What this is about:
Researchers at the University of Colorado Boulder have published what they describe as the first hands-on user study of generative ghosts: AI systems designed to simulate someone who has died. Sixteen bereaved adults held two short text conversations — one with an AI speaking about their loved one, and another speaking as them. Participants generally preferred the first-person “reincarnation”. More strikingly, they judged authenticity less by biographical accuracy than by tone, slang, terms of endearment and conversational rhythm. In grief, sounding right may carry more weight than being right.

Why it matters:
Bereavement research has long moved beyond the idea that healthy grief requires severing the relationship and “moving on”. Continuing bonds — talking to the deceased, maintaining rituals, revisiting memories or sensing their presence — can support meaning, identity and adaptation. But a systematic review of 79 studies found a complex picture: those bonds can bring comfort and distress, depending on their form, meaning and context. AI changes something fundamental. The bond now replies.

What’s happening:
Participants were sometimes willing to overlook invented facts, but a wrong phrase or unfamiliar pet name could puncture the encounter completely. Even when the AI was explicitly framed as a third-person representative, some users addressed it directly as “Grandma” or “Grandpa”. The formal boundary between memory of the person and the person apparently speaking proved remarkably easy to collapse. Participants found that closeness compelling, while also worrying that repeated use could become difficult to stop.

The catch:
This was a qualitative user-experience study, not evidence that AI improves grief outcomes. The sample was small, self-selecting and heavily weighted towards grandparent loss. Each interaction lasted around 20 minutes; there was no longitudinal follow-up, no comparison with bereavement therapy, and no measurement of prolonged grief, depression or functioning.

It was also unusually buffered. A facilitator remained present, mental-health resources were available, and a human operator reviewed and lightly edited GPT-4’s replies before participants saw them. The research tells us that a carefully staged, supervised encounter can feel powerful. It does not tell us what happens when a commercial griefbot is available indefinitely at 2am, with nobody checking what it says.

The bigger picture:
Earlier research with ten mourners already using chatbots found genuine perceived benefits: constant availability, a non-judgemental listener and, for some, a chance to say what had been left unsaid. But users generally positioned the technology as supplementary support, and the researchers also identified risks including inappropriate responses, isolation and over-reliance. The evidence remains almost entirely small-scale and qualitative. Feeling moved is not the same as demonstrating therapeutic benefit.

The ethics piece:
There are at least three people in every AI afterlife: the person whose data creates the ghost, the person who builds or controls it, and the person expected to interact with it. Their wishes will not always align. The deceased may never have consented; relatives may hold radically different memories; private messages may become training material; and a provider can alter the model, raise the price or close the service altogether.

Peer-reviewed ethics work recommends mutual consent, adult-only access, clear disclosure that the system is simulated, no unsolicited appearances or notifications, and meaningful procedures for ending or “retiring” a ghost. Otherwise, continuing bonds risk becoming commercial dependency — followed, potentially, by a second loss when the product disappears.

Reality check:
Using a griefbot is not inherently pathological, just as speaking to someone who has died is not inherently pathological. The clinically useful questions are about function and flexibility. Is the interaction helping someone remember, make meaning and remain connected to the living? Or is it becoming the only place they can tolerate the loss? Can they choose when to enter and leave? Do they understand that the answers are generated rather than recovered? And what happens when the bot places new words into the mouth of someone who can no longer correct it?

The point is not to police grief into one approved form. It is to avoid confusing a convincing simulation with an evidence-based intervention.

The interesting bit:
The study’s most important finding may be that affective fidelity outranked factual fidelity. A system may not need to know the deceased particularly well. It may only need to reproduce the rhythms that make a bereaved person feel recognised. That is the product opportunity and the safety problem in the same sentence.

The line:
A griefbot does not preserve what someone said. It generates what they might say next.

Sources: CU Boulder Today | Original DIS 2026 study | Death Studies: continuing bonds systematic review | CHI 2023: chatbots and grief support | Philosophy & Technology: responsible griefbot design | Generative Ghosts: benefits and risks of AI afterlives

Our subscriber drive wraps up this week, which means there are only a few days left to enter our book raffle.

