×
Google’s AI Overviews are cutting off the oxygen to the web

Google’s AI Overviews are cutting off the oxygen to the web

It’s official: Google’s AI Overviews are eating search.

Ever since Google first debuted the AI-generated search summaries, web creators have feared the overviews will siphon precious clicks and upend a search experience publishers have relied upon for years. Now, it seems they have their proof.

According to a new study from the Pew Research Center, Google users who are met with an AI summary are not only less likely to click through to other websites but are also more likely to end their browsing session entirely.

Researchers found that just 8% of users who were presented with Google’s AI-generated overviews clicked on a traditional search result link, as opposed to those who did not encounter an AI summary, who clicked on a search result nearly twice as often.

Just over a quarter of searches that produced an AI summary were closed without users clicking through to any links, compared with 16% of pages with only traditional search results.

The summaries are also becoming more common. According to Pew, around one in five Google searches in March 2025 produced an AI summary, with 18% of all the Google searches in the study producing an AI summary.

AI’s search revolution

It’s easy to see why the summaries are popular. Apart from a few minor user experience tweaks, search has remained largely untouched since its conception. Up until AI-powered search entered the scene, users had been presented with a list of links, ranked by an ever-changing Google algorithm, in response to what is normally a natural-language query.

After the launch of AI-powered chatbots such as ChatGPT, the logical jump to technology’s search potential was so obvious that Google declared a “code red” internally and began pouring resources into its AI development.

Fast forward three years, and Google’s AI Overviews are facing off against AI-powered search competitors like Perplexity and ChatGPT Search.

More often than not, users who come to search engines are looking for an answer to a question. AI allows for a new, cleaner way to provide these answers, one that utilizes natural language and speeds up the search process for users.

But the trade-off for this improved experience is the lack of click-through to other websites, potentially resulting in a catastrophic decline in website traffic, especially for sites that rely on informational content or rank highly for keywords. 

The study found that Google is far more likely to serve up an AI Overview in response to longer, more natural-sounding queries or questions. Just 8% of one or two-word searches produced an AI-generated summary. That figure jumps to 53% for searches containing ten words or more.

Queries phrased as full sentences, especially those that include both a noun and a verb, triggered summaries 36% of the time. Meanwhile, question-based searches were the most likely to invoke an AI response, with 60% of queries beginning with “who,” “what,” “when,” or “why” generating an Overview.

Common sources

While the overviews do link out and cite web sources, more often than not, the summaries lean heavily on a trio of Wikipedia, YouTube, and Reddit.

Collectively, these three platforms accounted for 15% of all citations in AI Overviews, almost mirroring the three sites’ 17% share of links in standard search results. Researchers found that AI Overviews were more likely to include links to Wikipedia and government websites, while standard search results featured YouTube links more prominently. Government sources represented 6% of AI-linked content, compared to just 2% in traditional results.

In another potentially ominous sign for publishers hoping to capitalize on the AI revolution, news organizations remain largely flat in both formats, making up just 5% of links in AI Overviews and standard search results alike.

Things are likely to get worse for web creators before they get better as Google leans further into AI in its search business. In May, Google unveiled a new “AI mode” search feature that intends to provide more direct answers to user questions. The answers provided by the new feature are similar to AI Overviews, and blend AI-generated responses with summarized and linked content from around the internet.

Google has continually brushed off concerns that the overviews could negatively affect web traffic for creators. The company did not immediately respond to Fortune’s request for comment on the Pew survey.

Source link
#Googles #Overviews #cutting #oxygen #web


Yesterday, OpenAI released its latest voice model, GPT-Live-1. The company called it “a new generation of voice models for natural human-AI interaction.” The full-duplex model allows ChatGPT to talk and listen at the same time, giving the back-and-forth a more conversational flow. It’s designed to be smarter, faster, and more natural.

Sounds great! Let’s see how it handles the internet’s one-man voice model red team, TikToker @huskistaken (aka Husk), when he tests out its capabilities by giving the model something simple:

Ooh, so close. Husk gave ChatGPT the straightforward request to tell him how many times the letter E appears in the number “seventeen,” and the voice model managed to get halfway there by answering “two.” It then offered a very awkward sign-off when Husk ended the conversation. Maybe that’s a query that GPT-Live-2 will be able to handle.

