Robot pets for dementia: what the evidence actually supports

Agitation is the one dementia symptom that cleared significance in both 2026 research syntheses, and the effect is small. Cognition does not move. Nothing here touches the disease. And the trial closest to a family's real situation did not use the clinical robot seal most of this research is built on. It used a $159.99 battery-powered cat and a $179.99 dog, left in the room.

The literature also disagrees with itself, in a way nobody selling robot pets for dementia will mention.

What has been measured in people with dementia

Outcome	Best available evidence	Result
Agitation	PARO trials in Australian care settings, Int J Med Inform, July 2026: 6 RCTs, 1,444 participants	SMD -0.44, 95% CI -0.70 to -0.18, p = 0.0008
Agitation	AI-based socially assistive robots, The Gerontologist, April 2026: 14 studies, 10 pooled	SMD -0.26, 95% CI -0.49 to -0.03, p = .027
Anxiety	Same PARO review	SMD -0.59, 95% CI -0.91 to -0.27, p = 0.0003
Depression	Same Gerontologist review	SMD -0.30, 95% CI -0.56 to -0.03, p = .027
Cognitive function	Same Gerontologist review	SMD -0.07, 95% CI -0.30 to 0.16, p = .555
Neuropsychiatric symptoms	Cluster RCT of Joy for All pets, JMIR Aging, 2022: 8 care homes, 83 residents	Fell in the robot homes, rose in the control homes, P < .001
Caregiver burden	Cluster RCT, Alzheimer's & Dementia, February 2026: 85 residents, 6 group homes	LS mean -3.29, 95% CI -6.26 to -0.32, p = 0.030
All of it, pooled to 2022	Ageing Research Reviews, June 2022: 66 studies	No clear evidence of benefit

An SMD of -0.26 is small by the usual convention and -0.44 sits at the lower edge of moderate. Against a symptom families call the hardest part of their week, a small measured shift is worth having. It is also smaller than the marketing photographs imply.

One row needs a warning label. The depression line in the July 2026 PARO review prints a confidence interval that excludes zero next to a p value of 0.09, and repeats the agitation point estimate exactly. Those authors call the depression evidence unclear regardless.

The same seal, two teams, opposite conclusions

In 2022 a team led by Gill Livingston, whose group also writes the Lancet Commission reports on dementia, reviewed 66 studies of socially assistive robots in this population. Pooling PARO against usual care and against an active control over five to twelve weeks, they found no significant improvement on agitation, cognition, neuropsychiatric symptoms, apathy, depression, anxiety or quality of life. Their conclusion: robots are feasible and acceptable, and there is no clear evidence that people with dementia derive benefit from them.

Four years later, two teams pooling a heavily overlapping set of trials reported significant reductions in agitation. What changed was mostly the arithmetic and the comparator: measured against another activity rather than against nothing, the robot's advantage shrinks. The largest trial in the field, covered in the wider therapy-evidence piece, ran PARO against an identical plush toy with the robotics switched off, and the plush toy came out slightly better value per point of agitation averted.

Then there is what the clinical literature leaves out. A 2024 network meta-analysis in International Journal of Mental Health Nursing ranked non-drug interventions for dementia symptoms across 43 randomized trials and 4,978 participants: exercise plus usual care best overall, aromatherapy with massage best for agitation. Robot pets are not among the ranked interventions at all.

The trial closest to a family's situation

Almost everything above ran on PARO, which sells by quote through clinical distributors. The exception is the study families should read. Hannah Bradwell and colleagues at the University of Plymouth ran a stratified cluster randomized trial in eight Cornwall care homes starting January 2020. Each intervention home got one Joy for All cat and one dog; the four control homes carried on as usual. Of 253 residents with the chance to interact, 83 enrolled for data collection and 63 were analysed at four months, 20 having died during the study.

Neuropsychiatric symptoms decreased in the robot homes and increased in the control homes (P < .001), with a significant difference in occupational disruptiveness too (P = .03), the subscale for how much those symptoms disrupt staff. Of the residents who interacted, 46 of 54 were reported to have positive experiences, and the team recorded no novelty effect across eight months, the most reassuring finding in the category. Collaborators could not be blinded, so the authors flag possible positive reporting bias.

Then the result that should change how families think. Residents who went on to interact with the robots had significantly higher dementia severity scores than those who did not (P = .001). The pets held the attention of people further along, not people in the early stage. Buying one as a gentle introduction at diagnosis is not what the evidence describes.

How often, not how long

PARO trials typically ran 15-minute sessions three times a week for six to twelve weeks. A Japanese cluster-randomized trial in Alzheimer's & Dementia, February 2026 tested frequency directly across six group homes and 85 residents, with no health professional running the sessions, and only the thrice-weekly arm cut caregiver burden significantly. Bradwell's homes scheduled nothing at all: the pets simply lived there, which is the easier protocol for a household and the one with a four-month result attached.

Whether you tell them it is a robot

Every family that buys one of these arrives at the same question, and a Canadian expert panel worked through it. Published in BMC Geriatrics on 30 January 2026, a modified Delphi with researchers, clinicians and family caregivers put 143 statements to 12 panellists, then 52 revised ones to eight, and settled on 135. Handed the statement that caregivers should tell residents the pet is not real, the panel rejected it and replaced it with this: "Care providers should enter the resident's reality when they use the robotic pets, as residents may believe it's a living pet or a robot."

The rest of that guide is practical in a way research rarely is. Introduce the pet with neutral language and wait: the paper's example is a staff member saying "Look at this" while gesturing to it. Let the resident lead. Sanitize hands before shared use, train staff on a written cleaning protocol, and get consent from family when the person cannot give it themselves. Shared pets are fine, so a home does not need one per resident.

An ethnography in Swedish nursing homes (Journal of Aging Studies, December 2024) lands nearby: a robot cannot lie by itself, and its deceptive potential comes from the people around it, which blurs the line between deceiving someone and following them empathically. A review published in March 2026 catalogued at least 60 distinct ethical concerns in this field, so nothing is settled.

What to buy, and what it costs

The pets in the Cornwall trial cost $159.99 for a cat and $179.99 for the Golden Pup on joyforall.com. No app, no WiFi, no subscription, four C batteries as the only recurring cost, and a product page that markets them for Alzheimer's and other dementias.

PARO publishes no price, so for most families it is something to ask a care home about rather than to buy, and Tombot's Jennie, the robotic Labrador built for this audience, remains waitlist-only with no published price. The expensive consumer tier is untested here: PubMed holds no dementia trial of Moflin at $429 or aibo above $3,000. Spending more buys more machine, not more evidence. The rest of the category sits in the robot pets guide for older adults, and the conversational tier in the companion robot buyer's guide.

What none of this shows

Cognition did not move, and no study claims it did. The longest follow-up anywhere above is eight months. A pilot trial in 38 hospitalized older adults with dementia reported fewer psychotropic medications and fewer instances of delirium among those visiting with PARO than with a human visitor, which is interesting and far too small to act on.

If your person is in residential care, the useful call this week goes to the home rather than to a store. Ask whether they already have robotic pets, whether those are shared or assigned, and who cleans them. If the answer is none, two pets covered a whole home in Cornwall, a small enough number for a family council to raise at the next meeting.