AI-SEO would trash the internet
Plus: The Guardian finds a clever way to protect whistleblowers.
I think Google is about to kill SEO and replace it with something worse
AI Overviews are decimating search traffic to news outlets — but the solution could be really, really annoying.
The thing that news outlets said would happen because of AI search has begun.
A Wall Street Journal report found that organic search traffic — that is, traffic from a search engine that isn’t from a paid placement — has tanked at major news outlets.
Outlets like HuffPost and the Washington Post have seen their search traffic cut in half over the last three years.
The effects seem to be the worst at outlets that relied more on search traffic, but some degree of decline is being felt everywhere: search’s share of traffic fell by roughly 8% at the New York Times and 5% at The Wall Street Journal.
Business Insider’s search traffic declined by 55%, a factor the outlet cited when it laid off 21% of its staff two weeks ago and promised to “go all-in on AI.”
Background: When Google first introduced AI Overviews, publishers instantly sounded the alarm. The fact that Google’s AI summarizes info from web pages and news stories right in results means there is little incentive for users to actually click through for the whole story, or scroll down the page to find other sources.
Big picture: What these early impacts show is that Google may effectively be killing SEO for huge sectors of the internet. SEO, short for search engine optimization, is a skill set that has evolved over the last two decades to help web pages show up higher in search results.
For news outlets, SEO has always been a bit of a necessary evil. Google’s search algorithms are ultimately a black box — SEO experts have very educated guesses about how it works, but no one knows for sure and Google can (and has) change things at a moment’s notice. But the traffic it provides has justified playing the game.
It has also contributed to the marked decline users have noticed in both search engine results and the quality of websites at large. Most news outlets apply a light touch with SEO, mostly focusing on things like headlines, ledes, and having links in a story to show up in search, without changing the content of a story itself. But other types of websites (as well as less-serious news outlets) have built their entire content strategy around SEO. Think of how-to guides that repeat paragraphs worth of keywords before actually getting to the steps, or food blogs that tell a whole story before a recipe. They are all competing to be at the top based on the same playbook, and it has resulted in a lot of similar, low-quality sites getting bumped up.
Where we’re headed: The obvious implication here is that news outlets are going to suffer more and more as AI keeps people from actually visiting their websites. Even if they start pivoting to channels with more sustainable engagement (like newsletters!), this is still going to be a big hit to an industry that saw search as a more viable alternative after social media networks pulled the rug out from under them, initially by feeding them misleading metrics and then by deprioritizing news content on their platforms (and, in some cases, banning it outright). Non-news websites are likely to face similar challenges.
But what I’m also wondering about is if websites are going to start trying to find ways to game AI search. AI is complicated and there’s a lot of unknown factors that go into becoming a situation in AI Overviews, but that didn’t stop publishers from doing the same with regular search and creating SEO in the first place. And if you think SEO-bait sites were bad, the ways websites are going to essentially say “hey please notice me” to AI could be brain-melting.
There’s no way a placement as a link in an AI summary is ever going to replace search traffic that’s been lost, but with few other alternatives, websites may feel compelled to fight for the crumbs that are left.
And let’s keep in mind: Relying on AI Overviews and similar AI search functions is not just bad for websites — it’s bad for anyone who enjoys little things like “getting correct answers to the things they are looking for.” AI Overview is still absolutely plagued by hallucinations and inaccuracies — some recent favourites include not knowing the current year, telling people to prepare for a deadline on March 61st, or believing that Cape Breton has its own time zone 12 minutes ahead of the rest of Nova Scotia because of a satirical Beaverton article.
