Home » Articles » Getting a Service Dog » Diabetic and Seizure Alert Dogs: What the Studies Actually Measured

Diabetic and Seizure Alert Dogs: What the Studies Actually Measured

By /
Featured card reading Diabetic and Seizure Alert Dogs. Three studies found 83, 56 and 36 percent on lows. None says a dog replaces a meter. Photo of a close-up of a dog's nose.
Photo: Anastassia Anufrieva / Unsplash

Published October 21, 2026Checked against the current regulations

Do diabetic alert dogs work? Some do, some of the time, and the number you get depends on who measured it. Three studies of trained diabetic alert dogs put their hit rate on low blood sugar at 83 percent, 56 percent and 36 percent. Seizure alert dogs have less behind them: a real finding that seizures have a smell, and no published proof I could find that a dog can reliably warn you before one starts. Under the ADA, none of that decides whether your dog is a service dog. Training for a task does.

This page puts all three numbers side by side, says who produced each one, and shows what the dogs have sold for. For the legal ground under every service dog, start with do I qualify for a service dog, and for what each state adds, service dog laws by state. Everything here is built from the studies and the court record, not from my own dog.

What the law counts as a task

The ADA definition of a service animal, at 28 CFR 35.104, is “any dog that is individually trained to do work or perform tasks for the benefit of an individual with a disability.” It gives examples, and the list is “not limited to” them. Two are on point here:

  • “assisting an individual during a seizure”
  • “alerting individuals to the presence of allergens”

Read the first one closely. The regulation’s own example is helping during a seizure. It does not say the dog has to see one coming. A dog trained to go get help, to fetch a phone or medication, or to stay with a handler on the floor is doing a task the regulation names. Whether a dog that alerts before a seizure is also doing one depends on whether that behavior is trained, not on whether it is guaranteed.

The same definition draws the other line: “the provision of emotional support, well-being, comfort, or companionship do not constitute work or tasks.” A dog that seems to sense when something is wrong but was never trained to do anything about it is not a service dog under this rule, however much it matters to you. The full test is in how do I qualify my dog as a service dog.

And one thing the law does not require at all: proof. A business may ask only whether the dog is required because of a disability and what task it is trained to perform. It may not ask how accurate the dog is.

Diabetic alert dogs: three studies, three numbers

Study How alerts were checked Hit rate on low blood sugar
Rooney and colleagues, PLOS ONE, 2019. 27 dogs Records kept by the dogs’ owners Median 83 percent
Wilson and colleagues, Frontiers in Veterinary Science, 2019. 8 dogs Flash glucose monitor plus video, scored blind Mean 55.9 percent, ranging 33.3 to 91.7 by dog
Gonder-Frederick and colleagues, Diabetes Research and Clinical Practice, 2017. 14 participants Masked continuous glucose monitors 35.9 percent

“Hit rate” here is sensitivity: of the low blood sugar episodes that happened, how many the dog alerted to.

I am not going to tell you which number is right. They measured different things in different ways. But you should know who produced each one, because that is the part most pages leave out.

  • The 83 percent study lists two authors as full-time employees of Medical Detection Dogs, a UK charity, and a third as a part-time paid employee of it. One author has no financial interest. It is also the top Google result for “do diabetic alert dogs really work.”
  • The 55.9 percent study came from largely the same group, using a method that did not depend on owners’ records. Its conflicts section says one author “was given a salary by Medical Detection Dogs (MDD) to develop this paper,” and that another “is a co-founder of MDD.” Its own finding: “substantial variation was seen despite all dogs having been trained by the same institution.”
  • The 35.9 percent study used masked glucose monitors and lists no tie to a training organization in what I read. Its conclusion: “the majority of DADs did not demonstrate accurate detection of low and high BG events.” It found 26.2 percent for high blood sugar.

Who benefits from you believing the highest number? Anyone selling a dog. That does not make the 83 percent wrong. It means the people closest to the dogs got the best result, and the people furthest from them got the worst, and both are worth knowing before you spend money.

A 2026 scoping review in Medical Sciences pulled ten studies together and landed where the evidence does: diabetic alert dogs are “complementary, context-dependent support strategies rather than as substitutes for conventional glycemic monitoring technologies.” The 2019 owner-independent paper argues the same. No study I opened says a dog should replace a glucose monitor.

Seizure alert dogs: a real smell, an open question

The strongest finding is a 2019 study in Scientific Reports by Catala and colleagues. Five trained dogs were given odor samples, cotton pads wiped on patients’ skin and breath collected in a bag, from five people with epilepsy, taken during seizures, after exercise and during calm activity. The dogs picked out the seizure samples with 67 to 100 percent sensitivity and 95 to 100 percent specificity.

The authors’ conclusion is careful: “This constitutes a first proof that, despite the variety of seizures and individual odours, seizures are associated with olfactory characteristics.”