Here is the deal: Share your unique referral link {{rp_refer_url}} with a therapist friend, colleague, supervisee, group chat member, or anyone in your life who enjoys emerging research and tech trends in mental health wrapped in a side of humor.

For every person who subscribes, you will receive one entry to win a $30 gift card to an independent bookstore of your choice.

You will also be helping more therapists find a community that takes the work seriously without pretending we have to be serious all the time.

One last housekeeping note: referrals only count after the new subscriber confirms their email address. Please remind your people to check their inbox and click the confirmation link.

The winner will be announced on July 22nd.

You can also copy & paste this link to others: {{rp_refer_url}}

Content Note: Misogyny, Racism, Public Sexual Harassment.

On Saturday we were invited to watch the World Cup in the pub. I am not an avid football fan. My Dad explained the offside rule to me and I both do and do not understand. I do, however, get the same empty-headed relief watching a game that I get with Wimbledon. I become so preoccupied with the ball that life itself ceases to exist for a bit. Maybe I am a dog.

I was debating what to wear, because-heat! And settled on a black fitted jersey maxi dress, with cooling tech. Granted, it was pretty fitted. Max has never said anything like this before, but in the same sentence that he said I looked very nice, he also commented that I might look too nice. He voiced concern about looking that nice in said pub context. To be very clear, Max does not control what I wear. He has never before made a comment like this. And to contextualise, British male football fans in a pub is a wholly different genre to a normal night out in Britain. Max knows this far better than me. Alas, I was indignant. Reflectively, this was very naïve.

During one of the “hydration breaks”, (I mean seriously, what even is that, FiFa) we rolled outside for some air. Cramped pub, heat, football fug. Outside we met two brothers, one of whom I know as he is well acquainted with some of our friends (he’s a scientist with a PhD). Their late father was an avid supporter of my political party. So my conversation started, quite normally, with how sorry I was for their loss. My naivety continued because there I was, deluded enough to assume that if someone’s father had certain values, said values must somehow pervade. That my friends would not be friends with people who did not hold anti-racist, anti-othering, anti-oppressive, anti-misogynistic views. Which is insane really because I know this is not true. One of my own brothers supports Reform.

We got into a conversation about life after death. I love a bit of banter like this and, frankly, it was a welcome relief from football brain numb. Everyone else went back into the pub but I stayed outside chatting. Max checked I was okay before he went back in and I brushed him off. Anyway, he could see me through the window. I don’t know exactly when I felt something change. It might have been around the derogatory comment about my friend/our local MP. And, when I pushed back on this, It might have been when they started repeating that their dad was “really a Reform supporter.” It might have been when PhD said I would “end up being raped in this town by a Muslim”. When I countered with evidence bases and data, alongside Full Fact, I was told by the other that I was a “fucking idiot woman” who had “been brainwashed”. When I steadily offered that this was rude, PhD brother agreed. Small mercies. Crumbs of civilisation.

When I got up to leave, PhD held onto my hand and said, “nah, stay here.” It wasn’t threatening. It wasn’t forceful. But it felt difficult to leave. It felt like I didn’t want to make a scene. There it is, isn’t it? That awful, familiar female calculation. Is this bad enough yet? Am I overreacting? Can I leave without escalating it? I sent a message to my friend saying “help”, because I was worried that if I sent it to Max he might come out and these two people, who I had radically underestimated, might ‘kick off’- because, football. And then the final straw: PhD said, “your tits look great in that dress.”

At this point my rage and biological brain kicked into gear. I told him that remarking in this way was out of order, misogynistic, rude and disgusting. He replied, laughing “it’s your fault for looking like that.” I had already got up by then. I wanted to say that their father would be ashamed, but I didn’t. Maybe because I didn’t want to weaponise grief. Maybe because I wanted to get back inside. Maybe because, despite everything, women are osmotically trained to manage the emotional temperature of a context that has just made us feel unsafe.