To be fair, Husk isn’t really testing any of the new features that OpenAI highlighted with the release of its new model, which, by most accounts, seems like it’s more capable when it comes to things like live translation than previous iterations. But he is giving the model a very simple benchmarking test that, despite being OpenAI’s flagship voice model, it still failed spectacularly.

It seems OpenAI knows just how bad this looks, too. Jason Liu, a Developer Experience Engineer on OpenAI Codex, reposted the video with a very succinct, “FUCK.”

Husk’s made a habit of terrorizing OpenAI over its voice model for some time now, making viral videos out of how poorly it handles certain prompts. One of his videos—in which he tasks ChatGPT with setting a timer and watching it fail—made it all the way to CEO Sam Altman, who tried to laugh it off in a very “I’m not mad, please don’t put in the newspaper that I got mad” kinda way.

Husk is not alone in giving GPT-Live-1 the old stress test. Another user on X gave the model the “Strawberry” test, asking it to count the number of times the letter R appears in the word “strawberry.” It’s a classic, and one that most AI models can answer correctly by now (whether that’s because they really know there are 3 Rs or because they’ve been trained on how to respond to prevent embarrassment is another question), but GPT-Live-1 gets tripped up by it.

Another user noted a new annoyance that has arisen from the model’s full-duplex functionality: While the model now allows ChatGPT to say things like “mhm” and “yeah” while you talk to confirm it is listening, it apparently is incapable of simply shutting up and not doing that. It seems ChatGPT has built an interrupting machine. Just what we’ve all been waiting for.

READ MORE:

The Future Is Always Listening: OpenAI Says Its New Voice Assistant Is ‘One Step Closer to a Truly Accessible AGI’

OpenAI Is Tired of Seeing All Those Videos of People Clowning on Its Voice Mode

#OpenAI #Beat #TikTokerartifical intelligence,ChatGPT,OpenAI,voice model">OpenAI Just Can’t Beat This TikToker
                Yesterday, OpenAI released its latest voice model, GPT-Live-1. The company called it “a new generation of voice models for natural human-AI interaction.” The full-duplex model allows ChatGPT to talk and listen at the same time, giving the back-and-forth a more conversational flow. It’s designed to be smarter, faster, and more natural. Sounds great! Let’s see how it handles the internet’s one-man voice model red team, TikToker @huskistaken (aka Husk), when he tests out its capabilities by giving the model something simple:  Had to give the new voice model the classic test pic.twitter.com/fQYHbBRNuL — Husk (@huskirl) July 8, 2026  Ooh, so close. Husk gave ChatGPT the straightforward request to tell him how many times the letter E appears in the number “seventeen,” and the voice model managed to get halfway there by answering “two.” It then offered a very awkward sign-off when Husk ended the conversation. Maybe that’s a query that GPT-Live-2 will be able to handle.

 To be fair, Husk isn’t really testing any of the new features that OpenAI highlighted with the release of its new model, which, by most accounts, seems like it’s more capable when it comes to things like live translation than previous iterations. But he is giving the model a very simple benchmarking test that, despite being OpenAI’s flagship voice model, it still failed spectacularly.

 It seems OpenAI knows just how bad this looks, too. Jason Liu, a Developer Experience Engineer on OpenAI Codex, reposted the video with a very succinct, “FUCK.” Husk’s made a habit of terrorizing OpenAI over its voice model for some time now, making viral videos out of how poorly it handles certain prompts. One of his videos—in which he tasks ChatGPT with setting a timer and watching it fail—made it all the way to CEO Sam Altman, who tried to laugh it off in a very “I’m not mad, please don’t put in the newspaper that I got mad” kinda way.

