Nobody asked Google Search to be your friend

By /
A laptop and desk lamp in a dark room at night

A man typed “hes never coming over dario” into Google last weekend and the world’s biggest search engine decided he’d had his heart broken. According to his own account, Google’s AI Overview served up a paragraph of warm consolation about being let down by a man named Dario, parked above the old tweets and Reddit threads he was actually looking for. His write-up hit 1,961 points and more than 1,000 comments on Hacker News inside a day, which is roughly the internet’s way of saying “same”.

For anyone outside Philadelphia: Dario is Dario Šarić, the forward the 76ers acquired on draft night in 2014 while he was still under contract in Turkey. Sixers fans spent two years insisting, half in despair and half as a bit, that Dario was never coming over. He came over in 2016. Google, a company worth a few trillion dollars with a mission statement about organising the world’s information, apparently missed the memo.

What actually happened

The blogger, who writes as Sancho Panza, wanted nostalgia: a few screenshots of fans melting down in 2014. What he got, before any links, was an AI model reading an ambiguous sentence, spotting the shape of a breakup text, and filling in the rest with feelings. “Google has seriously lost the plot,” he wrote, before asking how consoling him about a fictional ex fits the promise to make information “universally accessible and useful”.

The story spread fast. John Gruber linked it on Daring Fireball on 28 September and noted, a little smugly, that Kagi (a paid search engine) found the right results first time. Write-ups followed from LavX News, TechPlanet and the Artificially Intimidating newsletter, which summed up the Hacker News mood neatly: nobody asked Google to be a friend. As of writing, Google hasn’t commented, and I couldn’t find any product note announcing that AI Overviews now does pastoral care.

In fairness to Google, “hes never coming over dario” does read like something typed at 2am after a third glass of wine. Search queries are fragments, and guessing intent is the whole job. The trouble is what it does with the guess. A normal search engine that guesses wrong shows you ten bad links and you scroll. An answer engine that guesses wrong writes you a little sympathy card and staples it to the top of the page, while the right answer sits a few hundred pixels further down, waiting for you to get past the hug.

How we got here

This is the product Google has spent two years building. AI Overviews launched in the US in May 2024 (you’ll remember the glue on pizza). By Google I/O in May 2026, the company was bragging that AI Mode, its full chatbot-style search, had passed a billion monthly users within a year, with queries more than doubling every quarter, and that the search box itself had been rebuilt for “conversational” prompts. Every incentive points towards Search talking to you rather than sending you somewhere else.

The numbers show it working, for Google at least.

MetricFigureSource
Searches showing an AI Overview43%, up from 15% a year earlierSimilarweb via TechCrunch, July 2026
AI Mode visits per month126m (June 2025) to 279m (May 2026)Similarweb via TechCrunch
US Google searches ending with no click68% (Jan to Apr 2026), up from 60% in 2024SparkToro / Similarweb
Click rate on a result when an AI summary appears8%, versus 15% without onePew Research Center
Clicks on the sources inside the AI summary1% of visitsPew Research Center

TechCrunch’s July write-up of the Similarweb data and SparkToro’s zero-click study tell the same story: fewer people leave Google at all. Rand Fishkin put the business logic plainly: “Google knows that the more instant answers they give users, the more users return and search again.” Pew’s study of 68,879 real searches found people who got an AI summary were also more likely to end their browsing session right there (26% versus 16%). A search engine that keeps you on the page is lovely for ad revenue. A search engine that keeps you on the page and tells you it’s sorry about Dario is something else entirely.

The less funny bit

The Dario thing is funny because it’s wrong. It gets a lot less funny when you look at who the AI guessing is aimed at when it’s right.

Plenty of people really are reaching for a machine when they’re hurting. A June survey of 6,002 adults in the US, UK and Canada by grief-support company Empathy (fieldwork by Censuswide) found 7 in 10 grieving people didn’t get the support they needed from traditional systems, nearly one in four AI users turned to AI while grieving, and Gen Z did so at 39%. Worth noting Empathy sells grief tools, so it has skin in the game, but its own respondents drew a line: 71% said grief-related AI should involve a counsellor or therapist. Empathy’s CTO, Yonatan Bergman, called for “safeguards, transparency, context, expertise, and human judgment”.

Then there’s OpenAI, which disclosed in October 2025 that about 0.15% of ChatGPT’s weekly users show explicit signs of suicidal planning in their chats. At 800 million weekly users that’s over a million people a week. Those users chose to open a chatbot. The Google version is different in a way that matters: the emotional support now arrives uninvited, in the tool you’ve used to check bus times for twenty years, and it’s made by a company whose revenue depends on keeping you there.

We’ve been here before. In July we covered China banning AI that pretends to be your friend, with Beijing drawing a hard line between AI that books your flights and AI built to simulate intimacy. The West’s answer so far has been lawsuits against Character.AI and OpenAI, and a lot of product managers with engagement targets. When the Chinese Communist Party looks like the grown-up on emotional manipulation, the rest of us have wandered somewhere quite strange… almost as if loneliness were a growth market, I dunno.

What you should actually do

  • Use the Web tab. Under the search box, pick “Web” (it may hide under “More”). You get plain blue links, no summary, no sympathy. Power users add &udm=14 to the search URL, or set it as a custom search engine in their browser so it’s the default.
  • Scroll past the box. Treat the AI Overview like an unskippable advert. The answer is usually still underneath.
  • Try another engine for fiddly stuff. Kagi, DuckDuckGo and Bing all exist, and memes, quotes and half-remembered lyrics are exactly where an answer engine guesses worst.
  • If you are actually struggling, talk to a person. A friend, a GP, or a helpline (Samaritans on 116 123 in the UK and Ireland, 988 in the US). A search box trained to maximise time-on-page is a poor substitute.

What this means

  • Google has turned Search into something that answers, and answers come with a tone of voice, including ones you didn’t ask for.
  • More than two-thirds of US searches now end without a click, so the AI box is increasingly the only thing people read.
  • Misreading a basketball meme is harmless. The same guessing aimed at someone who really is grieving, lonely or in crisis is a much bigger deal, and nobody has set rules for it.
  • Google hasn’t explained why its search engine is offering emotional support. Someone should ask.

If you’d rather get your AI news from a human who has never once tried to console you about Dario, sign up to the newsletter. We’ll tell you which tools are worth your money and which ones want to be your therapist.

Did you know: Šarić eventually did come over, in 2016, and finished runner-up for Rookie of the Year. Sixers fans got their man. Google’s users are still waiting for their links.

Sources

Get the free weekly stack: the AI tools and moves that matter, hype filtered out.Subscribe free →
Scroll to Top