Bestie, You Sound Like ChatGPT: People Are Using AI to Answer Texts.

Sarah Chiappetta needed some career advice a few months ago, so she turned to a close friend. When she laid out her dilemma over text, her friend’s response was suspiciously helpful. Chiappetta says the message was “overly sympathetic,” acknowledged her feelings thoroughly, and used something she’d rarely, if ever, seen her friend use before — an em dash. All of it made her pause. The words were comforting, but they sounded a lot like ChatGPT.

“I wasn’t mad. I was a little weirded out,” Chiappetta, a 30-year-old product marketing manager in San Francisco, tells me. She wondered: “Is my text that hard that you need ChatGPT to help you with it?”

Millions of people are now yapping with chatbots as if they’re coworkers, BFFs, boyfriends, girlfriends, and even, in rare cases, spouses. Some see this as dystopian; others see it as a band-aid for the so-called loneliness epidemic; and to others, it’s a godsend. But as developing relationships with chatbots becomes more normalized, what remains more stigmatized, perhaps, is when people use generative AI to foster their human relationships, in texts to their IRL friends, partners, and families. Research shows that people think of people more negatively when they receive a text that appears AI-generated, and social media is abound with ChatGPT police. There are telltale signs, like em dashes or the word “delve,” which ChatGPT favors and has injected into the speech of its super users, the sleuths say.

Bringing GenAI into our personal conversation could transform how we perceive one another and how we socialize. We’ve long grown comfortable with autocorrect and predictive text helping us spit out emails and texts. But large language models can entirely outsource the labor of comforting and confronting our friends. New research from MIT suggests that people using ChatGPT to assist with writing essays may get lazier and increasingly dependent on the technology with time, and “consistently underperformed at neural, linguistic, and behavioral levels” compared to control groups that either used Google or no aid to write their essays. As we start using GenAI to flirt on dating apps, write wedding vows, or confront friends and family in tough conversations, will it similarly deteriorate our social muscles and ability to connect to one another?

Development and adoption of AI is far outpacing the study of its social impacts, says Jess Hohenstein, a former AI researcher at Cornell University. But there is concern that the tech might undermine trust in how we text each other. “We don’t know who we’re actually talking to, and I want to talk to my friends and have validation that it’s actually them that’s talking to me and giving me feedback and listening to me,” Hohenstein says. “Could we potentially be moving to a place where face-to-face interactions are the only interactions we can truly trust?”

The way AI will affect our convos depends on how the tech is chiming in. For each offended recipient of a text that reads like it was copied and pasted straight from ChatGPT instead of from your bestie’s heart, there are people who feel AI helps put your best, most polished self forward. The issue, experts tell me, depends on how authentic to you the sentiment remains. If you’re fully presenting “AI in conversation as yourself, then you’re not being authentic,” says Marisa T. Cohen, who is dating app Hily’s “authentic relationship expert.” But if you’re asking AI to coach you to be wittier or kinder, “that’s just maybe giving you a little bit of assistance” akin to asking friends or self-help articles for a hand.

A 2023 study out of The Ohio State University underscores the potential complications of texting friends with AI. Participants were told they had a years-long friendship with a fictional Taylor, and told to text her seeking support, advice, or to tell her their birthday was coming up. They then received replies and were told that they either came directly from Taylor, that Taylor had human help writing the message, or that Taylor used gen AI to craft the response. Both AI and human intervention led people to feel that their relationship with Taylor was not very close, and the AI messages were also rated as less appropriate responses. Similarly, a 2023 study from Cornell University found that conversations become more efficient and people use more positive language when conversing with the help of AI smart replies, but people also had more negative evaluations of others more who used algorithmic responses (or those they think might be using AI).

“We really just see this critical disconnect,” says Hohenstein, who worked on the research from Cornell. “Using AI to communicate actually can kind of improve the way that we talk to each other, but there’s this perception where it’s judged so negatively, and we think this could be due to these societal assumptions about AI kind of lacking authenticity and being less human.”

Reddit is flush with posts from people hurt to learn their loved ones are talking to them with AI. One person complained that receiving a clearly AI-generated text made it seem “like our friendship is a chore.” Another says: “It’s YOUR letter to me, so if I find out an AI (or friend) wrote it for you, that certainly won’t make me think highly of you.” According to a new survey from Hily, 45% of Gen Zers say using AI in dating makes the process less authentic.

