Is AI companionship good for older people?

You've probably pictured it already: your mum chatting to a friendly voice on the tablet while you're at work, or your dad telling it the same story for the fifth time and it listening as if it were the first. It's a lovely picture, and it raises proper questions. Is it good for them? Is it honest? Could it make the loneliness worse?

Those are the right questions. Here's what we found when we went through the research, in plain language. If you're still working out whether loneliness is the issue at all, start with our guide to the signs of loneliness in elderly parents.

What the research shows so far

A short chat with an AI can ease loneliness in the moment

The best-known study is by Julian De Freitas of Harvard Business School and colleagues, first released as a working paper in 2024 and published in the Journal of Consumer Research in 2025 (De Freitas et al.). Across several experiments, people who chatted with an AI companion felt less lonely afterwards, about as much as people who chatted with another person, and more than people who watched YouTube videos. The biggest reason seemed to be "feeling heard". Participants also underestimated how much better they'd feel.

Two caveats matter for families. First, the participants were recruited online and were mostly around 40 years old, not older adults. Second, in a week-long study the relief happened each time but didn't build up over the week. In an interview with Harvard Business School, De Freitas said the effect was comparable for chatbots and humans "at least for a 15-minute interaction", and the researchers stressed it isn't a substitute for human company (HBS Working Knowledge, 2025).

Real-world programmes: encouraging, but not proof

New York State's Office for the Aging (NYSOFA) has given ElliQ, a tabletop companion robot, to older people living alone. In 2023, NYSOFA and the robot's maker announced a "95% reduction in loneliness" among more than 800 users (NYSOFA, 2023).

The underlying report is worth reading closely. It's based on engagement data from the device, and the wording is that 95% of clients report ElliQ is helpful in reducing loneliness and improving wellbeing (NYSOFA and ElliQ engagement report). That's a satisfaction rating, not a measured drop in loneliness, and there was no comparison group. It's still meaningful that so many people said it helped, and that the average active user chatted on six days a week. But it isn't the same as a trial.

Robotic pets and social robots

There's more research on robots than on voice apps, mostly in care homes. A 2019 review led by the University of Exeter looked at 19 studies of robotic pets in care homes. People described real comfort and companionship, and pooled trials showed less agitation, but the meta-analysis found no statistically significant effect on loneliness, depression or quality of life (Abbott et al., 2019). As one of the authors put it, robopets are no substitute for human interaction (University of Exeter, 2019).

A 2025 meta-analysis in The Gerontologist pooled 19 studies with 1,083 older people and found social robots were linked with lower loneliness, with bigger effects in care homes and other institutions (Mehrabi and Ghezelbash, 2025). The authors were careful: 11 of the 19 studies had no control group, there were signs of publication bias, and long-term effects are unknown.

For people living at home, a small 2020 study interviewed 20 lonely older adults who'd been given robotic cats and dogs. Those who lived alone or had fewer social ties seemed to benefit most; some of the younger participants found the pets toy-like (Hudson et al., 2020).

Technology in general: mixed results

It's easy to assume any screen that connects people will help. The evidence is more sobering. A 2021 systematic review of digital interventions for loneliness in older adults, such as video calls and internet training, found six studies with 646 people and concluded there was no evidence that they reduced loneliness, with evidence quality rated very low to moderate (Shah et al., 2021). That doesn't mean technology can't help. It means that handing someone a device is not, by itself, a fix.

More use isn't always better

In 2025, researchers at the MIT Media Lab and OpenAI ran a four-week trial with 981 adults using ChatGPT by text or voice (MIT Media Lab, 2025). The type of chat they were assigned made little difference. But people who chose to use it the most had worse outcomes: more loneliness, more emotional dependence, and less time socialising with people. Because people chose how much to use it, we can't tell which came first. Participants were around 40 on average. It's still a useful warning: a companion that's always available, with no limits, is not automatically a good thing.

People still matter most

For comparison, ordinary human phone calls have good evidence behind them. In a 2021 randomised trial, Meals on Wheels clients in Texas who got short, friendly calls from trained young volunteers for four weeks were less lonely, less depressed and less anxious than those who didn't (Kahlon et al., 2021). If your parent would welcome a real voice, free befriending calls are a very good place to start. We compare the options in daily check-in calls for seniors.

The concerns worth taking seriously

Honesty: knowing it's a machine

An AI that's warm and remembers your birthday can feel like a friend. That's partly the point, and partly the risk. In a 2023 survey of 825 adults aged 25 to 88, 69.3% said they'd be uncomfortable being allowed to believe, if they had dementia, that an artificial companion was human, and most didn't think a companion robot would make them less lonely (Berridge et al., 2023). Pennsylvania's guidance for people who work with older adults warns that companion bots can lead older adults to think they're speaking with a human, and suggests explaining that bots are not human and are designed to engage and validate users (Pennsylvania Department of Education).

