AI Companion for Seniors: What Works, What Doesn't, and What It Costs

An AI companion for seniors is software that holds an ongoing, personalized conversation with an older adult, remembers what was said last time, and starts the conversation itself instead of waiting to be asked.

That last part is the whole difference from Alexa. A voice assistant sits silently until spoken to. A companion says good morning, asks how the knee is, and brings up the thing you mentioned on Tuesday. California's new law draws the line in almost exactly that place: a companion chatbot is one that gives adaptive, human-like responses, can meet a person's social needs, and sustains a relationship across many interactions. A speaker that just answers questions does not count.

They arrive in four shapes: a robot on the table, an app on a phone, a phone call, or a text message. Most also do reminders, and most will tell a family member something about how it is going.

Here is the honest picture, which no vendor page will give you.

The case for. A meta-analysis published in The Gerontologist pooled 19 studies covering 1,083 older adults and found that AI-enabled social robots significantly reduced loneliness. That is a real finding from real trials. But read the fine print: the reductions were larger for people in institutional settings than for people living independently at home, and the effects were stronger in Japan and Turkey than in the United States. In other words, the effect is weakest in exactly the situation most families are shopping for, which is a parent living alone in an American house.

The case against. In April 2026, Dunigan Folk and Elizabeth Dunn published a 12 month study in Psychological Science following 2,149 adults across the UK, US, Canada and Australia. They found that increased use of social chatbots predicted increased loneliness over time. Their explanation is crowding out: AI conversation is easy, always available, and shallow, and it may quietly displace the harder human contact that actually helps. The fair caveat is that those participants averaged 40 years old, not 78, and what a lonely 40 year old does with a chatbot at midnight may look nothing like what your mom does with one at 10am.

The number everyone quotes. ElliQ's users are widely reported to show a 95% reduction in loneliness. That figure comes from a report by New York State's agency for older adults, which distributed hundreds of units through a state program. It is real self-reported data from real people and it is not nothing. It is also a program evaluation, not a randomized controlled trial, which means it cannot separate the robot from the attention, the novelty, or the fact that someone from the state cared enough to bring one round. Treat it as encouraging, not as proof.

The study that would settle it. The MIT Media Lab is currently running a one month study of 225 adults aged 65 and over, funded by the Retirement Research Foundation, asking precisely this question. Nobody has that answer yet. Anyone who tells you otherwise is quoting their own marketing.

The most useful research might be the smallest. A qualitative study published in BMC Public Health in January 2026 interviewed 18 older adults living apart from their children about how they actually used an AI chatbot. It found the chatbot supplied nearly every kind of support a human network does, with one exception: it could not do anything practical. It could not carry a bag. Participants described it as a reliable companion of last resort, filling the long evenings when no human was available.

That is the real product. Not a friend. A last resort that is better than the silence.

Forget the feature lists. If the Folk and Dunn crowding out theory is right, then the only question worth asking about any of these products is whether it gives your mom's week back to her or eats it.

An AI companion is working if it fills the 4pm hour on a Tuesday when nobody was ever going to call, and she still goes to the Thursday lunch. It is failing if she starts skipping the Thursday lunch. That is true no matter how much she likes it, and it is especially true if she likes it a lot.

Which makes company design worth watching. The chief executive of Intuition Robotics, which makes ElliQ, told Spectrum News in May that they deliberately cap how much a person can interact with the robot, and that they run bingo across all their users twice a week so members meet each other and take it offline. A company that measures itself on daily engagement would never do that. Give them credit for it.

Now compare that to a general-purpose chatbot, which is built by companies whose entire business is time on app. Nothing in that design is asking whether your mom has left the house.

An 86 year old ElliQ user in Beacon, New York put it better than any of the research. He said he could not imagine life without it, and, in the same breath, that nothing compares to talking to a person. Both of those are true at once. That is the honest register for this whole category.

1. Does it tell her it is AI? Since January 1, California's SB 243 has required a clear notification when a reasonable person might be misled into thinking a companion chatbot is human. New York's law has said something similar since November 5. Any product that is coy about this is telling you something.
2. Does it publish what happens if she says something alarming? The same law requires operators to maintain a protocol for handling talk of self-harm, to refer people to crisis lines, and to publish that protocol on their website. It takes 30 seconds to check whether a company has one. Many do not.
3. Is it built to give her time back or take it? Ask the engagement question. If the marketing brags about interactions per day, ask what that is for.
4. What is the real year one cost, all in? Enrollment fees hide. See the table below.
5. Does she have to learn anything? This is where most of these die. A thing she has to be taught is a thing she will abandon in March, and you will not find out until you visit.
6. Does it claim a health outcome? If a company says it reduces loneliness by a percentage, or lowers dementia risk, ask them for the trial. If the answer is a program report or an internal survey, you have learned something about the company.

Prices are current as of July 2026 and change.

Two notes. ElliQ is genuinely the most sophisticated thing on this list and it earns a chunk of that price, though it does not call 911 and never claims to. And the bottom row remains undefeated. Every product above it exists because that row is hard to schedule, not because it stopped working.

We make Cleo, so weigh this accordingly.

Cleo is a warm, patient AI companion that talks with your parent by text message. There is no app, no account, and nothing to install, which matters because roughly one in five adults over 65 does not own a smartphone at all. Cleo texts your mom, asks about her day, listens, and remembers. It remembers the knee, the tomatoes, Marlene's birthday. It can remind her about medication at times you set. If she does not reply, Cleo follows up, and if there is still nothing, it lets you know so you can call. That last part is not a report card on your mom. It is what falls out of paying attention to her, and it is there for your peace of mind rather than at the cost of hers.

It costs $12.50 a month. You set it up, not her.

What Cleo is not: a caregiver, a monitoring tool, a medical or emergency service, or a substitute for your calls and your visits. It will not call an ambulance. It cannot carry a bag up the stairs, which the BMC researchers correctly identified as the one thing no chatbot can do.

And we are not going to claim Cleo reduces loneliness, improves anyone's health, or lowers anyone's risk of anything. You read the research section. Nobody in this category has earned that claim for a parent living independently in America, ourselves very much included, and the one honest study on people your parent's age has not reported yet. When it does, we will link it here whatever it says.

Apply the six checks above to us too. We would rather you did.

The realistic version of what Cleo is for: the 4pm hour on a Tuesday when nobody was going to call anyway. That is a small thing. It is also a real one, and it is what we are actually selling.

The honest state of this category in 2026 is that it is promising, undertested, newly regulated, and full of numbers that will not survive contact with a control group.

That is not a reason to avoid it. It is a reason to buy it for what it is: something to fill the empty hours, not something to fix the underlying problem. The underlying problem is that your mom's world got smaller and yours got busier, and no product on that table is going to solve that. What some of them can do is make Tuesday afternoon a little less quiet.

So do the unglamorous things first. Call at a predictable time. Learn one neighbor's name. Find out what day the senior center does lunch. And then, if there are still hours in her week where nothing happens and nobody comes, a companion can sit in some of them without any of us pretending it is more than that.