Your Website Was a Brochure. Now It's the Record.

The most-discussed technology story of the past week was not a product launch. It was an essay in The Walrus by Vass Bednar, published under the headline "Google Search Is Dying. What Comes Next Is Worse," which drew 937 points and 985 comments on Hacker News after being posted on August 10.

The argument is about archives and search infrastructure. But read it as a hospital system, a county government, a university, or a professional-services firm, and it is about something much closer to home: the two assumptions your website was built on have both stopped being true, and they stopped at the same time.

The two assumptions

For roughly twenty years, an institution's website rested on a pair of quiet premises.

The first: your site is a marketing brochure. It exists to make an impression, generate inquiries, and look current. Its job is persuasion.

The second: search is a reliable gateway. Somebody types your name or their question, and a search engine routes them to your page, where your words appear in your order with your context around them.

Both premises are now failing, and they are failing in ways that compound.

Search stopped routing and started answering

The clearest evidence isn't a market-share chart. It's a courtroom.

On June 10, 2026, the Regional Court of Munich held that Google can be directly liable for false statements produced by its AI Overviews feature. Two Munich publishers had been wrongly connected by the AI summary to scams, subscription traps, and questionable business practices — accusations that, the court noted, did not appear in the sources the summary linked to. The court found that AI Overviews are not neutral pointers to third-party content the way a list of blue links is. They are Google's own statements. It issued an injunction; Google has said it will appeal.

Set aside the liability question and look at what the court described. An answer layer now sits between the public and your website, generating new claims about you that appear nowhere in the material it cites.

The traffic consequence is visible at the largest scale available. The Wikimedia Foundation reported in October 2025 that human pageviews on Wikipedia had fallen roughly 8 percent year over year, attributing the decline to generative AI and search engines "providing answers directly to searchers." If the single most-cited reference site on the internet is losing human visitors to an answer layer built partly on its own content, no institutional website is exempt from the same mechanism.

And the answer layer can be gamed. 404 Media has documented companies planting content on Reddit specifically to shape what AI search tools say — a practice with its own acronym now. Cornell Tech researchers found that planted text as short as thirteen words could steer AI research agents toward recommending fabricated products, in controlled tests, with success rates as high as 100 percent.

So: a layer that rewrites you, occasionally invents things about you, and can be influenced by parties who are not you.

The backstop is going too

The reasonable response is: fine, the copy of record still exists somewhere. It doesn't, reliably.

In May 2026, Nate Silver reported that Disney had erased the FiveThirtyEight archive — a decade of work, which he estimated at around 200,000 person-hours, published under Disney's ownership from 2014 until the site was shut down in 2025. The URLs now redirect. The organization that owned the content decided it no longer justified its keep, and the record went with it.

The Internet Archive, the default assumption behind "well, it's archived somewhere," is under pressure from both directions. In September 2024, the Second Circuit ruled against it in Hachette v. Internet Archive, rejecting its controlled digital lending program as fair use; the Archive declined to appeal to the Supreme Court that December. And as of April 2026, Wired found 241 sites blocking the Internet Archive's crawler — 23 of them news organizations, including the New York Times and USA Today — largely out of concern that AI companies would use archived pages as an indirect route to copyrighted material.

The archival backstop is contracting at precisely the moment the front door became unreliable.

What this actually changes for an institution

Put the two halves together and the conclusion is not that AI is coming for your website. It is narrower, and more useful:

Your website is no longer your marketing brochure. It is your durable public record — and increasingly the only copy of it that you control.

For a hospital publishing patient guidance, a county publishing ordinances and notices, a university publishing program requirements, or a firm publishing its regulatory positions, that reframes the entire investment. The site stops being a persuasion asset with a three-year refresh cycle and becomes infrastructure with a stewardship obligation.

Four things follow, and none of them are exotic.

Structure is no longer a technical nicety. When a machine is the first reader, unstructured content gets summarized loosely and structured content gets quoted correctly. Clear headings, real semantic markup, explicit dates, named authorship, and schema on the pages that carry facts about you are what make an answer layer repeat you accurately instead of paraphrasing you badly. This is the same discipline that makes a site accessible to a screen reader — which is not a coincidence, and which most institutions are already obligated to do.

Permanence has to be a policy. URLs that survive a redesign. Content that is superseded rather than silently deleted. A record of what you said and when you said it. FiveThirtyEight is what the alternative looks like when the decision is made by someone optimizing a different balance sheet.

Canonical beats rented. Every fact about your organization that lives only on a platform you don't own — a directory listing, a social profile, a third-party portal — is a fact whose availability is somebody else's decision. The version on your own domain is the one you can still stand behind in five years.

Verification is part of the job now. Periodically ask the major AI assistants what they say about your organization, and read the answer as a public statement made in your name. The Munich publishers found out the hard way that this is not a hypothetical.

The part worth sitting with

None of this is a reason to panic, and it is not a reason to buy anything urgently.

It is a reason to reclassify something. For two decades, institutional websites were funded as marketing and evaluated on impressions. The work now in front of hospitals, agencies, school systems, and firms is to fund them as public infrastructure and evaluate them on whether the record they hold is accurate, structured, accessible, and still there in ten years.

The organizations that get this right won't be the ones with the largest budgets. They'll be the ones that noticed the brochure quietly became the archive.

Sources: Vass Bednar, "Google Search Is Dying. What Comes Next Is Worse", The Walrus; Hacker News discussion (engagement as of August 17, 2026); Regional Court of Munich ruling on AI Overview liability, June 10, 2026; Marshall Miller, "New User Trends on Wikipedia", Wikimedia Foundation, October 17, 2025; 404 Media on Reddit and AI search manipulation; Nate Silver, "Disney erased FiveThirtyEight", May 19, 2026; Hachette v. Internet Archive, Second Circuit, September 4, 2024; news outlets blocking the Wayback Machine (Wired figures), April 2026.