What the Dog Actually Does

The effect everyone promises is the one the evidence supports least. The effect nobody mentions is the one it supports best.

Does pet therapy reduce agitation?

Probably less than you have been told. This is the effect most often promised and the one the direct evidence supports worst.

The confusion is real, and it comes from two respectable bodies of research disagreeing.

On one side, a Bayesian network meta-analysis of 65 randomized trials comparing eleven non-drug approaches to agitation found animal-assisted intervention among the approaches associated with the largest reductions, alongside personally tailored activity and massage. That finding is genuine, and it is why animal therapy appears high in a lot of guidance.

On the other side, when researchers pool the trials of animal-assisted therapy on its own, the agitation effect thins out. A 2022 systematic review and meta-analysis of randomized trials found significant reductions in behavioral and psychological symptoms overall and in depression, but not sufficient evidence of benefit for agitation, cognition, activities of daily living, or quality of life. A separate review of dog-assisted therapy specifically reached the same place, with very low certainty across every outcome.

The practical version: if someone is telling you a therapy dog will settle your mother's late-afternoon agitation, that is a hope, not a finding.

What does the evidence actually support?

Mood. Specifically depression, and the broader cluster of behavioral and psychological symptoms.

This is the part that gets undersold, and it is the better reason to do it.

The 2022 pooled analysis found statistically significant improvement in depression in people with dementia receiving animal-assisted therapy. Depression in dementia is common, badly under-detected, and frequently mistaken for the disease progressing when it is a treatable condition sitting on top of the disease.

There is also a signal, from one small study, on apathy — the flatness that families read as not caring and that is in fact one of the most common symptoms of the illness. One small study is not a finding, but it is a reasonable thing to watch for.

• Mood is the target. Judge it on whether she is brighter on the days the dog comes, not on whether the shouting stopped.
• Depression is worth chasing separately. If the animal helps mood, that is a hint the mood was low to begin with, and that is worth raising with a doctor rather than filing under dementia.

Is a robotic pet as good as a real one?

On agitation, the robot may actually do better. This surprises everyone, including the people who ran the study.

A network meta-analysis comparing animal-assisted therapy with pet-robot therapy found that pet-robot therapy marginally benefited agitation compared with control, while animal-assisted therapy did not. Neither improved cognition, depression, or quality of life in that particular analysis.

Two honest caveats before anyone buys a robot seal. The comparison rests on a small evidence base, and "marginally" is the authors' word, not a softening of mine. And the outcome a robot wins on is exactly the outcome the whole field measures least reliably.

But it is enough to retire the assumption that a machine is obviously the lesser option. It never gets tired, never has an off day, and cannot be hurt by rough handling — three things that matter more in practice than they sound.

Does any of it help memory or slow the disease?

No. There is no evidence for that and no plausible reason to expect it.

Every review that has looked has found no effect on cognitive function. Animal contact does not slow the underlying disease, does not improve memory, and does not delay progression.

This matters because it is the claim most likely to appear in marketing material for a facility or a program. A well-run animal program is a quality-of-life offering. A facility presenting it as a treatment for the disease is telling you something about the facility.

How good is this research, honestly?

Weak, and everyone in the field says so. That does not make it worthless, but it does change how much weight the numbers can carry.

The certainty of evidence across these outcomes has been graded very low using GRADE, the standard system for rating confidence in findings. The reasons are structural rather than sloppy:

• You cannot blind anybody. Everyone knows whether a dog just walked in — participants, staff, and the person scoring the outcome.
• The trials are small and short. Most run for weeks, with modest numbers, often in a single facility.
• "Animal-assisted therapy" is not one thing. A trained therapy dog visiting weekly with a handler and a resident cat living on the unit are different interventions pooled under one heading.
• Publication favors the positive. Nobody is eager to publish the trial where the dog made no difference.

The most recent overview of systematic reviews on this question was published in 2026 and searched the literature through March of that year, which tells you the field is still actively trying to settle it.

So should we do it anyway?

Usually yes, for reasons that do not require a trial. Just be clear about which reasons those are.

If someone lights up when a dog comes in, that is worth doing on its own terms. An hour of pleasure is not a placebo and does not need a p-value. The trap is only in what you expect from it — and in what you are asked to pay for it.

• Do it because she enjoys it. That is sufficient.
• Expect mood, not behavior control. If the goal is a specific behavior, the evidence points to tailored activity, treating undiagnosed pain, and correcting hearing loss before it points here.
• Be skeptical of a fee attached to a promise. Enjoyment is a fair thing to buy. Symptom control is not, at this level of evidence.
• Watch the person, not the literature. Trials measure averages across strangers. You are looking at one person you know well, and you will see a real effect faster than any scale will.