection and monetization surface rather than a retrieval surface. Every layout choice — AI Overviews at the top, indistinguishable sponsored results, the Discussions module surfacing Reddit content Google is training on — is explained by the twin goals of keeping users inside google.com for more ad impressions and harvesting query reformulations as Gemini training signal.
In screenshot form, sancho documents a typical late-2026 SERP: AI Overview, People Also Ask, shopping carousel, Reddit block, YouTube block, Discussions block, four ads — with the actual answer link pushed below the fold. The implicit argument is that the page is no longer designed to answer the query but to trap attention and cannibalize the open web Google indexes.
The dominant thread reaction is not defense of Google but matter-of-fact migration stories: Kagi for general search (18+ months in with no regrets), Perplexity for research, Claude for code. The consensus framing is that Google is now a single-purpose tool for local/commercial lookups ('finding a restaurant') rather than the default entry point to the web.
A recurring resigned comment is that users now append 'site:reddit.com' to nearly every query to escape SEO-farm content. The position is that if the most reliable way to use Google is to constrain it to a single third-party forum, the core ranking product has already failed — and Google's own 'Discussions and forums' module is a UI-level acknowledgment of that failure.
The post flags that the long tail of low-quality SEO content has proven immune to Google's core updates because it was purpose-built to game the ranking signals those updates measure. This reframes the quality decline as a structural problem — Google can't algorithmically distinguish content optimized for its algorithm from content optimized for humans — rather than a temporary lapse.
A short post on Bearblog titled "When did Google get so weird?" — by an author writing as *sancho* — climbed to roughly 1,100 points on Hacker News. It isn't a scoop. It isn't even new reporting. It's a screenshot-driven rant about what a typical Google results page looks like in late 2026: an AI Overview at the top, a "People also ask" accordion, a shopping carousel, a Reddit block, a YouTube block, a Discussions block, four ads, and — somewhere below the fold — the actual link the user was looking for.
The reason it hit the front page isn't that the observation is novel; it's that every working developer opened it, scrolled through the screenshots, and thought "yes, that's exactly what my last five searches looked like." The comment thread is the tell. Instead of the usual HN pile-on where half the room defends the incumbent, the top comments are variations of "I switched to Kagi eighteen months ago and I'm never going back," "Perplexity for research, Claude for code, Google for finding a restaurant," and the resigned "I just type `site:reddit.com` on everything now."
The post itself flags the specific shifts: AI Overviews eating the top of the page, sponsored results that no longer look meaningfully different from organic ones, a "Discussions and forums" module that mostly surfaces the Reddit answer Google is simultaneously training on, and a long tail of SEO-farm content that survived every core update because it was engineered to.
The interesting part isn't that Google's results are worse — that story has been running since the 2019 core updates. The interesting part is that the product has clearly stopped optimizing for the query and started optimizing for two other things: keeping you inside google.com long enough to serve another ad, and harvesting your reformulations as training signal for Gemini. Once you see the SERP as a data-collection surface rather than a retrieval surface, the layout choices stop being confusing.
Compare the incentives. A ten-blue-links page sends you off-site in under a second. An AI Overview answers in place, and if the answer is wrong or thin, you rephrase and query again — which is exactly the interaction pattern a foundation-model team wants to log. The Discussions module surfaces Reddit content Google licensed for $60M/year in 2024; the AI Overview then paraphrases that same content back at you without the click-through that used to fund Reddit and the individual answerers. It's a closed loop that only makes sense if the destination is the model, not the user.
The competitive picture has quietly moved too. Kagi crossed 55,000 paying subscribers earlier this year and publishes its ranking philosophy as a manifesto rather than a black box. Perplexity, whatever you think of its scraping ethics, has become the default "look something up with citations" tool inside a lot of engineering orgs. And the biggest shift: for a large slice of technical questions, developers now open Claude, ChatGPT, or a local model *first*, and only fall back to a search engine when they need a primary source. Google's moat was never the algorithm — it was the habit of typing into the omnibox — and habits are what LLM chat interfaces have been steadily eating for two years.
The SEO ecosystem tells the same story from the other side. Talk to anyone running content marketing and they'll admit organic-search traffic to their long-form is down 30–60% year over year, and that the drop correlates almost exactly with AI Overview coverage of their query space. Publishers are suing. Stack Overflow's traffic collapse is now a case study. The web that Google indexed is being hollowed out by the interface Google built on top of it, and there's no version of the next five years where that reverses.
If you build anything that depends on being discoverable through Google — a docs site, a SaaS landing page, a technical blog, a package registry — you should already have accepted that the old SEO playbook is a depreciating asset. The practical moves: get your content into places LLMs actually cite (GitHub READMEs, canonical docs, well-structured Wikipedia-adjacent references), invest in a real newsletter or RSS audience you own, and stop treating a Google #1 ranking as a durable moat. A first-page result that sits underneath an AI Overview paraphrasing your own content is worth roughly what the click-through rate says it's worth, which is single digits and falling.
If you're a developer doing daily research: it's worth running the experiment of paying for Kagi for a month, or defaulting to an LLM with web search for anything that isn't navigational. Most people who make the switch don't come back, and the reason isn't ideology — it's that the friction of parsing a modern Google SERP has quietly become higher than the friction of typing a question into a chat box. If your team still standardizes on Google for internal research, that's a policy worth revisiting.
And if you're building a product that has a "search" feature, the ceiling on what users will tolerate has moved. Users who have been trained by ChatGPT to ask a full-sentence question and get a full-sentence answer will not accept a keyword-matching search bar with a filter sidebar. Retrieval-augmented answers, with citations, are becoming the baseline — not the premium tier.
Google isn't going to un-ship AI Overviews or the Discussions module; the revenue math and the model-training math both point the same direction. What's actually in play is whether the alternatives — Kagi at the paid end, Perplexity and the chat interfaces at the AI end, and a resurgent "just read the docs" instinct in the middle — carve out enough share that a working developer in 2028 uses Google roughly the way they use Yahoo today: occasionally, out of habit, and slightly embarrassed about it. The HN reaction to a two-page blog post about SERP layout suggests that shift is further along than the market-share numbers say.
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.