 Husk is not alone in giving GPT-Live-1 the old stress test. Another user on X gave the model the “Strawberry” test, asking it to count the number of times the letter R appears in the word “strawberry.” It’s a classic, and one that most AI models can answer correctly by now (whether that’s because they really know there are 3 Rs or because they’ve been trained on how to respond to prevent embarrassment is another question), but GPT-Live-1 gets tripped up by it.  OpenAI’s brand new voice model vs counting the r’s in strawberry. I really tried to help it along. pic.twitter.com/Pm0RfLyxIA — Himelstech (@himelstech) July 9, 2026  Another user noted a new annoyance that has arisen from the model’s full-duplex functionality: While the model now allows ChatGPT to say things like “mhm” and “yeah” while you talk to confirm it is listening, it apparently is incapable of simply shutting up and not doing that. It seems ChatGPT has built an interrupting machine. Just what we’ve all been waiting for. READ MORE: The Future Is Always Listening: OpenAI Says Its New Voice Assistant Is ‘One Step Closer to a Truly Accessible AGI’ OpenAI Is Tired of Seeing All Those Videos of People Clowning on Its Voice Mode      #OpenAI #Beat #TikTokerartifical intelligence,ChatGPT,OpenAI,voice model

called it “a new generation of voice models for natural human-AI interaction.” The full-duplex model allows ChatGPT to talk and listen at the same time, giving the back-and-forth a more conversational flow. It’s designed to be smarter, faster, and more natural.

Sounds great! Let’s see how it handles the internet’s one-man voice model red team, TikToker @huskistaken (aka Husk), when he tests out its capabilities by giving the model something simple:

Ooh, so close. Husk gave ChatGPT the straightforward request to tell him how many times the letter E appears in the number “seventeen,” and the voice model managed to get halfway there by answering “two.” It then offered a very awkward sign-off when Husk ended the conversation. Maybe that’s a query that GPT-Live-2 will be able to handle.

To be fair, Husk isn’t really testing any of the new features that OpenAI highlighted with the release of its new model, which, by most accounts, seems like it’s more capable when it comes to things like live translation than previous iterations. But he is giving the model a very simple benchmarking test that, despite being OpenAI’s flagship voice model, it still failed spectacularly.

It seems OpenAI knows just how bad this looks, too. Jason Liu, a Developer Experience Engineer on OpenAI Codex, reposted the video with a very succinct, “FUCK.”

Husk’s made a habit of terrorizing OpenAI over its voice model for some time now, making viral videos out of how poorly it handles certain prompts. One of his videos—in which he tasks ChatGPT with setting a timer and watching it fail—made it all the way to CEO Sam Altman, who tried to laugh it off in a very “I’m not mad, please don’t put in the newspaper that I got mad” kinda way.

Husk is not alone in giving GPT-Live-1 the old stress test. Another user on X gave the model the “Strawberry” test, asking it to count the number of times the letter R appears in the word “strawberry.” It’s a classic, and one that most AI models can answer correctly by now (whether that’s because they really know there are 3 Rs or because they’ve been trained on how to respond to prevent embarrassment is another question), but GPT-Live-1 gets tripped up by it.

Another user noted a new annoyance that has arisen from the model’s full-duplex functionality: While the model now allows ChatGPT to say things like “mhm” and “yeah” while you talk to confirm it is listening, it apparently is incapable of simply shutting up and not doing that. It seems ChatGPT has built an interrupting machine. Just what we’ve all been waiting for.

READ MORE:

The Future Is Always Listening: OpenAI Says Its New Voice Assistant Is ‘One Step Closer to a Truly Accessible AGI’

OpenAI Is Tired of Seeing All Those Videos of People Clowning on Its Voice Mode

#OpenAI #Beat #TikTokerartifical intelligence,ChatGPT,OpenAI,voice model">OpenAI Just Can’t Beat This TikTokerOpenAI Just Can’t Beat This TikToker
                Yesterday, OpenAI released its latest voice model, GPT-Live-1. The company called it “a new generation of voice models for natural human-AI interaction.” The full-duplex model allows ChatGPT to talk and listen at the same time, giving the back-and-forth a more conversational flow. It’s designed to be smarter, faster, and more natural. Sounds great! Let’s see how it handles the internet’s one-man voice model red team, TikToker @huskistaken (aka Husk), when he tests out its capabilities by giving the model something simple:  Had to give the new voice model the classic test pic.twitter.com/fQYHbBRNuL — Husk (@huskirl) July 8, 2026  Ooh, so close. Husk gave ChatGPT the straightforward request to tell him how many times the letter E appears in the number “seventeen,” and the voice model managed to get halfway there by answering “two.” It then offered a very awkward sign-off when Husk ended the conversation. Maybe that’s a query that GPT-Live-2 will be able to handle.