Now the part that matters for anyone shopping. Those dogs sniffed cans in a lab, not people at home, and the samples were taken during the seizure, not before it. The paper’s own next step: “Further studies will aim to look at potential applications in terms of anticipation of seizures.” In other words, warning ahead of time was the question the study left for later.

I went looking for a published study that settles it. I could not open two review papers on this subject from here, so I am not going to summarize what they say. What I can tell you is what I read: a seizure has a smell, and the best study of it does not claim a dog can predict one.

That is exactly why the regulation’s example matters. A dog trained to respond during and after a seizure is doing a named task, whether or not it ever alerts in advance.

What the dogs have sold for

In May 2018 the Virginia Attorney General sued a company that sold “diabetic alert dogs” that could “purportedly identify and alert individuals to life-threatening low and high blood sugar.” Families paid $18,000 to $27,000. The complaint said many received poorly trained puppies.

The consent judgment in August 2021 ordered:

  • $514,406.39 in restitution to consumers
  • $1,447,919.18 to the Commonwealth for charitable organizations
  • $822,500 in civil penalties, suspended if the founder complies
  • a lifetime bar on the founder being paid by, or holding a fiduciary position in, organizations that breed, train, sell or place companion animals

The 2021 judgment describes dogs sold to help with diabetes, autism, seizure disorders or post-traumatic stress. The full story of that case is on what service dog training costs, and what a service dog costs from every route is in how much does a service dog cost.

Put the prices next to the table above, and ask which of the three numbers a seller is quoting you.

What no one can promise you

  • That a dog will catch every low. No study I opened found that, including the one most favorable to the dogs.
  • That a dog will warn you before a seizure. The best study of seizure odor says that question is still open.
  • That one dog will perform like the average. In the owner-independent study, dogs from the same trainer ranged from 33.3 to 91.7 percent.
  • That a dog replaces your meter. The 2026 review and the 2019 owner-independent study both say the opposite.

None of that means an alert dog is worthless. It means the dog is one layer, not the only one, and anyone who sells it as the only one is selling something the studies do not support. If insurance is part of how you are planning to pay, does insurance pay for a service dog covers that honestly.

Size, breed and the nose

Scent work has no weight requirement. A small dog’s nose is not meaningfully worse at this than a large one’s, which is laid out in can small dogs be service dogs. In the studies above, the variation was between individual dogs, not between breeds.

The test to bring to any seller

Ask one question: how was the dog’s accuracy measured, and by whom? If the answer is owner reports or the seller’s own records, ask for anything checked against a glucose monitor. If the seller cannot answer, or answers with a percentage and no study, you have learned what you needed to. And for a seizure dog, ask what the dog is trained to do during and after a seizure, because that is the part the law names and the part you can test.

Frequently asked questions

Do diabetic alert dogs really work?

Some do, some of the time. A 2019 study by researchers tied to the training charity found a median 83 percent sensitivity to low blood sugar from owner records. A 2019 owner-independent study found 55.9 percent, and a 2017 study with masked glucose monitors found 35.9 percent. Individual dogs vary widely.

Can a dog predict a seizure?

Not reliably, as far as the published evidence I could read shows. A 2019 study found five trained dogs could identify seizure odor samples, which shows seizures have a smell. The authors said using that to anticipate seizures is a question for future studies. Responding during a seizure is a task the ADA regulation names.

Is a seizure response dog a service dog under the ADA?

Yes, if it is individually trained to perform a task related to the handler’s disability. The ADA definition at 28 CFR 35.104 lists assisting an individual during a seizure as an example of work or tasks. A dog that only provides comfort is not a service dog, because the definition excludes emotional support.

Are diabetic alert dogs worth it?

That depends on what you expect. Every study and review opened for this page describes alert dogs as a support alongside glucose monitoring, not a replacement for it. In the studies, dogs from the same trainer varied from about a third of lows to more than nine in ten. Ask how any dog’s accuracy was measured before you pay.

What is the average cost of a diabetic alert dog?

There is no independent published average. The one documented range on this page comes from a Virginia lawsuit, where families paid $18,000 to $27,000 for dogs sold as diabetic alert dogs. The company’s founder was later barred for life from being paid by organizations that breed, train, sell or place companion animals.

Can a business ask how accurate my alert dog is?

No. Staff may ask only whether the dog is required because of a disability and what work or task it has been trained to perform. They may not ask for proof of accuracy, certification or training, and they may not ask about the nature of your disability. Accuracy is a question for you and the person selling the dog.

I am a veteran and a service dog handler, not a lawyer and not a clinician, and nothing here is legal or medical advice. Every link above goes to the regulation, the agency or the study itself, so you never have to take my word for it.

Sources

Every source below was opened and re-read on 20 October 2026.

Scroll to Top