You don’t need AI to write a love letter

For the past year or so, I’ve been checking in semi-regularly with the corner of Reddit in which people crowdsource variations of the same question: How to feel when someone in your life uses generative AI for human-to-human communication — the suspiciously well-argued apology letter from a mother-in-law, the stilted wedding vows, the Sunday sermons that suddenly seem both more earnest and more generic. Sometimes the communication in question features what are now recognizable as LLM tells, like an overuse of triadic phrasing (“My feelings for you are not X, but Y”) and em dashes. But more often, people on the receiving end just feel like something is . . . off.

AI companies have been working tirelessly to stuff generative and agentic AI into every nook and cranny of our online lives. These might be helpful, anodyne retail chatbots; they also might be utter menaces like Google Docs’ Gemini pop-up. (It’s hard not to feel undermined when the AI is asking what changes it should make to a sentence I haven’t finished typing out.) The question of whether these interventions are good — and yes, I am aware that they can be turned off — is almost beside the point: The companies forcing generative AI into everyone’s life are shifting the Overton window to normalize using AI in the one place where, at least for those not using it for accessibility, it has no reason to be: personal interactions.

Want more from culture than just the latest trend? The Swell highlights art made to last.Sign up here

Last week on “The View,” co-host Sunny Hostin mentioned using ChatGPT to make notes to her husband “nicer.” Veteran tech journalist and podcaster Kara Swisher, sitting in for Joy Behar, had some choice words for her, namely, “Do you not have friends [you] can do that with?” When Hostin enthused that she has “a PhD in ChatGPT” and regularly uses the LLM for personal and marital counseling, Swisher held up a warning finger: “Please don’t do that.”

The broader discussion was about the influx of high-profile influencers and Hollywood celebrities — Demi Moore, Reese Witherspoon, Mel Robbins, — who appear to be spokespeople for GenAI and have been using their sizable platforms to deliver a message that’s half doomsaying, half cheerleading. Women need to get on board with GenAI tools or risk being left behind. Also, it’s empowering! Just ask Robbins, who peppered a recent Instagram post sponsored by Microsoft Copilot with feminist-lite language while urging women to share their financial records with the AI. Or Sheryl Sandberg, whose Lean In Foundation now hawks AI to women; in March, Sandberg joined the board of directors of Nscale, a British data-center startup.

The value proposition of generative AI is that it streamlines the process of communicating by skipping connective thought processes. Without what’s called cognitive friction, using tools like ChatGPT yields fluency without a grounding in knowledge; writing in The Walrus, Sheldon Fernandez likens it to building a house on sand — it looks great in the moment but won’t hold up for long.

Earlier this year, a mother I know read one of her son’s school assignments and felt a surge of disappointment: “It didn’t sound like him at all, and when I asked him about it, he explained that it was an overdue assignment and he needed to get it done fast,” she said. Recalling that she grew up being warned away from copying off someone else’s test because “cheating is only cheating yourself,” she sounds glum. “They’re learning that people who don’t cheat are suckers.” This is borne out by reports from within higher education, where the use of AI is often not considered cheating at all.

For someone already using AI in school or the workplace, using it everywhere, for everything, is also frictionless. A June 2025 MIT paper mentioned by Fernandez studied a group of students writing the same essay, with one using an LLM, one using a search engine, and one using just their brain. It found that “83% of participants who used generative AI could not recall a single quote from their own essay, whereas 89% of participants who did not use AI assistance were able to do so.” Other studies of students have concluded that reliance on LLMs is associated with procrastination, memory loss and diminished academic performance. Human brains function on a pretty strict use-it-or-lose-it policy: When they’re not engaged, they start to atrophy.

“Each time I rely on AI to think through a problem, my own critical thinking muscles are getting a bit weaker,” wrote Ines Lee in a 2024 Substack dispatch titled “Oops, I’m becoming codependent on AI.” “It’s sort of like outsourcing your workout to someone else and expecting to get stronger. Muscles — physical and cognitive — just don’t work like that.”