 To be fair, Husk isn’t really testing any of the new features that OpenAI highlighted with the release of its new model, which, by most accounts, seems like it’s more capable when it comes to things like live translation than previous iterations. But he is giving the model a very simple benchmarking test that, despite being OpenAI’s flagship voice model, it still failed spectacularly.

 It seems OpenAI knows just how bad this looks, too. Jason Liu, a Developer Experience Engineer on OpenAI Codex, reposted the video with a very succinct, “FUCK.” Husk’s made a habit of terrorizing OpenAI over its voice model for some time now, making viral videos out of how poorly it handles certain prompts. One of his videos—in which he tasks ChatGPT with setting a timer and watching it fail—made it all the way to CEO Sam Altman, who tried to laugh it off in a very “I’m not mad, please don’t put in the newspaper that I got mad” kinda way.

 Husk is not alone in giving GPT-Live-1 the old stress test. Another user on X gave the model the “Strawberry” test, asking it to count the number of times the letter R appears in the word “strawberry.” It’s a classic, and one that most AI models can answer correctly by now (whether that’s because they really know there are 3 Rs or because they’ve been trained on how to respond to prevent embarrassment is another question), but GPT-Live-1 gets tripped up by it.  OpenAI’s brand new voice model vs counting the r’s in strawberry. I really tried to help it along. pic.twitter.com/Pm0RfLyxIA — Himelstech (@himelstech) July 9, 2026  Another user noted a new annoyance that has arisen from the model’s full-duplex functionality: While the model now allows ChatGPT to say things like “mhm” and “yeah” while you talk to confirm it is listening, it apparently is incapable of simply shutting up and not doing that. It seems ChatGPT has built an interrupting machine. Just what we’ve all been waiting for. READ MORE: The Future Is Always Listening: OpenAI Says Its New Voice Assistant Is ‘One Step Closer to a Truly Accessible AGI’ OpenAI Is Tired of Seeing All Those Videos of People Clowning on Its Voice Mode      #OpenAI #Beat #TikTokerartifical intelligence,ChatGPT,OpenAI,voice model

Yesterday, OpenAI released its latest voice model, GPT-Live-1. The company called it “a new generation of voice models for natural human-AI interaction.” The full-duplex model allows ChatGPT to talk and listen at the same time, giving the back-and-forth a more conversational flow. It’s designed to be smarter, faster, and more natural.

Sounds great! Let’s see how it handles the internet’s one-man voice model red team, TikToker @huskistaken (aka Husk), when he tests out its capabilities by giving the model something simple:

Ooh, so close. Husk gave ChatGPT the straightforward request to tell him how many times the letter E appears in the number “seventeen,” and the voice model managed to get halfway there by answering “two.” It then offered a very awkward sign-off when Husk ended the conversation. Maybe that’s a query that GPT-Live-2 will be able to handle.

To be fair, Husk isn’t really testing any of the new features that OpenAI highlighted with the release of its new model, which, by most accounts, seems like it’s more capable when it comes to things like live translation than previous iterations. But he is giving the model a very simple benchmarking test that, despite being OpenAI’s flagship voice model, it still failed spectacularly.

It seems OpenAI knows just how bad this looks, too. Jason Liu, a Developer Experience Engineer on OpenAI Codex, reposted the video with a very succinct, “FUCK.”

Husk’s made a habit of terrorizing OpenAI over its voice model for some time now, making viral videos out of how poorly it handles certain prompts. One of his videos—in which he tasks ChatGPT with setting a timer and watching it fail—made it all the way to CEO Sam Altman, who tried to laugh it off in a very “I’m not mad, please don’t put in the newspaper that I got mad” kinda way.

Husk is not alone in giving GPT-Live-1 the old stress test. Another user on X gave the model the “Strawberry” test, asking it to count the number of times the letter R appears in the word “strawberry.” It’s a classic, and one that most AI models can answer correctly by now (whether that’s because they really know there are 3 Rs or because they’ve been trained on how to respond to prevent embarrassment is another question), but GPT-Live-1 gets tripped up by it.

