Before anyone built a companion that could talk, people spent four decades studying the companion that could not. The pet literature has already run the experiment the AI companion field is running now, with the same funding model, the same self-reports, and the same split result.
Hal Herzog, a psychologist at Western Carolina University, has spent much of his career on a question that sounds settled: are people with pets healthier and happier than people without them? In 2011 he published a review in Current Directions in Psychological Science under the title “The Impact of Pets on Human Health and Psychological Well-Being: Fact, Fiction, or Hypothesis?” His answer, after hundreds of studies, was the third option. The evidence was “at best, inconclusive.” Most studies, he said later, “do not show the pattern of results that the pet products industry claims.”
That sentence could be lifted into a review of AI companions without changing a word except the industry. The parallels are close enough to be useful, because the pet researchers have had thirty more years to find out what their early results were made of.
The pet effect, and who paid for it
The idea that animals are good for you entered the research literature in the 1980s with studies that found new pet owners reporting fewer minor health complaints and dog owners walking more. The earliest of that work was funded by Mars, the pet-food company, through its Waltham institute; the lead researcher later said there was “no other source of funding, really.” A 2008 partnership between Mars and the National Institutes of Health put industry money and federal money into the same program. A foundation created in 2010, backed by Petco, Purina, and Zoetis, has since spent more than three million dollars on research into the human-animal bond and, by its own executive’s admission, tends to emphasize the positive findings.
A 2021 survey from that foundation found that nearly nine in ten pet owners say their pet benefits their mental health. The number is real, and it measures what the owners believe. When researchers measured loneliness itself, with validated scales rather than a question about the pet, the picture changed.
What the loneliness studies actually found
In 2015, two researchers reviewed every quantitative study they could find comparing loneliness in pet owners and non-owners, thirteen studies spanning nearly three decades, and concluded that none of them provided convincing evidence that companion animals alleviate loneliness. Herzog later extended the tally to 21 studies through 2019. Six found pet owners less lonely as a group. Eleven found no difference. One large British study found pet ownership associated with more loneliness in older people, by about a quarter, and by half among women. Nine of the 21 compared cat owners with everyone else, and not one found cat owners less lonely.
The pattern inside the mixed results is the interesting part. In one study, 82 percent of owners said their pet had a strong impact on their lives, and their loneliness scores were indistinguishable from non-owners’. Pet ownership reduced loneliness in women living alone and did nothing for women living with others. Women living with a dog were less lonely in a German study; men were not. An Australian randomized trial that gave people dogs found loneliness dropped at three months and stayed down, until the researchers adjusted for age, education, and gender, after which the effect disappeared.
The effect, where it exists, is not a property of the animal. It appears when the animal changes what the person does: who they talk to on the walk, whether anyone else is in the house, whether the pet is a reason to leave it. A dog takes you places. A cat does not, and nine out of nine studies could not find a cat that made anyone less lonely.
The robot pets ran the experiment again
Before AI companions, there were robot ones, and they were studied with more rigor than most of what has come since. Paro, a robotic harp seal built for dementia care, was tested in a randomized trial in New Zealand rest homes: residents who spent twelve weeks with it, twice a week, reported less loneliness than those who did not. The detail that mattered came from watching the sessions. Residents talked to each other more around the seal than they did around the facility’s resident dog. A 2008 American study found a robotic dog reduced loneliness in nursing-home residents about as well as a living one.
Both trials were small, and both were run in institutions where the residents already shared a room and a schedule. But the mechanism they surfaced is the same one the pet literature keeps finding from the other side. The robot did not reduce loneliness by being company. It reduced loneliness by being something several people could be around at once, a reason to speak to the person in the next chair. The seal was a social object. The benefit was the other residents.
The prediction
Put the two literatures side by side and the pet research makes a specific forecast about AI companions. Self-reported benefit will be high, because people are reporting on something they chose and like. Industry-funded and vendor-collected studies will be more positive than independent ones. Validated outcome measures will show a smaller, patchier effect than the surveys. And where a real effect shows up, it will be in the people for whom the companion changed what they did with other humans, not in the people for whom it replaced them.
The early AI companion evidence already fits that forecast. The studies with the strongest effects are short, self-reported, and vendor-adjacent. The longer and more careful ones find a paradox in which comfort and withdrawal arrive together, and the one randomized comparison against a human stranger found the stranger worked and the chatbot did not. Where companions appear to help, it is when they make someone more social rather than less, which is the dog-walk mechanism with the dog removed.
There is one place the forecast breaks, and it breaks in the wrong direction. A pet cannot talk. Its company is real but bounded, and the boundary is what pushes an owner back toward people for everything a cat cannot give. An AI companion has no such boundary. It can hold the conversation, remember the day, ask the follow-up, and do all of it at two in the morning, which removes the one feature of pet ownership that reliably sent people back toward other people. The question the pet literature never had to ask, whether a companion can become hard to leave, is the first question for this one.
The pet researchers needed forty years to learn that the companion that helps is the one that changes what you do with people, and that the industry would rather fund the studies that do not check. If the next decade of companion research finds the same split, high satisfaction, patchy outcomes, and a benefit that lives in the humans around the device, the field will have rediscovered the pet literature forty years late, with the answer on the shelf the whole time.
Sources: Herzog, “The Impact of Pets on Human Health and Psychological Well-Being: Fact, Fiction, or Hypothesis?” Current Directions in Psychological Science (2011). Herzog, “Can Pets Relieve Loneliness in the Age of Coronavirus?” Psychology Today (April 2020; tally of 21 studies 1986 to 2019, Gilbey and Tani’s 2015 systematic review, Antonacopoulos and Pychyl 2010, Zasloff and Kidd, Antonacopoulos 2017, Powell and colleagues 2019). Gilbey and Tani, “Companion Animals and Loneliness: A Systematic Review of Quantitative Studies,” Anthrozoös (2015). Undark, “The Fuzzy Science on Whether Fido Is Actually Good for You” (2024; Serpell, Mars and Waltham, the NIH partnership, HABRI funding and the 2021 survey, Nestle, Feldman). Robinson, MacDonald, Kerse and Broadbent, “The Psychosocial Effects of a Companion Robot: A Randomized Controlled Trial,” Journal of the American Medical Directors Association (2013). Banks, Willoughby and Banks, “Animal-Assisted Therapy and Loneliness in Nursing Homes: Use of Robotic Versus Living Dogs,” JAMDA (2008).