And this is the bit I keep coming back to. Max wasn’t wrong to clock the room. I wasn’t wrong to be furious that the room needed clocking. The dress was not the problem. My body was not the problem. Their racism was the problem. Their misogyny was the problem. The casual entitlement to my attention, my politics, women’s bodies, women’s compliance, was the problem. And yet the whole machinery of the moment still tried to drag me into self-audit: what I wore, why I stayed outside, whether I was too trusting, too polite, too naïve.

I sit with young women every week who are trying to build some sort of relationship with their bodies in a world that treats those bodies like public consultation documents. Commented on, legislated over, blamed, assessed, desired, judged, reduced.

And then we wonder why rage, shame and disappearance become so seductive.

Insight Timer continues to offer its series of free workshops for therapists, and this month’s lineup includes our friend and TB reader Daniel Fleshner, LPC, CST, who will be leading:

Sex with Presence: Mindfulness for Intimacy Anxiety
Today, Wednesday, July 15
12:00–12:30 PM PDT

Daniel’s workshop will focus on using mindfulness to support clients navigating intimacy anxiety, which is definitely one of those topics we probably all encounter but do not have enough language, tools, or training around.

Register here.

Last week, I hosted an event with Mental Health Match where we talked about therapist mistakes, repair, and reducing shame. I was nervous going in. The plan was for a few therapists, including me, to share vulnerable stories about moments when we got it wrong, and then open the conversation to the larger group.

While yes, it was nerve-wracking to share about microaggressions and other mistakes I’ve made, honestly, it was an incredibly uplifting experience - both to share and to listen and learn from.

Not uplifting because the stories were easy. They weren’t. The other speakers shared about burnout, addiction, perfectionism, and countertransference.

It was uplifting because everyone in the room seemed to resonate with what the speakers shared.

Most of us aren’t used to talking about our mistakes openly, which makes doing so all the more powerful.

The tool this week is simple, but not always easy:

When you make a mistake, seek out your community (Specifically, I’m recommending a supportive community, not the toxic Facebook group).

The right kind of community helps you stay honest without collapsing into shame. It helps you ask, What happened? What was mine to own? What repair is possible? What needs to change? What support do I need so this does not keep happening?

Mistakes can be isolating because shame tells us to hide. Community interrupts that.

So this week, consider:

The big picture: A new Nature Human Behaviour study found that, within a nationally representative cohort of Americans now in middle adulthood, conservatives’ measured health diverged from liberals’ during the 2010s. By the early 2020s, conservatives were also dying at higher rates—and COVID-19 did not fully explain the difference.

Why it matters: Previous research has often relied on self-reported health or compared mortality across red and blue counties. This study followed individuals over time using measured biomarkers and cause-of-death records, making it possible to examine how health changed within political groups rather than relying only on what people reported or where they lived.

By the numbers:

What may be contributing: In a separate 2024 survey, conservatives were just as likely as liberals to report seeing a primary care provider, but were less likely to trust that provider or follow their advice. Republicans and Trump voters were also less likely to report having visited a primary care provider.

Among people with chronic illnesses, conservatives, Republicans, and Trump voters expressed less confidence that their prescribed medications were useful, even for conditions such as hypertension that had nothing to do with COVID-19.

Political realignment only explains part of the story.

Bottom line: The authors argue that political beliefs may increasingly function as a social determinant of health by shaping how people seek care, trust clinicians and follow medical advice. But medical mistrust remains a proposed explanation, not a proven cause.

One caveat worth flagging: The findings are correlational, the health analysis focused on one cohort largely in its forties, and the medical-trust survey involved a different group of respondents.

The study shows that the gaps coexist, but it cannot yet establish that mistrust caused the differences in health or mortality.

Sources: Nature Human Behaviour, May 14, 2026

This may be in terrible taste after the Fresh Findings section above, but humor is how I’ve lasted in the field this long.

I present to you the Saturday Night Skit: MAHA-spital.

Yes, I did just do that. I bloody love writing this, it brings me an enormous amount of joy and allows my brain an epic break.

Please do share us, oh and share your pet photos too, so I can keep my ‘meme machine’ rolling!

Stories from the community

The results:

Therapist Brief viewership’s dominant position = less fairytale wedding and more have we considered funding several public services?