Another user noted a new annoyance that has arisen from the model’s full-duplex functionality: While the model now allows ChatGPT to say things like “mhm” and “yeah” while you talk to confirm it is listening, it apparently is incapable of simply shutting up and not doing that. It seems ChatGPT has built an interrupting machine. Just what we’ve all been waiting for.

READ MORE:

The Future Is Always Listening: OpenAI Says Its New Voice Assistant Is ‘One Step Closer to a Truly Accessible AGI’

OpenAI Is Tired of Seeing All Those Videos of People Clowning on Its Voice Mode

#OpenAI #Beat #TikTokerartifical intelligence,ChatGPT,OpenAI,voice model

Friendly home robots have been the dream of sci-fi for a long time. Docile, helpful machines that do your laundry, take out the trash, maybe make dinner, and clean up afterward too. But if robots are going to do all that, they’re going to need some hands to make all that happen.

1X, a Norwegian-American robotics company, today revealed details about the five-finger hands attached to its soft, helpful robot companion, Neo.

The hands are built with actuators designed to replicate how tendons in the arms move human hands. 1X says this gives Neo’s hands 25 degrees of freedom of movement, which is just a little less than the 27 degrees of freedom human hands usually have. Cameras and AI smarts help sort out the broader context of what the fingers are trying to grab. It’s a dexterous mix that gives the Neo bot a very broad range of motion. 1X says the hands can grip odd shapes and detect when something is slipping out of its grip. The fingers can also move extremely quickly and hyperextend in directions human digits can’t. They also have an IP68 waterproof rating, meaning the robot can wash its own hands.

Courtesy of 1X

It’s a range that the engineers have aimed to dial in to align with what a human can actually do. Jonathan Terfurth, 1X’s director of actuators and hands, says the range of motion might even be better than a human’s, enabling it to open doors, lift heavy objects, and even plug itself in when its battery starts to die.

“You want to be able to operate with a human who has never worked or interfaced with a robot, and you still want it to be safe and compliant and soft,” Terfurth says. “Range of motion can be a bit extreme, but we try to be very close to what humans can do so that we can live in the world.”

Together, this is part of a growing ChatGPT-esque moment for robotics, where bots have gone from clumsy claw machines to capable handlers, able to gently carry fragile objects and become increasingly helpful for managing menial daily tasks and boring office chores.

The humanoid robot market tends to be dominated by tough, hulking Terminators meant to be paraded out for some defense industry contract or another. 1X is taking a different tack for its robot. The Neo is a soft, supple robot wrapped in a 3D lattice shell. Its design is inspired by characters like Baymax, the affable robot from the Disney movie Big Hero 6. In limited quantities, early access pricing is $20,000 or $500 per month to have it in your home, though the lump sum will prioritize delivery for 2026.

#Neo #Robot #Freaky #Fast #Fingersrobots,robotics,artificial intelligence">The 1X Neo Robot Has Freaky Fast FingersFriendly home robots have been the dream of sci-fi for a long time. Docile, helpful machines that do your laundry, take out the trash, maybe make dinner, and clean up afterward too. But if robots are going to do all that, they’re going to need some hands to make all that happen.1X, a Norwegian-American robotics company, today revealed details about the five-finger hands attached to its soft, helpful robot companion, Neo.The hands are built with actuators designed to replicate how tendons in the arms move human hands. 1X says this gives Neo’s hands 25 degrees of freedom of movement, which is just a little less than the 27 degrees of freedom human hands usually have. Cameras and AI smarts help sort out the broader context of what the fingers are trying to grab. It’s a dexterous mix that gives the Neo bot a very broad range of motion. 1X says the hands can grip odd shapes and detect when something is slipping out of its grip. The fingers can also move extremely quickly and hyperextend in directions human digits can’t. They also have an IP68 waterproof rating, meaning the robot can wash its own hands.Courtesy of 1XIt’s a range that the engineers have aimed to dial in to align with what a human can actually do. Jonathan Terfurth, 1X’s director of actuators and hands, says the range of motion might even be better than a human’s, enabling it to open doors, lift heavy objects, and even plug itself in when its battery starts to die.“You want to be able to operate with a human who has never worked or interfaced with a robot, and you still want it to be safe and compliant and soft,” Terfurth says. “Range of motion can be a bit extreme, but we try to be very close to what humans can do so that we can live in the world.”Together, this is part of a growing ChatGPT-esque moment for robotics, where bots have gone from clumsy claw machines to capable handlers, able to gently carry fragile objects and become increasingly helpful for managing menial daily tasks and boring office chores.The humanoid robot market tends to be dominated by tough, hulking Terminators meant to be paraded out for some defense industry contract or another. 1X is taking a different tack for its robot. The Neo is a soft, supple robot wrapped in a 3D lattice shell. Its design is inspired by characters like Baymax, the affable robot from the Disney movie Big Hero 6. In limited quantities, early access pricing is ,000 or 0 per month to have it in your home, though the lump sum will prioritize delivery for 2026.#Neo #Robot #Freaky #Fast #Fingersrobots,robotics,artificial intelligence