Part of the awkwardness of using AI as a mediator in a conversation may stem from the very nature of how large language models work, which is based on prediction and probability. “Telling you what is the most likely next best word is just so different from telling you what I think,” says Quinn White, a professor of philosophy at Harvard University. When AI becomes a mediator in the conversation, “it’s fundamentally different from what we’re doing when we talk with each other,” White says. People come to friends when they want to be heard, and ultimately to receive an opinion from that friend, not from a bot trained on all of the available information on the internet. And a good friend will listen, validate you, and tell you what you need to hear, not just what you statistically may want to hear.

Sarah has been relying on ChatGPT to translate acrimonious texts to her soon-to-be ex-husband. “It does a good job of separating yourself from a hostile situation,” she says.

Chiappetta says she called her friend out, nicely, asking if the text was generated by AI. Her friend copped to using ChatGPT for parts of her response, but not all, and Chiappetta dropped the inquisition from there. She says she has since noticed some other longer texts from the friend that also have hints of AI in them. But that doesn’t make Chiappetta less likely to turn to the friend for advice or value her less. “She’s still a really good friend,” Chiappetta says, and is comforted by the thought that she is still getting through to her friend, even if there’s an AI mediator creeping in the chat. “Generally, I like ChatGPT’s advice. It’s still helpful to get,” she says.

For some who are neurodivergent, contending with social anxiety, or facing a tough conversation, GenAI tools can make for helpful dress rehearsal for conversations with friends. “Incorporating generative AI is a sign of caring a lot about the other person,” David Deal, a marketing consultant, tells me. The 62-year-old has used ChatGPT recently to work through conversations with much younger people; one a mentee, and one a relative. Deal has used ChatGPT to workshop a response to his mentee, telling the bot he didn’t want to come off as a “mansplaining jerk,” and asking for help crafting an affirming reply to a young woman — the large language models, he says, have evolved to understand context like that, which was top of mind in his response. The chatbot, Deal says, suggested he lead the response with more outright empathy and restate how she was feeling to show he had been actively listening to her concerns. “I don’t know that I would have done that in my first reply,” he tells me.

There are also times when we may be better off showing up as a muted, less authentic version of ourselves. That’s the version of herself Sarah, who asked me not to use her last name because she’s still in the process of finalizing a divorce, has been trying to be as she works through a custody agreement and the splitting up of assets with her soon-to-be ex-husband. She’s relying on ChatGPT to translate acrimonious texts to him. “Whenever you’re in anything that’s contentious, or you’re worried about anybody reading your texts, you just want to be really polite,” she tells me. “There are moments when I, as myself, cannot do that.”

Sarah will type up what she wants to say, with all the emotionally-charged snark and frustration over coordinating coparenting, and ask ChatGPT to rephrase the ask as “respectful, but with firm boundaries” to get across the logistical info she needs to say. Sarah gets the satisfaction of still putting her rage into words, but also the feeling of “moral superiority” as she sends texts to her ex that are calm, cool, and collected, and he responds with cheap jabs, sometimes made even angrier by her composure. “It does a good job of separating yourself from a hostile situation,” she says.

I’m in a very active group chat, where six of us jump between topics every day. If the chat really takes off and you’re the only one on do not disturb, it’s common to pick up your phone to hundreds of unread messages. I wondered if ChatGPT could step in and summarize when this happens, so I fed it a few days of group chat content. The chatbot was accurate, but missed the mark on the kind of humor only humans, and sometimes close friends, can get. One friend jokingly wished us all a happy “Hoagie Day,” which ChatGPT summarized as a “culturally significant” celebration.

But the bot did a decent job of summarizing a section where we all went back and forth trying to work out details for a camping trip. I asked then what I should say next, ChatGPT suggested something on topic and something “humorous”: “I’m only coming camping if someone promises to take a dramatic candid of me staring into the woods like I’m in a folk album cover” — which, if I ever sent this, I’m trusting my friends would call a wellness check on me to make sure I haven’t been kidnapped. For a TL;DR on what your friends are saying, chatbots might help in a pinch. But if you want your friends to maintain their respect for you, it’s still best to type with your own two thumbs.

Amanda Hoover is a senior correspondent at Business Insider covering the tech industry. She writes about the biggest tech companies and trends.