'When Did Google Get So Weird?' Just Topped Hacker News. Should Your YC W27 Application Still Say 'SEO'?

The top story on Hacker News this weekend, with over 900 points, was a short essay titled "When did Google get so weird?" It struck a nerve because it put words to something founders have been feeling all year: Google search is no longer a reliable place to be found. AI Overviews now sit on top of results, sometimes misreading the query entirely. Commenters traded examples of searches for song lyrics being answered with congratulations, and of sites launched in 2025 that still have a single page indexed. And thread after thread, people described where they actually search now: ChatGPT, Claude, Perplexity, Kagi.

If you are filling out a YC Winter 2027 application right now, this is not just an amusing thread. It is a live update to one of the questions every application has to answer: how will you get users?

Why the distribution answer matters more than founders think

YC partners read thousands of applications per batch, and the "how will you get users" question is where weak applications quietly die. A great product description with "we'll do SEO and content marketing" underneath reads like a placeholder, because it was the default answer for a decade. It worked, too. Zapier (YC S12) famously grew on programmatic SEO, generating a landing page for every pair of apps it connected. Thousands of B2B SaaS companies copied that playbook.

The HN thread is a 200-comment argument that this playbook's foundation is cracking. Not because content stopped mattering, but because the front door changed. When Google answers the query itself, the click that used to reach your landing page never happens. When the person skips Google entirely and asks ChatGPT, your programmatic landing pages are not even in the room.

What is actually breaking, concretely

Three specific failure modes came up repeatedly in the discussion, and each one maps to a distribution assumption founders still write into applications:

Indexing is slower and stingier. Commenters described new sites waiting months with a handful of pages indexed. If your launch plan assumes Google will surface your content in weeks, budget for quarters.

AI Overviews absorb the click. For informational queries, the answer increasingly appears above the links. Traffic to the pages that used to win those queries drops even when rankings hold.

Users are leaving the building. The most-upvoted comments were not complaints, they were migration notes: Kagi for search, ChatGPT and Perplexity for questions. Your future users are increasingly asking an AI assistant, not a search box.

What should replace 'SEO' in your W27 application?

You do not need a novel growth hack. You need an answer that shows you understand where your specific users actually are in 2026. Three patterns hold up:

1. Be the answer, not the result

If your users ask questions, the goal is to be cited by the AI engines answering them. That means answer-shaped content: specific questions as headings, direct answers underneath, real data and named examples that a model can quote. It also means being present in the sources those engines retrieve from. This is the same discipline as old SEO, aimed at a new reader. In your application, name the questions your users ask and explain why you will be the cited answer.

2. Go where the query never happens

Airbnb (YC W09) did not wait to be searched for; they went to where their users already were, one Craigslist listing at a time. The 2026 version is the same move: the vertical Slack and Discord communities, the subreddit where your users complain about the problem, a Hacker News launch that earns its upvotes. YC has always trusted "do things that don't scale" more than a channel strategy, and in a post-Google-weirdness world it is also simply more defensible.

3. Make your product legible to agents

A growing share of "users" arriving at products are AI agents acting for a person: researching tools, comparing options, completing signups. If an assistant can read your docs, try your API, and complete onboarding without a human, you get recommended when someone asks "what should I use for X?" This is early, which is exactly why it is a credible wedge to claim in an application, as long as you can show you have actually tested it.

How to phrase it in the application

Weak: "We will acquire users through SEO, content marketing, and social media."

Stronger: "Our users ask [specific question] in ChatGPT and in [specific community]. We publish the canonical answer to that question, we are active in the two communities where it gets asked daily, and our docs are structured so AI assistants complete our onboarding successfully. Our first 50 users came from [specific place], manually."

The difference is not sophistication. It is evidence that you know where your users are and have already gone there.

Does this mean SEO is dead for startups?

No, and writing "SEO is dead" in an application would be its own red flag. Transactional and local queries still route through Google, and Google still carries the majority of search volume. What is dying is the assumption that a content moat on Google is a sufficient distribution answer for a new company. Treat search as one channel with a falling ceiling, not the plan.

Before you submit, it is worth having someone who has actually been through YC pressure-test your distribution answer. That is what YC Roaster is for: YC alumni review your application and tell you which answers read like placeholders before a partner does.

The Google-got-weird thread will scroll off the front page by the weekend. The shift it describes will still be underway when W27 interviews happen. Applications that acknowledge it will read like they were written in 2026. Applications that say "SEO" alone will read like they were written in 2019.