But AI use can also affect processes we don’t necessarily think of as load-bearing, like the trust in our own abilities that naturally develops over time. Cary, 45, started to use LLMs at her workplace because doing so was heavily encouraged; a year or two later, she found herself struggling to write a remembrance of her father for his memorial service. She’d never worried about it before, but suddenly, the task of pulling something both heartfelt and coherent from her seemed overwhelming: “It was like, Do I even remember how to do this?”

The awareness that reliance on GenAI has cognitive impacts (and can be addictive) means that we’re now being told how to stop relying so much on it. Create dedicated “thinking spaces” in your home or workplace. Give yourself 30 minutes of thinking on your own before hitting up Claude. Track how you feel when you’re using an LLM for something just as easily done without one. It’s not that these aren’t helpful reminders. The issue is that they accept the premise that AI is so necessary, so all-seeing and so inevitable that we’re using it even in situations where our knowledge and experience have always been sufficient.

And this is a problem, given that large-language models might, at this point, be best known for being sycophantic flattery machines. The stories of chatbots being so affirming of human feelings that they end up as accessories to suicide are outliers, but studies of chatbot interactions with humans have all pointed out that they prioritize agreeability, if not actual sycophancy.

Whenever I type a question — say, “How common is chatbot sycophancy?” — into Google, it responds with what seems like an excess of enthusiasm: “Great question!” Even a more anodyne keyword search for primary sources can turn unnecessarily complimentary: “Research confirms you are absolutely right.” (Again, I am aware that these features can be turned off and, again, that is beside the point.) Like most women and a whole lot of men, I’ve lived most of my life being told how not-enough I am, continually reminded to notice how much I lack and also notice the things I can buy that will make me lack slightly less. So when my Google searches treat my request for info about sitcom intros of years past like splitting the atom, I get suspicious.

Lindsey Hall’s recent viral essay “I stumbled across my boyfriend’s ChatGPT and it ended our relationship” is an example of how many people are willing to farm out their thinking for no reason other than, well, why not? Hall describes stumbling on the ChatGPT logs of her boyfriend of 5 months and finding that he had asked the LLM to decide, based on the concerns he input, whether he should continue their relationship.

“[H]e laid out his doubts in clipped, almost clinical fragments: my lifestyle, my sensitivity, my past, my van, my online writing, my eating disorder history, my cats . . . I was, in short, being methodically assessed.” This activity itself isn’t new: People frequently work out the big picture of romantic compatibility by listing pros and cons, and the discovery of such a list is a recurring trope in sitcoms and rom-coms. But seeing herself reduced to raw data by ChatGPT hit very different for Hall: “It was like accidentally reading someone’s diary, except the diary was a f*cking robot — predisposed to agree with him, ready to take his cruel thoughts and shape them into something that sounded reasonable.”

There’s no evidence that well-known women using social media and public appearances to sell their audiences on GenAI have any ill intent, or that they’re shills for AI companies. But given what we do know makes it worth asking, as Swisher did, why their support is so unquestioning — and why they frame it as empowering without mentioning the many ways in which it’s the opposite. (Let’s remember that the last tech Witherspoon hyped to women was NFTs.)

The more that the likes of Robbins and Sandberg and Witherspoon warn that women miss the AI train at their peril, the more it seems like the new Leaning In. Women recruit women into a molding themselves to a corporate structure that wasn’t built for them, tell them that structural workplace inequality is actually just insufficient personal investment, and turn negativity away at the door. Meanwhile, the men founding and running AI companies continue to prove how little they value women as fellow humans: They’ve knowingly perpetuated body anxiety in girls in order to capitalize on it; they’ve let their AI chatbots sexualize and exploit minors; they’ve said, out loud and approvingly, that their technology will erode the cultural and political power of women. If there’s a reason these women are intent on convincing women to adopt (but not question) this technology, I’m all ears. But so far, they’re promising things that they not only can’t deliver, but also can’t articulate.