Google AI Overviews

[Dani]

I know how I feel about this as a publisher. However, as a searcher, do you like the Google AI overviews at the top of the search results? Or would you prefer things went back to the old school way they were years ago where it was just 10 search result links on a page?

[Reverend Jim]

About 50-50. Some times I just want the information, not a link to a website full of ads and menus I have to wade through.

[Dani]

And when, for example, would be times when the other is true?

[Reverend Jim]

Some times I just need the info, like the current score during the Grey Cup game or the World Series. Some times I need a link for "download acoustica".

[SCBWV]

Recently, it hasn't been helpful when I had a programming question, which I find puzzling. I would think that would be one of the easiest to get correct.

[Salem]

If it's a simple fact that I already know the answer to and I want to just check it, then the summary is fine.

If it's complicated, where there may be differing opinions, then absolutely not. The biased and often wrong AI can take a hike.

[Dani]

... and this was a year ago. I think I read somewhere the number was more like 80% in 2025. And growing. Zero-click searches are a real problem for publishers who still give away all of their content for free to the LLMs but get no advertising dollars in return. Hey, at least they don't have to pay staff writers anymore because they now only have the budget to publish articles generated by LLM.

Soon enough, every news and editorial publication on the web will be entirely LLM-generated content. We will do searches and read answers generated by LLMs who ingested and trained off of LLM-generated content.

At least right now, the majority of what LLMs have trained off of was human-written. Have a programming question? LLMs learned to fix coding bugs from ingesting 20+ years of Stack Overflow and DaniWeb. I wonder what will happen when there are new technologies and programming languages? LLMs won't have any information about them to train on. Will the first round of articles written about them be 100% hallucinations? Will search results and LLM code just be based on hallucinated articles?

[rproffitt]

Zero click searches. Dwell on that.
On top of that, look at what is now being ingested which was previously off limits. Gmail for example.

With so much AI generated slop, and without any disclaimer such was AI generated I say let AI do the Ouroboros.

[Dani]

On top of that, look at what is now being ingested which was previously off limits. Gmail for example.

Gmail was never off limits. Back in 2008-ish, you could pay to show contextually targeted ads on Gmail with Google AdWords. That means that Google was ingesting the content of emails and serving ads in the right sidebar of the interface based on contextual relevancy.

[rproffitt]

Further down the Google rabbit hole we read https://www.documentcloud.org/documents/24527110-google-unopposed-settlement

So they lied about it. My take is that anything in any cloud is now food for the machines.

[Dani]

Ingestion by LLM is recent (for Google AI). Ingestion for contextually targeted advertising is decades old.

[Salem]

AI gets indigestion from ingesting too much of it's own dog food.

Circular arguments are good because they are.

[john_111]

If it's something that I've forgotten the fine details about, the AI summary usually contains what I want.
If it's help with a problem, the AI sometimes gives what I need, but the actual sites returned can be more useful.
If I'm unsure that the AI is giving me enough info, or looks as if it is a bit biased, I look at the search results. The deeper the info I want, the more likely I am to dig into the search results.

A year ago I asked a question or five about usability engineering of websites, and it give some very good results and listed its sources, which were all quite good. I actually knew most of what it summarised, so I thought this would be a good test. I then asked a question where I knew the top two sites it should return (one was a site I run) and both were quoted as places to go for info in the AI summary. In the search returns, many sites were listed but none were as good as the two mentioned in the AI summary.

So basically, it all depends on the question you ask, and the amount of info you require.

[Dani]

This morning I was on my phone and I had Googled a medical question. One of the bullet points in the AI Overview sounded interesting and I specifically wanted to learn more about why it said what it said. There was a little link icon to click on to go to the source of what was summarized in the bullet point. Clicking on the link icon didn't work. It seemed to be a javascript bug of some sorts? Odd for Google. Either way, Google was giving me health information without letting me read the source to confirm its interpretation was even correct.

Update: Despite being <button>s in the DOM, the link icons here on my desktop don't seem to do anything either right now.

[rproffitt]

Now ask Gork.

[Dani]

Who is Gork?

[SCBWV]

The manipulation suggests a different nickname is warranted.

Grok Manipulation

[SCBWV]

rproffitt commented: Are you suggesting Elon has not tweaked GORK to praise him as superior to every other living person? +0

In case you weren't being facetious, I'm suggesting Elon needs a new nickname for his manipulation of Grok, but he's being doing that for a while. I'll also state my sadness that a large part of society will put too much faith in AI.

[SCBWV]

I didn't take it as being flippant. I just wanted to make sure you knew what side I was on. lol I didn't see the news but it doesn't surprise me. I'll check it out. Kids can't even be kids anymore.

[Trisha_6]

The AI overviews are convenient when I just want a quick summary, but there’s something about the old-school results that felt more open and less filtered. Back then, I could explore multiple sources, compare different viewpoints, and feel like I was actually doing my own research. Now it sometimes feels like Google is deciding the answer for me before I even get to look. Helpful, yes. But I do miss the simplicity and freedom of those classic ten blue links.

[john_111]

This morning I was on my phone and I had Googled a medical question. One of the bullet points in the AI Overview sounded interesting and I specifically wanted to learn more about why it said what it said. There was a little link icon to click on to go to the source of what was summarized in the bullet point. Clicking on the link icon didn't work. It seemed to be a javascript bug of some sorts? Odd for Google. Either way, Google was giving me health information without letting me read the source to confirm its interpretation was even correct.

Update: Despite being <button>s in the DOM, the link icons here on my desktop don't seem to do anything either right now.

The little link symbol in an AI summary opens links to the actual sites it mentions to the side of the summary, in a little box. Using chrome, on laptop.

[Dani]

I have since noticed that it does that sometimes. For the vast majority of time, for me at least, they do nothing.