IN OTHER NEWS
Disney and Universal sue AI developer Midjourney. The studios’ suit included examples of copyright infringing material created by Midjourney’s platforms that featured characters from franchises like Toy Story, Shrek, Marvel comics, and minions from Despicable Me. Until now, major Hollywood studios — which have traditionally been highly protective of their IP — have largely stayed out of the various copyright lawsuits brought against AI companies. (CNBC)
Streamers go to court to fight Online Streaming Act. The case, which began on Monday, consolidates appeals made by Apple, Amazon, Spotify, and the Motion Picture Association-Canada, which represents the likes of Netflix and Paramount. The Act passed last year and forces foreign-owned streamers to pay into funds that support production of Canada content, similar to what Canadian media companies are obligated to do. (Global News)
Nintendo Switch 2 sets a new sales record. The company sold 3.5 million units within four days of launch, a mark it took more than a month to hit for the first Switch and Nintendo Wii. This happened despite a steeper price for both the console and its games compared to its predecessor. While Nintendo was able to avoid further tariff-related price hikes due to the U.S.’s 90-day pause, some consumers may have been getting their console early out of fear that re-implemented tariffs could drive the price higher in the future. (Engadget)
Competition Bureau sues DoorDash for deceptive “drip pricing.” The lawsuit alleges that the delivery service marketed prices that were unobtainable by users, due to extra fees and charges. The watchdog has been more aggressive about curtailing drip pricing since changes to the Competition Act gave it more power to do so, pursuing suits against the likes of Cineplex, SiriusXM, Discount Car & Truck Rentals, and Canada's Wonderland. (Canadian Press)
Guardian invents a new way to give anonymous sources a cover
CoverDrop is open source, so other news outlets are free to adopt it too.
Whistleblowers who pull back the curtain on the unethical things their employers do have more ways than ever to get in touch with the journalists that could expose wrongdoings to the public. But that also means there’s more ways for employers, investigators, or even hackers to expose their identities to the public, especially if they didn’t use a secured platform to reach out.
There may be a way to fix that. Working with the University of Cambridge’s department of computer science and technology, the Guardian has developed CoverDrop, a technology built into its existing news app that allows people — namely, whistleblowers, leakers, and other would-be sources — to securely message the outlet’s journalists without fear of reprisal.
How it works: In addition to encryption similar to secure messaging apps, CoverDrop gets every Guardian app user to periodically send a “decoy” message, unseen in the background. If an “adversary” — which could be an employer, investigator, hacker, law enforcement, or someone else — finds a way to access message logs, they will not be able to distinguish whistleblower messages from the decoys being sent by everyone else.
All messages — real or decoy — are encoded to use a similar, very small amount of data, so as to not tip someone off that a message is longer and, therefore, a real communication.
A user’s encryption keys are also one-way. Put simply, this means they can lock but not unlock messages, so they remain secure even if someone else were to come into possession of a user’s phone.
Why it’s needed: The “first contact problem.” Apps like Signal fully encrypt messages, as does SecureDrop, a platform already used by both big and small news outlets for sources to exchange files. But in their study, the Cambridge users found that many sources make their outreach through a more traditional, less secure channel, leaving them vulnerable until a reporter instructs them on how to cover their tracks.
Journalists, especially those experienced working with sensitive sources, are trained to protect those sources, and will often leave instructions for reaching out safely in their stories and social media profiles. However, would-be sources may not see those instructions, or know how to do them properly.
The fact that someone downloaded secure messaging apps in the first place, or began using them more than usual, could also point to someone being a whistleblower.
One small problem (and a solution): If someone didn’t already have the Guardian app, the act of downloading it could now be seen as incriminating. But the code for CoverDrop has been open-sourced, meaning any news organization can incorporate it into their own apps. If CoverDrop gets more widely used, that provides even more cover for sensitive sources.
ALSO WORTH YOUR TIME
A profile of Ross Minor, a blind gamer who consults for big studios to make games more accessible (and it’s not always an inspirational story).
Wikipedia has paused an AI experiment after “overwhelmingly negative feedback” from editors.
Snap is planning to sell AR glasses (again).
Go beyond the Beach Boys’ greatest hits records and check out this list of the late Brian Wilson’s greatest songs.
Security researchers found the first known “zero click” flaw on an AI agent — allowing hackers to take control of an agent without the user clicking anything.