G starting to replace news headlines with AI-generated ones in Search

[rcjordan]

from Bsky. I haven't seen confirmation, but Debbie says it's likely true.

Google is literally starting to replace news headlines with AI-generated ones in Search. If you won't SEO-ify your headlines to their satisfaction, they'll do it for you.

[littleman]

I've been pondering the strength and meaning of SEO on 2026.  I remember Brett claiming SEO was dead in the early 2000's.  I don't think it is "dead" on 2026, but it is certainly diminished and much less dominant in an internet marketing strategy today.  Odds are much of those headlines are already written by AI before Google throws their SERP hack into the mix.

[ergophobe]

>> Brett claiming SEO was dead in the early 2000's

Isn't someone always claiming SEO is dead? It became a lot less fun and lot harder to find tricks that move the needle so it became like real work. Not even interesting real work in many cases.

>diminished

I have been wondering what happens to SEO now - I still have some SEO news in my feeds and most of it seems like echoes from a disappeared world. I've been asking myself if that's just my interests that have changed or if there's some objective meat to that observation.

Are we moving to a world where Google is for old people? If you're targeting 21 year olds, is SEO even a relevant strategy at all?

Unless it's a shopping or current events query, I usually don't look below the AI summary anymore.

Recent example: I tried 5-6 searches to find the original research article where researchers were able to identify users' political leanings based on mouse movement. Because of Google's recency bias, I came up with nothing but slop from the past year, none of which was on a topic even close to what I was asking about. Most of it was affiliate slop for people pitching mouse-tracking software.

I then typed my question into Claude and in 30 seconds, it had the exact reference to the original 2016 research paper, a summary of the paper, a summary of criticisms of the paper and links to 3-4 follow-up studies that add qualifications and caveats to the original paper.

Google Search (the 10 blue links part) is diminishing in importance every time a new model is released. If not for Gemini-generated headers, traffic would be collapsing, because Google SERPS are still a cesspool.

I think Google Search *IS* a form of AI and has been for roughly 15 years. I also think Google search is vastly better than it was in 1998.

BUT

At no point in *this* century has Google felt magical like it did when you did those first searches and the results were vastly better than Alta Vista and Ask Jeeves. Google has improved steadily, yes, but at a rate where it feels like the improvements are consistently outrun by the accumulation of slop and the enshittification of Google itself.

So though Google has improved for 28 years, it always *feels* like it's getting worse. The Farmer/Penguin/Panda era was the one period where search briefly felt better for end users. But what alternative was there until very recently? I switched one browser to default to Bing and it's at best slightly worse. LLMs did not become a viable alternative until the second to last round of model updates (e.g. Gemini 3, Claude Opus 4.5)   

Now I wonder whether *this* is the magical period for LLMs. Is this the 1998-2000 Google where it feels magical because people haven't figured out how to game it and spam it to death?

The LLM you use today is the dumbest, least powerful, least capable LLM you will use for the rest of your life. But I wonder if it will ever feel this magical again.

[rcjordan]

My question is whether we will end up with 90%⁺ Gemini dominance?

[ergophobe]

I hate to say it, but I already did above: advances in Gemini are keeping me using Google search and the fact that I already pay for Workspace accounts means that I get it for "free" just like the "free" movies I get with Amazon Prime.

Option #2 is that Anthropic's strategy of focusing on corporate customers with big budgets might keep them afloat and ad-free.

Option #3 is that the open weight models are good enough that most people use them with ads for consumer use, but for serious work, people use paid frontier models.

Option #4 is that open weight models are good enough that most people use them with ads for consumer use, but for serious work, frontier models use frontier models.

I'm going with option 4

[rcjordan]

>dominance

Gemini task automation is slow, clunky, and super impressive | The Verge
https://www.theverge.com/tech/898282/gemini-task-automation-uber-doordash-hands-on