ChatGPT-esque moment for robotics, where bots have gone from clumsy claw machines to capable handlers, able to gently carry fragile objects and become increasingly helpful for managing menial daily tasks and boring office chores.

The humanoid robot market tends to be dominated by tough, hulking Terminators meant to be paraded out for some defense industry contract or another. 1X is taking a different tack for its robot. The Neo is a soft, supple robot wrapped in a 3D lattice shell. Its design is inspired by characters like Baymax, the affable robot from the Disney movie Big Hero 6. In limited quantities, early access pricing is $20,000 or $500 per month to have it in your home, though the lump sum will prioritize delivery for 2026.

#Neo #Robot #Freaky #Fast #Fingersrobots,robotics,artificial intelligence">The 1X Neo Robot Has Freaky Fast Fingers

Friendly home robots have been the dream of sci-fi for a long time. Docile, helpful machines that do your laundry, take out the trash, maybe make dinner, and clean up afterward too. But if robots are going to do all that, they’re going to need some hands to make all that happen.

1X, a Norwegian-American robotics company, today revealed details about the five-finger hands attached to its soft, helpful robot companion, Neo.

The hands are built with actuators designed to replicate how tendons in the arms move human hands. 1X says this gives Neo’s hands 25 degrees of freedom of movement, which is just a little less than the 27 degrees of freedom human hands usually have. Cameras and AI smarts help sort out the broader context of what the fingers are trying to grab. It’s a dexterous mix that gives the Neo bot a very broad range of motion. 1X says the hands can grip odd shapes and detect when something is slipping out of its grip. The fingers can also move extremely quickly and hyperextend in directions human digits can’t. They also have an IP68 waterproof rating, meaning the robot can wash its own hands.

Courtesy of 1X

It’s a range that the engineers have aimed to dial in to align with what a human can actually do. Jonathan Terfurth, 1X’s director of actuators and hands, says the range of motion might even be better than a human’s, enabling it to open doors, lift heavy objects, and even plug itself in when its battery starts to die.

“You want to be able to operate with a human who has never worked or interfaced with a robot, and you still want it to be safe and compliant and soft,” Terfurth says. “Range of motion can be a bit extreme, but we try to be very close to what humans can do so that we can live in the world.”

Together, this is part of a growing ChatGPT-esque moment for robotics, where bots have gone from clumsy claw machines to capable handlers, able to gently carry fragile objects and become increasingly helpful for managing menial daily tasks and boring office chores.

The humanoid robot market tends to be dominated by tough, hulking Terminators meant to be paraded out for some defense industry contract or another. 1X is taking a different tack for its robot. The Neo is a soft, supple robot wrapped in a 3D lattice shell. Its design is inspired by characters like Baymax, the affable robot from the Disney movie Big Hero 6. In limited quantities, early access pricing is $20,000 or $500 per month to have it in your home, though the lump sum will prioritize delivery for 2026.

#Neo #Robot #Freaky #Fast #Fingersrobots,robotics,artificial intelligence

Post Comment