Dependence, and less time with people

The worry isn't that your mum enjoys a chat. It's that the chat becomes the easy option and the lunch club slowly drops away, as the MIT and OpenAI findings hint. If she's going out less, not more, it's worth a gentle conversation.

Families visiting less

There's a quieter risk on our side too. Once a companion is in place, it's tempting to call a bit less. The World Health Organization has warned that AI in older people's lives may reduce contact between generations, and calls for AI to be designed with older people, respecting their right to consent and to contest (WHO, "Ageism in artificial intelligence for health", 2022). Any companion should be extra company, not a reason to visit less.

Privacy and data

Companions hear a lot: family worries, health, money, grief. Pennsylvania's guidance notes that companion bots can collect user preferences, location and communication history. In an October 2026 survey for AARP's Senior Planet, 70% of older adults said they lacked confidence about what's safe to share with AI tools (AARP, 2026). Regulators are paying attention: in September 2025 the US Federal Trade Commission opened an inquiry into AI chatbots acting as companions, asking companies how they test for harms, make money from engagement, and use what people tell them (FTC, 2025). That inquiry focuses on children, but the questions apply at any age.

There's a second privacy question that's about family, not companies. Reading every word your parent says to a companion would be surveillance, however kindly meant. Agree with them what you'll see.

Dementia, capacity and consent

Researchers looking at a tablet-based pet avatar for older people with cognitive impairment raised concerns about deception, monitoring and consent, noting there are no uniform guidelines and that someone's ability to consent may change as dementia progresses (Portacolone et al., 2020). If your parent has dementia, involve their doctor or care team, sit with them for the first chats, and keep checking that they still enjoy it and understand what it is.

Scams

Older people are heavily targeted by fraud. The FBI's Internet Crime Complaint Center received 201,266 complaints from people aged 60 and over in 2025, with reported losses of about $7.7 billion, up 59% on 2024 (FBI IC3, 2025 report). Any companion your parent trusts should never ask for money, bank details or passwords, and should be on their side when someone else does.

Questions to ask any AI companion

Whatever you're considering, these questions sort the thoughtful products from the rest. A good provider will answer them plainly.

• Does it say it's an AI? Always, including when asked directly?
• Who decides what family can see? Does my parent get a say, on their own device?
• What is kept, for how long, and who can read it? Are voice recordings stored? Is anything used to train AI models or shared with advertisers?
• Can we see and delete what it remembers?
• Are there limits, or is it designed to keep my parent talking as long as possible?
• Does it show ads or sell anything?
• What happens if my parent mentions a fall, chest pain or feeling hopeless? Does it point them to real help? Does it claim to be an emergency service?
• Will it ever ask for money, card details or passwords? Does it warn about scams?
• Does it encourage real-world contact, or try to be the only friend?
• Can we try it first, and leave easily?

How Dottie answers them

We built Dottie with these concerns in mind. Here are our answers, plainly. You can check each one on our safety and privacy page.

• Honest about being an AI. Dottie always says she's an AI, never pretends to be a person, and won't pretend to be a doctor.
• Consent on their device. By default, family see when your parent chatted, for how long, and a sentence or two about how it went. Full conversations are only shown if your parent agrees on their own device and you turn sharing on, and Dottie knows which setting is on so she can answer honestly. Summaries, not surveillance.
• Memory you can see. Dottie keeps short notes about names and the things that matter. Family can read and clear them at any time.
• Daily limits. Each plan has a daily chatting time, so Dottie stays part of the day, not all of it. The Emergency button keeps working when the time is used up.
• No ads, no selling. We don't keep voice recordings, and our AI provider, OpenAI, doesn't use the conversations to train its models.
• Real help, honestly framed. If your parent mentions a fall, chest pain, trouble breathing or feeling very unwell, Dottie tells them to call for help and shows a help card with buttons for the emergency number and the family contact. She isn't an emergency service, doesn't call anyone by herself and doesn't detect falls.
• Scam-aware. She never asks for money, bank details or passwords, and gently warns if someone else has.
• She doesn't ring them. Your parent taps to talk when they feel like it. Nothing nags.
• Their stories, their choice. Dottie asks before keeping a story for the family story book, and a no is always fine.

None of that makes Dottie a replacement for you. It's meant to be company between your calls and visits.

If you decide to try one

• Ask your parent first. Present it as something to try, not something decided for them. Our guide to when an elderly parent refuses help may be useful if they're wary.
• Tell them plainly what it is: a computer programme with a friendly voice, not a person.
• Agree what family will see, and stick to it.
• Keep your own calls and visits steady. Put them in the diary before the companion arrives.
• Check in after a few weeks. Are they enjoying it? Going out more or less? Would they rather stop?

For a broader plan, our guide on how to help a lonely elderly parent has twelve ideas, most of which involve people, not technology.

Questions families ask